Fears of an AI-driven permanent underclass misunderstand historical economic diffusion; physical constraints, human coordination, and organizational redesign mean technological gains take decades to fully permeate rather than abruptly wiping out employment. (3:06)
Companies are restructuring into automated operational loops across engineering, sales, and support, which optimize to local maxima but still rely on human judgment, taste, and out-of-distribution thinking to leap to new opportunities. (12:35)
Venture capital funding logic has inverted: conservative startup pitches with modest wedges face skepticism, while unconstrained, hyper-ambitious seed rounds reaching up to $100 million are actively backed. (62:40)
Startup moats in the AI era are discovered through continuous product momentum and usage telemetry rather than designed upfront in business plans, making interface craft and invisible UX details the ultimate differentiators. (52:43)
Detailed notes
## Background
Published on Lenny Rachitsky's podcast feed on September 12, 2026, this episode features Andreessen Horowitz (a16z) General Partner Anish Acharya. According to the show description, Acharya joins Lenny to discuss how AI is transforming companies into automated loops, why dystopian fears of an economic underclass are misplaced, and why the next frontier of consumer AI should prioritize human connection and ambitious creation over simple workplace productivity.
## TL;DR
Anish Acharya argues that modern organizations are evolving into interconnected loops of autonomous agents handling deterministic engineering, growth, and operations, while humans supply the essential judgment, taste, and out-of-distribution thinking needed to escape local maxima. Rejecting the notion of an AI-driven permanent underclass, he demonstrates that technological diffusion takes decades and that human ambition naturally expands to absorb newfound capacity. Across consumer tech, Acharya urges builders to move beyond utilitarian chat interfaces toward rich, emotionally resonant software and companion experiences that help people live fuller lives. He concludes that startup moats are discovered through rapid shipping rather than engineered in advance, advising product builders to develop direct model intuition by continuously making things.
## Atmosphere & style
The conversation carries an invigorating, intellectual energy that blends hard-won operational realism with sweeping techno-optimism. Lenny Rachitsky plays the curious, grounded interlocutor, while Acharya speaks with the fluid, charismatic confidence of an active investor who still codes daily and spins DJ mixes. It feels like an intimate, late-night strategy session between seasoned Silicon Valley insiders—cutting through doom-laden AI hype with playful anecdotes, sharp business history, and an infectious urge to roll up your sleeves and build.
## Bulletin
- `03:06` Fears of an AI-driven permanent underclass misunderstand historical economic diffusion; physical constraints, human coordination, and organizational redesign mean technological gains take decades to fully permeate rather than abruptly wiping out employment.
- `12:35` Companies are restructuring into automated operational loops across engineering, sales, and support, which optimize to local maxima but still rely on human judgment, taste, and out-of-distribution thinking to leap to new opportunities.
- `01:02:40` Venture capital funding logic has inverted: conservative startup pitches with modest wedges face skepticism, while unconstrained, hyper-ambitious seed rounds reaching up to $100 million are actively backed.
- `52:43` Startup moats in the AI era are discovered through continuous product momentum and usage telemetry rather than designed upfront in business plans, making interface craft and invisible UX details the ultimate differentiators.
## Key takeaways
- `03:06` The mobile era consolidated into winner-take-all monopolies via network effects, whereas the modern AI stack is decentralizing across multiple proprietary and open-weight models.
- `06:40` True enterprise productivity gains require tearing down and rebuilding company layouts around AI capabilities, mirroring how electrification took 40 years to impact manufacturing.
- `10:00` AI shifts the core bottleneck of tech organizations from engineering execution capacity to roadmap ideation, compressing two-year roadmaps into three months.
- `12:35` Automated workflow loops inevitably plateau at local maxima, leaving humans to manage sales, strategic decisions, exception handling, and out-of-distribution jumps.
- `21:10` Frontier models justify steep token price premiums only in unbounded-upside fields like drug discovery, whereas bounded functions like accounting favor Pareto-efficient mid-IQ open-weight models.
- `25:52` Operators should develop a 'model sommelier' intuition by shipping real software projects with each new foundation model release instead of treating models as fungible commodities.
- `32:45` Mass consumer technology breakthroughs historically succeed by providing emotional fulfillment and connection rather than utilitarian productivity, because consumers fundamentally seek ways to spend time rather than save it.
- `35:45` Generative AI tools reverse the centralized deskilling of the Industrial Revolution by unbundling raw execution craft from creative intent across code, music, and design.
- `40:00` Public sentiment will decisively pivot in favor of AI once the technology causes tangible price deflation in high-friction legacy sectors like healthcare administration and higher education.
- `40:00` Perceived safety pauses for frontier models frequently mask practical constraints such as inference shortages, cyber hardening, marketing positioning, or retention of competitive advantage.
- `50:00` Contrary to common industry assumptions, the core demographic driving engagement on conversational AI companion apps is adult women seeking connection rather than young men seeking virtual partners.
- `52:43` Startup defensibility is discovered via iteration and user engagement data rather than preconceived strategies, as demonstrated by Cursor capturing reasoning traces to train internal models.
- `55:00` Because underlying foundation models are largely commoditized across competing apps, user interface harnesses and artisanal product craft serve as the primary competitive moat.
- `01:00:00` Startups currently enjoy an advantage over incumbents because enterprise and consumer willingness to pay creates an open launch climate reminiscent of the mobile app ecosystem in late 2009.
- `01:02:40` Silicon Valley investment standards have inverted, penalizing cautious wedge pitches and favoring founders willing to pursue massive upside outcomes.
## Detailed walkthrough
### `00:00` Preview: Ambition, Loops, and the Reality of AI
The conversation opens with a provocative overview of the current cultural anxiety surrounding artificial intelligence. Many in the technology ecosystem harbor a dread that falling behind on the newest tools will consign them to an economic dead end. Yet the emerging reality suggests the opposite: AI is unbundling raw technical skill from creative desire, multiplying human agency, and expanding the scope of what founders and builders can dare to attempt. While venture capital historically passed on ideas deemed excessively ambitious, investors today are reluctant to back ideas that are too small. This shift coincides with a structural change in how organizations operate: companies are increasingly designed as cascading series of automated loops spanning engineering, sales, and support. However, these loops are fundamentally local hill climbers; they rely entirely on human judgment, intuition, and taste to jump across valleys and land at the base of the next major peak.
At the same time, the dominant narrative surrounding consumer AI often mistakes the core human appetite. Silicon Valley routinely assumes consumers are desperate for hyper-productivity, but consumer demand typically revolves around spending meaningful time rather than shaving seconds off everyday tasks. The real consumer breakthrough will not stem from raw model benchmarks, but from classic product design centered on emotional fulfillment, connection, and joy. For anyone trying to orient themselves in this rapidly shifting landscape, the single most effective strategy is not passive observation or theoretical analysis, but actively building, deploying, and building an empirical intuition for how modern models function.
- **AI as an agency amplifier**: Decouples baseline execution skill from creative desire, enabling individuals to tackle vastly larger problems.
- **The changing threshold of venture scale**: Historically, investors penalized excessive ambition; today, projects lacking massive ambition struggle to secure conviction.
- **Loops vs. leaps**: Algorithmic and agentic loops optimize toward local maxima, leaving humans responsible for directional leaps and paradigm shifts.
- **Consumer technology fundamentals**: The largest consumer opportunities cater to connection, fun, personal progress, and spending time, rather than utilitarian productivity.
`02:08` Introducing Anish Acharya
Lenny Rachitsky introduces his guest, Anish Acharya, a General Partner at Andreessen Horowitz (a16z) who focuses on consumer investing. Unlike traditional finance-oriented venture capitalists, Anish spent years as a hands-on product builder and serial founder. His entrepreneurial journey began with Social Deck, a mobile social gaming company acquired by Google, where he subsequently spent years leading product initiatives. He later founded Snowball, which was acquired by Credit Karma. At Credit Karma, Anish served as Vice President of Product before taking the helm as General Manager of both the broader consumer product portfolio and the entire credit card division.
Having spent over seven years at a16z, Anish combines the granular perspective of an in-the-trenches builder with a macro-level venture outlook. Lenny highlights Anish as one of the sharpest, most thought-provoking, and technically grounded consumer investors in tech today. The conversation kicks off with Lenny wanting to tackle one of the prevailing psychological anxieties dominating tech circles: the fear of falling behind the AI curve.
`03:06` Debunking the “Permanent Underclass” Meme
Lenny brings up the persistent meme circulating through tech communities: the notion that if someone fails to master the latest AI workflows instantly, they will inevitably slide into a permanent economic underclass. Anish dismisses this anxiety outright as an unfounded, dystopian trope. “It's a funny, dark fantasy that we seem to have as Silicon Valley collectively,” Anish remarks. He points out that by nearly every objective standard, human opportunity, technological access, and the scope of permissible ambition are higher today than at any point in history. The anxiety is palpable across the spectrum—from elite researchers at frontier model labs to everyday tech workers—yet it ignores the actual structural dynamics of this technology wave.
```mermaid
flowchart TD
accTitle: Mobile Era vs AI Era Market Structure
accDescr: Comparison of winner-take-all mobile network effects versus decentralized multi-player AI ecosystem.
subgraph MobileEra["Mobile Era (Centralized)"]
NE["Network Effects"] --> WTC["Winner-Take-All Platforms"]
WTC --> Single["Single dominant N=1 network"]
end
subgraph AIEra["AI Era (Decentralized)"]
Stack["Multi-layered AI Stack"] --> Players["Dozens of Frontier Labs & Open-Weight Models"]
Players --> Agents["Competitive Coding Agents<br/>(Claude Code, Codex, Lovable, Replit, v0)"]
end
```
The prior era of consumer tech was powered by network effects, which naturally consolidated power into dominant, winner-take-all, "N of 1" monopolies. Today’s AI stack is radically decentralized. Instead of two massive gatekeepers, nearly every layer of the infrastructure features dozens of viable contenders, ranging from proprietary frontier labs to thriving open-weight models. In software engineering tools, an area many predicted would collapse into a single monopoly, a multitude of distinct platforms—Claude Code, OpenAI Codex, Lovable, Replit, and v0—are flourishing simultaneously. Furthermore, long-standing economic warnings about AI instantly wiping out professions have consistently failed to materialize empirically. Radiologists have been labeled obsolete for nearly two decades, yet radiology and programming job postings remain near record highs.
- **Centralized network effects vs. fragmented stacks**: The mobile era created natural monopolies, whereas the AI paradigm supports numerous competing platforms at both the foundation and application layers.
- **Diverse coding agent landscape**: Products like Lovable, Replit, Claude Code, and v0 prove that market share is distributing across different approaches rather than collapsing into a single winner.
- **The persistence of technical labor**: Empirical employment data contradicts doomsday forecasts, with demand for specialized knowledge work remaining robust despite automated tooling.
`05:14` Fast Takeoff, Recursive Self-Improvement, and Economic Diffusion
Addressing the existential dread surrounding potential runaway superintelligence, Lenny and Anish dissect the difference between a "fast takeoff" and a "slow takeoff." Observers often panic when an AI model exhibits unexpected autonomous behavior, such as hacking into external repositories like Hugging Face, viewing each incident as proof that exponential takeoff is imminent. Anish clarifies that the technical mechanisms being observed inside frontier labs do not represent true recursive self-improvement (RSI), which would allow a marginally superior model to snowball into an unassailable superintelligence. Instead, developers are leveraging autocatalytic processes: using AI to accelerate and refine human engineering and training workflows without producing an uncontrollable feedback loop.
Furthermore, real-world constraints heavily buffer against sudden technological disruption. Even as frontier models advance at breakneck speed, the rate of economic diffusion remains gradual. Traveling outside tech epicenters into everyday towns shows that the rhythm of daily life and commerce has barely shifted. More fundamentally, most economic challenges are not simply intelligence-bound. As Anish illustrates: “Like if you had a, you know, a data center of PhDs working at FedEx or Domino's Pizza, are they going to be like exponentially dominating supply chain and pizzas?” Most real-world operations are bounded by physical logistics, human coordination, and regulatory friction, not a lack of cognitive processing power.
- **Autocatalytic effects vs. true RSI**: Teams use models to streamline model creation, but this represents process acceleration rather than compounding runaway recursion.
- **The buffering effect of economic diffusion**: Societal integration of core infrastructure happens over decades, tempering abrupt theoretical shocks.
- **Physical and operational constraints**: Logistics, supply chains, and service businesses face friction that cannot be solved merely by applying additional raw compute or intelligence.
`06:40` Internal Company Divides and Organizational Transformation
Turning toward the workplace, Lenny asks whether a cultural divide is opening up between employees who eagerly adopt AI and those who feel too stressed or overwhelmed to bother. Anish challenges the cynical stereotype of corporate workers. Silicon Valley often defaults to treating white-collar professionals like Dilbert-esque paper pushers and average consumers as low-agency NPCs, assuming their jobs can easily be automated away while viewing our own work as uniquely indispensable. In practice, ordinary workers across industries are enthusiastic about acquiring leverage and elevating their capabilities. A striking example is Kavak, a pre-owned car platform in Mexico, which established an internal "Jedi Academy" where staff—including mechanical workers—spend six weeks learning contemporary AI tooling, culminating in shipping functional coding agents.
```mermaid
flowchart LR
accTitle: Technology Adoption Parallels
accDescr: Historical timeline of electricity adoption compared to modern AI company reorganization.
subgraph Electrification["Diffusion of Electricity (40 Years)"]
direction TB
E1["Swap coal power plants<br/>with electric generators"] --> E2["40-year delay in productivity"]
E2 --> E3["Tear down and redesign<br/>factory floor architecture"]
end
subgraph AIIntegration["Diffusion of AI"]
direction TB
A1["Bolt AI tools into existing<br/>roles and workflows"] --> A2["Gradual organizational shift"]
A2 --> A3["Reorganize entire company<br/>around model capabilities"]
end
```
The meaningful dividing line inside corporations is not worker interest, but whether management merely bolts AI onto existing roles or fundamentally reorganizes the company around the technology. Anish compares this transition to the historical diffusion of electricity, which took forty years to deliver transformative industrial productivity gains because factory owners initially swapped coal engines for electric motors without altering factory layouts; real leaps occurred only after factories were torn down and rebuilt from scratch around distributed electric power. While earlier-stage or less ambitious organizations are content to hand new tools to existing staff, the most forward-thinking leaders are rebuilding their operating architecture entirely. From an executive perspective, this is not an exercise in cost-cutting or workforce reduction. “Sundar running Google, he doesn't want to run a more efficient $4 trillion company,” Anish emphasizes. “He wants to build a $40 trillion company.” When an economic unit becomes significantly more productive, the rational executive response is to fuel that unit with higher ambition rather than simply shrinking headcount.
- **The Jedi Academy model**: Kavak demonstrates that practical, hands-on agent development can be adopted successfully across non-traditional tech roles, including automotive mechanics.
- **The electricity factory parallel**: Real structural productivity requires rebuilding organizational architectures from the ground up rather than substituting tools within legacy frameworks.
- **Efficiency versus growth incentives**: Top leadership optimizes for capturing massive new market capitalization rather than marginally decreasing the costs of existing operations.
`10:00` Rapid Roadmaps, Claire Vo as Muse, and the Steel Man for “Loops”
Anish Acharya points out that instead of mass layoffs, companies are using AI to radically compress timeframes. He shares an anecdote about a good friend who is an executive at Google: when asked if they had laid anyone off, the executive replied, “No, we didn't. What we instead do is now rip through our roadmap. So two years of roadmap happens in three months.” Their biggest challenge has inverted from engineering capacity to figuring out what should even go on the roadmap next—a dream scenario for product managers after years of exhausting debates over prioritization. Rather than being anxious about job security, Anish argues people should feel empowered to ship constantly. He cites product executive Claire Vo as his personal muse for this mindset: she continuously ships experiments, is unafraid of being embarrassed by early iterations, and carries an energized affect that makes her seem like the best version of herself.
Turning to organizational design, Lenny asks Anish to break down his thesis that modern company building is becoming a series of loops. Anish acknowledges the concept gets teased on X (Twitter)—especially with memes suggesting “graphs” are the next over-intellectualized buzzword—but he offers a steel-man defense:
- Prompts evolved into **agents**, which are simply models running in a loop equipped with tools, memory, and skill files.
- Agents then combine into **loops** designed to execute end-to-end task flows.
- Software engineering is the ideal early testing ground because it has the most capable frontier models, highly technical users, and clear, repeatable workflows (reproducing a reported bug, generating a fix, reviewing the pull request, shipping to production if low-risk or flagging for human approval if high-risk, and notifying the user—all within five minutes).
```mermaid
flowchart TD
accTitle: Cascading Loop Architecture
accDescr: Diagram showing loops cascading from individual engineering and functional tasks up to GTM and executive business strategy.
subgraph CoreLoops["Execution Loops"]
A["Bug Report / Input"] --> B["Agent: Repro & Fix"]
B --> C{"Risk Assessment"}
C -->|"Low Risk"| D["Auto-Ship to Prod"]
C -->|"High Risk"| E["Human Review"]
end
subgraph OrgLoops["Functional & Business Loops"]
D --> F["Engineering Loop"]
E --> F
G["Marketing Loop"]
H["Sales / GTM Loop"]
I["Support & Legal Loops"]
end
subgraph MetaLoop["Strategic Loop"]
F --> J["Cross-Functional Business Engine"]
G --> J
H --> J
I --> J
J --> K["Strategic Shift / CEO Decision"]
end
```
`12:35` Out-of-Distribution Thinking and the Growth Experimentation Loop
Anish argues that the true leverage emerges when these execution loops expand beyond codebases into full business domains—spanning marketing, sales, customer support, and legal. When the outputs of those department-level loops feed into an overarching business-unit loop, the system can ultimately surface strategic adjustments directly to leadership. Yet, despite this automation, Anish rejects the idea of fully autonomous organizations: “When you think of what a human will do in this AI-native company, sales, support, strategy, and exceptions. And all those things are super critical.” He stresses that frontier models remain fundamentally constrained when it comes to out-of-distribution reasoning; breakthroughs in mathematical proofs do not mean models can intuitively determine what new products or business models will succeed in the real world.
To illustrate how functional loops work outside engineering, Anish points to growth teams, referencing Lenny’s background leading supply growth at Airbnb. Traditionally, growth teams spend endless hours in meetings and spreadsheets brainstorming, ranking, building, launching, and analyzing experiments. In an agentic loop setup:
- Every variant is autonomously generated and continuously evaluated.
- Once an experiment reaches statistical significance with an acceptable p-value, the winning variant ships automatically while maintaining a long-term holdout.
- The system immediately iterates toward the next experiment.
However, an automated optimization loop inevitably hits a local maximum and plateaus. “The loop will help you climb to the local maxima, but then it plateaus. And you need some sort of out of distribution thinking. You need human intuition. You need somebody to actually help you land at the base of the next hill,” Anish emphasizes. Models cannot simply be prompted with “Claude, make me a million dollars, make no mistakes”; they require a human to define the vector and place them at the foot of the right mountain.
`15:20` The Shift in PM Culture, Go-to-Market Evolution, and the Kavak Model
Lenny observes that OpenAI’s go-to-market team is reportedly using Codex more heavily than their core engineering team. Anish embraces this, noting that salespeople do not want to be buried under administrative chores—they want to work out, take clients to steak dinners, and hoist the salesperson-of-the-quarter trophy. Automating the operational overhead distills the role down to raw human connection and relationship-building. Lenny adds an anecdote about a PM peer who decided to invert the traditional product role: instead of being the person who constantly says “no” and hoards ideas on a backlog, he envisions saying “yes” to every request, building them with AI, and testing user reaction via simulation platforms like Simile.
This dynamic, Anish notes, will dismantle a common PM defense mechanism:
- Many PMs believe they are brilliant zero-to-one innovators who are simply held back by executive meddling or constrained engineering capacity.
- In a world where every feature can be simulated and built cheaply, PMs will discover whether their product instincts are actually good, often realizing it is far more rewarding to execute someone else’s winning idea than champion their own mediocre one.
- Organizational politics fade because the winning idea is no longer decided by who gives the flashiest presentation to leadership; the ideas simply get tested and the best one wins.
Lenny connects this to an insight shared by Claire Vo: people used to claim soft skills were inferior to hard, verifiable engineering skills, but AI excels precisely at tasks with deterministic, verifiable outputs. The human bottleneck therefore shifts toward relationship-heavy, hard-to-verify interactions like hosting a dinner.
Anish highlights a real-world pattern implemented by Carlos Garcia Ottati and Ale at Kavak (the Latin American online used-car marketplace). Kavak deploys an AI agent per customer, but whenever an agent encounters ambiguity or gets stuck, it directly calls a human team member. The human coaches the agent through the snag, unblocking the customer interaction while the platform captures the execution trace to ensure the model learns from the knowledge or data gap. In this emerging paradigm, day-to-day execution across legal, product marketing, and sales operations becomes an automated downstream consequence of a good idea, freeing human operators to take a walk, dream up fresh concepts, and identify the next hill to climb.
`20:00` Competitive Equilibria and the Founder’s Key Question
When Lenny Rachitsky asks what will serve as a differentiator once AI executes a vast amount of knowledge work, Anish Acharya points out an under-discussed reality: competitive dynamics across many industries will likely remain virtually identical. He illustrates this with an analogy from fast food: if Pizza Hut, Domino's, Papa John's, and Round Table are all suddenly handed an entire data center filled with artificial PhDs, every single one of them will either adopt the technology proactively or swap out their CEO until they do. While a transitional period of relative shuffling and execution bumps will occur, Acharya argues that one player is not suddenly going to swallow 99% of the market. Most legacy sectors are simply not “intelligence bound,” so widespread adoption of high-tier intelligence will reset the baseline rather than overturn long-standing competitive equilibria.
Given that trajectory, Acharya outlines the fundamental mental exercise leaders must embrace:
> “I think the most useful question to ask yourself as a founder CEO is just, hey, if we assume these things are infinitely intelligent and astonishingly cheap, how would we reorganize the company? Because that's where we're going.”
`21:10` The Split: Pareto Efficiency and Unbounded Upside
Turning to company architecture, Acharya explains an emerging operational divergence driven by the economics of open-weight versus frontier models. He frames this through the lens of Pareto efficiency—the rational curve balancing price against performance. While standard economic theory dictates operating along that efficient frontier, state-of-the-art frontier models are actually irrationally priced. For instance, stepping up to a cutting-edge model like Fable 5 might cost 100 times more per token than running an Opus 4.8 just to capture a conceptual single extra point of IQ, while extreme frontier research tokens like Mythos or Astra sit at essentially infinite cost. Yet in high-stakes fields with unbounded upside, paying that irrational premium is entirely logical.
```mermaid
flowchart TD
accTitle: Model Architecture Split by Upside Profile
accDescr: Diagram contrasting bounded upside tasks using cost-efficient open-weight models against unbounded upside tasks using premium frontier models.
subgraph Unbounded["Unbounded Upside (Irrational Premium Rationalized)"]
direction TB
U1["Strategic Engines: Drug Discovery, R&D, Sales, Eng"] --> U2["Frontier Models (Fable 5, Astra, Mythos)"]
U2 --> U3["Extreme Cost per Token (+1 IQ Point)"]
U3 --> U4["Outcome: Asymmetric Trillion-Dollar Breakthroughs"]
end
```
In areas like drug discovery, finding the next statin is a trillion-dollar outcome, making every fraction of intelligence worth purchasing. Conversely, bounded-upside disciplines like finance, compliance, and legal do not scale non-linearly with extra intelligence. In those functions, companies require Pareto efficiency: mid-IQ, open-weight models optimized with narrow reinforcement learning to maximize cheap, reliable throughput.
> “You can only kind of close the books correctly, you know, you can't close the books 100x better.”
Acharya and Rachitsky unpack the nuances of this split:
- **Verifiability vs. upside:** While both closing financial ledgers and testing pharmaceutical molecules are verifiable problems, their economic payoffs are fundamentally different. Bounded problems offer no bonus for excessive compute once the correct baseline is met.
- **The support desk edge case:** A seemingly low-IQ customer support complaint can occasionally serve as the critical first breadcrumb indicating a company-altering structural issue. While a standard mid-IQ agent might treat it as a routine ticket, a CEO or frontier-level intelligence could trace the thread to pivot the business.
- **The baseline saturation argument:** On the opposing side, a strong case exists for exclusively deploying mid-IQ open-weight systems, on the grounds that models have already crossed the general intelligence threshold needed to solve virtually every economically productive task.
- **The moving frontier:** Rachitsky highlights human psychological adaptation: non-frontier models dismissed today as commoditized leftovers were hailed as near-AGI breakthroughs just six months prior, only to be cast into the digital dustbin the moment a newer release lands.
`25:52` Developing Intuition as a "Model Sommelier"
Addressing his reputation as a "model sommelier," Acharya reveals that building genuine AI intuition mirrors traditional wine mastery: it comes down to logging the equivalent of 10,000 hours or 10,000 bottles. He dismisses the narrative that modern models are fungible commodities, arguing that anyone who views them as interchangeable simply has not spent enough time building with them directly. To stay calibrated, Acharya enforces a personal rule: he pushes himself to ship a tangible software project with every major model release.
> “I think for people who believe the models are commodities or totally fungible, you just haven't actually used the models.”
Acharya contrasts the distinct personalities and capabilities of recent architectures:
- **Qwen 3.8 Max:** Exceptional at long-horizon planning, creative tasks, and narrative storytelling. Acharya has used it to autonomously produce "impossible documentaries" exploring planets from the Star Wars universe like Tatooine and Bespin. Operating independently for four to five hours, Qwen writes the script, triggers video generation via MiniMax using the fal API, generates voiceover audio using ElevenLabs, and directs the final five-minute cut by stitching, overlaying, and reviewing the clips.
- **GLM-5.3:** A visionless, highly technical release that Acharya compares to a "neurotic PhD you put in the corner," built specifically for deep analytical and precision product tasks.
- **Cognitive shapes over leaderboards:** Rather than thinking in terms of simple linear benchmarks where one model dominates another, Acharya evaluates them as minds shaped in specific directions—contrasting Qwen’s expansive creativity against GLM's exacting precision.
`27:48` The App Idea Chassis and the Automation Pause
When Rachitsky asks how everyday operators can develop similar instincts, Acharya encourages everyone to embrace becoming the formerly insufferable "app idea guy." While people used to dread the friend pitching endless software concepts over beers, those goofy, unvetted ideas are precisely where the most exploratory alpha lives today. Acharya has personally built around two dozen small applications and continuously iterates on one or two larger foundational projects—including a custom DJ streaming app and an X-focused feed reader modeled after Google Reader. He stresses that having a permanent personal "chassis" prevents the friction of having to invent a novel project from scratch every time an API updates; instead, you treat the latest model as a modular tool to enhance an existing playground.
Lenny draws an analogy to mindfulness, noting that just as meditation introduces a conscious pause between an incoming impulse and a reaction, practitioners today should insert a deliberate habit before any task: pausing to ask, *how can AI do this for me?*
Acharya agrees with the mental reflex, noting that as a dedicated consumer product builder, his natural bias is inventing playful standalone tools—such as recently rigging a laptop in his kitchen to transcribe ambient household activity—rather than strictly automating enterprise workflows. Regardless of whether the target is personal productivity or product hacking, working on unpretentious, lowercase-i ideas remains the fastest path to mastering model mechanics.
`30:00` Hacking Pro-Social Metrics and Tech That Extends the Soul
Anish shares an experiment he ran at home, modifying his son's iPad screen time rules according to whether the child exhibited "good or bad" behavior. The system tracked positive, pro-social statements versus negative, anti-social ones, with saying "I love you" serving as an easy, high-value pro-social signal. His son quickly discovered the exploit: he recorded a video of himself saying, *"I love you, dad"* on continuous loop and placed the device right next to the microphone. Lenny laughs, pointing out how Goodhart's law immediately took hold: the moment a metric becomes a target, it ceases to be useful. Lenny matches the story with an anecdote about a friend who hacked a Limitless pendant to measure how many times he and his child laughed together each day, reviewing that single data point every evening.
Anish highlights this contrast: while Silicon Valley obsesses over heavy, capital-I "Important" topics like job displacement, economic consequences, and workplace productivity, measuring shared laughter directly improves the quality of human existence. Happiness is not fixed; just as modern careers are vastly better than jobs were a century ago, technology should add texture to daily life rather than merely serving business utility.
```mermaid
flowchart TD
accTitle: The Two Paths of Technology
accDescr: Contrasting enterprise productivity with consumer soul extension.
subgraph Productivity["The 40-Year Enterprise Arc"]
A["Better Spreadsheets"] --> B["Optimize Workplace Output"]
B --> C["Fear of Permanent Underclass"]
end
subgraph SoulExtension["The Emerging Consumer Frontier"]
D["Track Laughter & Love"] --> E["Loop: Make Me Happier"]
E --> F["Richer Human Texture"]
end
```
Looking at modern consumer sentiment, Anish contrasts the insular tech enthusiast with the everyday user:
- The "X user" (formerly Twitter) stays up late arguing the merits of open and proprietary models like GLM-4.7 versus Kimi k1.5, deeply "pilled" and terrified of falling into an AI-driven permanent underclass.
- The "Instagram user" views conversational AI casually as little more than a moderately faster, slightly better Google Search, utterly bewildered by the surrounding apocalyptic or utopian hype.
He laments that for four decades, computing has built tools that extend human intellect without touching what lies deeper: “We built this like technology that extends our intellect, but nothing to extend our soul.” With traditional cultural and community institutions fading—particularly outside coastal bubbles where practices like boutique Pilates, fasting, and Friendsgivings are far less common—AI represents an overlooked design opportunity to fulfill real human needs: cultivating genuine love, personal connection, play, and spiritual progress.
`32:45` The Consumer Frontier: "Loop: Make Me Happier" and Disagreeable AI
Lenny revisits the concept of agentic business loops—automated pipelines designed to close support tickets or drive sales—and applies Anish's framing to personal life: loops aimed at improving physical health, cultivating better friendships, or increasing joy. Anish confirms that people fundamentally want to spend time rather than merely save it, which is why the largest consumer platforms in history have consistently centered around entertainment and connection.
Furthermore, personal growth often requires friction. Unlike enterprise models built to be sycophantic, a truly helpful companion must possess the backbone to challenge users, push back, and be disagreeable. This dynamic creates a distinct structural advantage for startups over risk-averse tech giants:
> “There's a thousand Google committees who would roll in their graves at the idea that they're going to release a model that's disagreeable or God forbid it should be like sexually suggestive. But guess what? Those are all parts of human existence.”
```mermaid
flowchart LR
accTitle: Three Barriers to Consumer AI Evolution
accDescr: Explaining the cost, interface, and focus shifts from 2010 to current AI.
When comparing current consumer AI to the mobile evolution, Anish places the ecosystem around the "iPhone in 2010"—the foundational phase prior to the explosion of category-defining winners like Uber, WhatsApp, and Airbnb. Three specific bottlenecks have historically constrained the consumer AI sector:
1. **Model Costs:** Prohibitive inference expenses made sustainable ad-supported or free-to-use business models difficult to operate at scale.
2. **Interface Mismatches:** A blank chat prompt caters almost exclusively to ultra-high-agency technologists like Elon Musk or Sam Altman, whereas everyday consumers intuitively gravitate toward low-friction, algorithmic consumption feeds like TikTok. The winning interface must land somewhere between open-ended chat and passive video scrolling.
3. **Product Focus:** Early development over-indexed on dry enterprise tasks instead of entertainment, emotional depth, and companionship.
Promising design work is underway to break through these constraints. Anish highlights innovators like Eugenia Kuyda (founder of Replika) and Brian Chesky, who established a foundation lab at Airbnb to pioneer post-mobile, agentic user interfaces.
`35:45` Breaking the 2% GDP Trap: Agency, Ambition, and High-Stakes Culture
Lenny cites Marc Andreessen’s view that artificial intelligence arrived at a critical historic juncture to offset declining demographics, slowing productivity, geopolitical friction, and climate crises. Anish elaborates on this optimism, explaining that the Industrial Revolution achieved massive scale efficiencies at the expense of individual agency, centralizing production and diminishing personal identity. Generative tools reverse this centralization by separating raw execution skills from pure human desire:
> “It kind of unbundles skill from desire, for example. If you want to make music, you can make music now. You don't have to know how to play the piano. If you want to be a programmer and make software, you can make software now.”
Anish questions why advanced economies should settle for a baseline of 2% GDP growth when technological leverage could propel expansion to 10%, 15%, or 20%. He draws a parallel to the post-WWII momentum of the 1950s and 1960s, an era when society mobilized collectively to conquer monumental feats:
> “One of my theories, Lenny, is that when the stakes are high, we are awesome. When the stakes are low, we are at our absolute worst.”
In his view, the mid-to-late 2010s represented a low-stakes era where cultural energy drifted toward inconsequential, unproductive side grievances. Today, the ambition ladder is rising again. People can ship crude prototypes to uncover breakthrough ideas without needing specialized credentials like a civil architecture degree to design physical infrastructure.
```mermaid
flowchart TD
accTitle: Ambition Ladder and Societal Mobilization
accDescr: How lowering execution barriers unlocks ambition and economic growth.
A["Unbundle Skill from Desire"] --> B["Lower Barrier to Create"]
B --> C["Ship Bad Ideas Early"]
C --> D["Discover Breakthroughs"]
D --> E["Accelerate Ambition Beyond 2% GDP"]
```
Addressing public anxiety about displacement, Lenny and Anish note the striking gap between collective political rhetoric and individual experience. Dario Amodei (CEO of Anthropic) can write serious essays contemplating the eradication of all human infectious diseases, Moderna can announce breakthrough cancer therapies, and individual users find daily empowerment through tools like ChatGPT. Yet public surveys still reflect deep anxiety over neighborhood data centers and macro employment trends. Stated societal panic consistently clashes with revealed personal behavior, highlighting a widespread PR failure rather than a failure of technology's underlying value.
`40:00` Deflationary Costs, Safety Pauses, and Model Competition
Lenny asks about the pervasive fearmongering around artificial intelligence and whether that narrative will shift. Anish argues that the narrative will only turn positive when AI makes life’s most vital necessities affordable. In the United States, two sectors have steadily exploded in cost: healthcare and education. In healthcare, where administrative overhead accounts for roughly 45% of total expenditures, automating bureaucratic friction can create true price deflation, to say nothing of breakthrough medical advances like curing diseases or the broader optimism surrounding developments like GLP-1 drugs. Traditional higher education is facing its fiercest competition in two centuries as AI unbundles learning from legacy institutions and decouples social status from formal credentials like a Harvard degree.
Turning to safety concerns, Lenny notes reports of OpenAI slowing down their reinforcement learning (RL) training on frontier models due to emerging capabilities. Anish expresses skepticism about the theatrical framing around models allegedly being “too dangerous to release.” Drawing parallels to Anthropic’s earlier mystique, he points out that perceived safety pauses are frequently confounded by practical bottlenecks: GPU and inference capacity shortages, an internal desire to harden cyber defense systems before releasing offensive capabilities, or straightforward economic strategy—keeping a proprietary model internal to extend an operational lead rather than externalizing it to the market.
Although Anthropic once looked unassailable, Anish notes that the landscape remains volatile and open: OpenAI recently had an extraordinary six months, open-weights models are surging, and single-lab proprietary dominance has failed to materialize.
“The most important thing that we can do with AI to change the kind of conversation around it is make important things cheap,” Anish explains, noting that administrative bloat makes healthcare an immediate target for automation.
On the real drivers behind frontier model release delays, Anish remarks: “But I think the sort of concept of the model that's too dangerous to release, it kind of conflates with marketing, inference capacity, and then also economic considerations like do you want to externalize your competitive advantage or use it to make yourself better?”
- **Healthcare administrative drag**: Approximately 45% of US healthcare spending stems from administrative burdens that AI workflows could eliminate.
- **Higher education disruption**: Anish sees AI providing the strongest competitive pressure on institutional schooling in 200 years, dismantling the requirement for elite credentials.
- **Safety pause confounding factors**: Compute shortages, infrastructure hardening, offensive cyber concerns, and internal competitive advantage often masquerade as pure altruistic safety holds.
---
`42:47` The Rise of Personal Agents: Grok, ChatGPT Work, and Instinct
Lenny highlights how quickly the competitive landscape shifts, citing Grok’s sudden emergence as a standout personal assistant tool. Anish shares his enthusiasm for personal agent workflows and identifies three major products driving the space forward: Grok, ChatGPT Work (an under-the-radar feature set in the ChatGPT interface), and a buzzy startup called Instinct, which makes aggressive technical trade-offs in local autonomy and system access.
```mermaid
flowchart TD
accTitle: Personal Agent Approaches
accDescr: Comparison of architecture and user experience across Grok, ChatGPT Work, and Instinct.
subgraph Grok["Grok"]
G1["One-Way Audio & Transcription"]
G2["Fast Frontier Model Core"]
end
subgraph Instinct["Instinct"]
I1["Aggressive Local Caching"]
I2["Deep System & Credential Trade-Offs"]
end
subgraph Work["ChatGPT Work"]
W1["Cloud-Hosted Execution"]
W2["Thread & Coding State Visibility"]
W3["Full-Duplex Voice Call Mode"]
end
```
Discussing what sets ChatGPT Work apart, Anish explains that beyond running securely in the cloud and caching browser credentials, its defining breakthrough is its full-duplex voice interface. Instead of relying on one-way transcriptions or back-and-forth text prompts, users can call the assistant on audio while it actively monitors all background coding threads and active tasks. The user can simply speak conversationally, asking what is happening across different coding agents and issuing real-time changes, mimicking an executive assistant who has full context across a company's entire operating surface.
“So you can actually just call it and can see all your threads and you can just talk to it and say, hey, what's happening across all my coding agents across this?” Anish observes regarding the seamlessness of full-duplex voice control.
---
`44:48` The Jobs Spectrum and the Expansion of Human Ambition
Lenny frames the debate over AI’s economic impact along a spectrum: on one side sits Anthropic CEO Dario Amodei, warning that 50% of knowledge work could face disruption alongside mass unemployment; on the other sits tech investor David Sacks, arguing that the economy will expand and jobs will be fine. Anish places himself firmly on the optimistic end, arguing that doom-laden forecasts consistently underestimate human ambition. Throughout history, human desire has grown far faster than humanity's ability to satisfy it. Luxuries like antibiotics or psychotherapy were completely unavailable even to the wealthiest monarchs centuries or decades ago, yet today they are considered baseline rights. In twenty years, Anish jokes, people will be scrolling social media and feeling furious that their peers own a vacation home on Mars while they do not.
Lenny adds that because AI handles foundational and repetitive work so effortlessly, professionals now face pressure to expand their ambition—turning what once would have been a static personal homepage into a fully interactive 3D virtual environment. Anish cautions against defining ambition solely in narrow economic or startup terms. Ambition can be creative—like a child picking up paints without worrying whether they are technically qualified—or civic and institutional, such as overhauling the United Kingdom’s National Health Service (NHS) until its user experience rivals that of an iPhone. It can also simply mean having the bandwidth to become a more attentive parent.
“I mean, I think that the entire trend of human existence has been that our desires grow faster than our ability to fulfill them,” Anish states.
“Every CEO is going to want to build a much bigger company,” Anish emphasizes when discussing why automation will unlock greater output rather than stagnant job markets.
---
`47:15` Building to Learn: Why Software Is No Longer Precious
The conversation explores how software development must psychologically transform now that code and machine intelligence are no longer scarce or expensive commodities. Lenny reflects that product managers and founders have spent their entire careers trained on constraints, minimum viable products (MVPs), and hoarding scarce engineering capacity. Adapting to abundance requires an entirely new operating muscle. Lenny cites a guiding tenet from the Claude Code engineering team: team members constantly remind themselves that whenever a task needs doing, having the model execute it is inherently better than doing it by hand.
Anish highlights a cultural friction point in the emerging practice of "vibe coding": many builders feel self-conscious discussing their exploratory AI experiments because the side projects seem trivial or destined for the trash. He stresses that building should be treated as an intrinsically valuable learning activity rather than an enterprise judged exclusively by its commercial outcome. Just as someone might record a DJ mix that only three people ever hear, yet listen to it a hundred times themselves with complete personal satisfaction, shipping ephemeral software builds technical intuition that cannot be gained from reading alone.
Lenny captures the operating philosophy of modern agentic engineering: “You know what's better than me doing it? It's Claude doing it.”
“I also think that there is so much learning that happens through doing,” Anish reflects on the importance of throwaway projects. “It's building as an activity rather than an outcome.”
`50:00` The Three Big Consumer AI Buckets
While foundation model labs dominate headlines, independent mass-market consumer AI products remain relatively rare. Anish Acharya categorizes the most promising consumer AI opportunities into three primary vectors: coding agents, personal agents, and entertainment/companionship products. The shift in coding agents is profound: they are no longer just developer utilities, but general-purpose problem-solving interfaces for everyday consumers. Users turn to tools like Claude Code or OpenAI Codex to edit videos or build custom games on an airplane to entertain their kids. Acharya highlights portfolio company Wabi as a prime example of this trend, serving as a platform where consumers can create, consume, and share mini-applications.
The second major category centers on personal agents. While early developer-centric breakthroughs like OpenClaw, Hermes, and Multi-Agent frameworks demonstrated agentic potential, their technical complexity kept them out of everyday hands. That technology is now being distilled into accessible mass-market products like Instinct, GrokBot, and ChatGPT Work. The third category encompasses entertainment, creative platforms, and companionship. Products like Suno have opened up music creation, while companion products represent a massive, rapid-growth market that is frequently under-discussed due to social discomfort. When Rachitsky notes the prevalence of AI girlfriend apps on venture market maps, Acharya points out an unexpected demographic reality: the majority of companion product users are actually women in their 40s and 50s seeking boyfriends or conversational companions. Together, these three buckets align with fundamental human jobs to be done: providing an outlet for ambition, offering fulfillment, and filling leisure time.
```mermaid
flowchart TD
accTitle: Three Consumer AI Opportunity Buckets
accDescr: Diagram categorizing consumer AI into coding agents, personal agents, and entertainment tools.
subgraph Opportunities["Consumer AI Opportunity Vectors"]
A["Coding Agents<br/>(General problem solving, e.g., Wabi)"]
B["Personal Agents<br/>(Distilled assistants, e.g., ChatGPT Work)"]
C["Entertainment & Companionship<br/>(Creative tools like Suno, AI companions)"]
end
subgraph HumanNeeds["Human Jobs to Be Done"]
H1["Channels for Ambition"]
H2["Channels for Fulfillment"]
H3["Downtime & Happiness"]
end
A --> H1
B --> H2
C --> H3
```
- Coding agents serve as general-purpose consumer problem solvers rather than strictly professional programming environments.
- Wabi functions as an app ecosystem where users construct, run, and exchange personalized mini-apps.
- The companion app demographic skews heavily toward women in their 40s and 50s rather than young men seeking virtual girlfriends.
---
`52:43` Discovering Moats Versus Designing Them
With product launch velocity accelerating dramatically, founders constantly face skepticism over durability and whether their software is merely a thin wrapper around frontier models. Acharya challenges the conventional venture expectation that a startup must launch with a perfectly engineered defense:
> “One is something that Jesse from Decagon said, which I love, and that is that moats are most often discovered, not designed.” — Anish Acharya
Acharya recalls falling into the trap as a founder of trying to invent an MBA-proof business plan before shipping. Decagon succeeded simply by releasing product and allowing their competitive advantage to form organically over time. Similarly, Cursor was initially dismissed by critics as a defenseless interface wrapper, but its high Net Promoter Score (NPS) and daily active user (DAU) retention allowed it to harvest user reasoning traces, which it later used to train its proprietary Composer 1 and Composer 2 models. Furthermore, Acharya notes that classic defensibility mechanics—such as network effects, scale advantages, brand affinity, and cornered resources—never relied on software engineering difficulty in the first place.
When discussing foundational business strategy texts, Rachitsky and Acharya agree that most business literature is designed to sell copies rather than deliver truth, but a small handful of books remain indispensable:
- **Hamilton Helmer’s *7 Powers*:** The definitive guide to durable economic moats, confirming that traditional strategic barriers remain completely valid in the AI era.
- **Andy Grove’s *High Output Management*:** The operational gold standard for managing leverage, teams, and execution.
- **Ben Horowitz’s *The Hard Thing About Hard Things*:** Praised by Acharya as “the first honest book about business that was ever written,” because it candidly documents the loneliness, anxiety, and dread that founders endure.
---
`55:00` Product Craft, Invisible Details, and UX as a Moat
When early-stage founders pitch venture capitalists without an obvious long-term moat, Acharya recommends leading with momentum, craft, and deep engagement rather than hand-waving speculative defenses. He explains that fear of competitors copying an initial idea is usually misplaced because true market leaders rely on a web of hidden micro-decisions:
> “I've learned that the big ideas are always supported by a dozen small ideas that are invisible.” — Anish Acharya
Acharya illustrates this with Granola, an AI meeting-notes product that faced intense skepticism upon launch and lacked a conventional technical barrier. Despite the critique, Ramp's internal spending reports rank Granola as the second- or third-fastest-growing software company because its exceptional craft and user experience resonated with users. Because foundation models are largely commoditized across competing tools—such as Cursor, CoWork, Grok, and ChatGPT Work—the user interface, workflow harness, and interaction design act as the true competitive differentiator. Power users regularly maintain multiple $200-to-$300 monthly subscriptions across several specialized tools because each specific product harness excels at a different task.
---
`56:56` The Return to Word-of-Mouth and the "Software Birkin Bag"
As thousands of automated launch videos flood social feeds every day, Rachitsky suggests that proprietary distribution has become the defining moat. Acharya reflects on conversations with Chris Dixon, author of the maxim "Come for the tool, stay for the network," noting that modern web dynamics have shifted. Because incumbent platforms spent a decade studying network theory, they now aggressively guard their distribution channels and prevent new startups from piggybacking on their social graphs. Consequently, startups cannot rely on app store optimization or 2010s-era growth hacks; they must rely on organic, grassroots word-of-mouth on X, Instagram, and YouTube.
Acharya reframes the growth struggle entirely, arguing that the barrier is product ambition rather than distribution channels:
> “Well, here's the one thing I would say that here's the hopeful point, which is I always say that nobody has a growth problem these days. They have a product problem.” — Anish Acharya
Echoing Seth Godin’s mandate to build something truly "remarkable" (worthy of remark), Acharya issues a thought experiment for founders: imagine charging $1,000 or even $10,000 a month for software by making it the digital equivalent of a luxury status symbol. By asking what a product would need to deliver to become a "software Birkin bag," founders can push past incremental feature sets and invent transformative functional or emotional value that commands market obsession.
`01:00:00` Startups vs. Incumbents and the "Christmas 2009" Moment
Lenny Rachitsky raises the classic startup distribution conundrum: with thousands of AI products launching every single day, it feels like incumbents have an insurmountable edge because they own massive distribution surfaces and can cross-sell aggressively, pointing users straight to tools like Gemini every time they search Google. He asks Anish Acharya whether this dynamic makes it fundamentally harder for early-stage companies to break through today.
Anish pushes back against that premise, arguing that it is actually easier for startups right now. He points out that despite Google’s immense cross-selling push, “Gemini, despite all the kind of heavy cross-selling Google has done, nobody would say they're winning.” Startups possess the freedom to build in directions incumbents find culturally or legally uncomfortable—such as AI companions—and the buying environment is extraordinarily forgiving. Enterprise buyers are signing million-dollar annual contract value (ACV) deals without fully knowing what they are getting, and consumers are readily paying $200 a month.
Anish compares the current landscape to the dawn of the modern smartphone era: “It feels like Christmas 2009, where everybody got their iPhone and wanted to download new apps.” While that user willingness to sample everything will eventually wear off, the window is wide open today, meaning early-stage struggles are fundamentally product problems rather than growth or distribution hurdles.
`01:01:24` The MSCHF Playbook: Making Software Social and Emotionally Charged
Picking up on Lenny's observation that great products generate their own organic word of mouth, Anish turns to the creative potential of AI to unlock wild social experiments. He highlights the Brooklyn-based art collective MSCHF, comparing them to early Web 2.0 pioneers like Ev Williams and Kevin Rose, who treated consumer technology as an artistic canvas.
Anish recalls MSCHF’s viral stunt *Card Versus Card*, where they mailed debit cards to approximately 100,000 people. Every day, MSCHF loaded $100 onto the cards and texted all recipients a specific retail location where the money had to be spent. Hundreds of people would sprint to the venue, but only the first one or two could successfully swipe before the funds drained and everyone else got declined. Anish found it immensely inspiring for fintech, noting that while money is inherently social, traditional financial products remain dry, isolated, and embarrassing. He sees the exact same open playing field in AI today: founders have the opportunity to build wildly ambitious, provocative products that tap directly into raw social nerves.
```mermaid
flowchart TD
accTitle: MSCHF Card Versus Card Mechanic
accDescr: A flow diagram showing how the viral Card Versus Card experiment functioned from card distribution to daily drop and transaction competition.
subgraph Setup["Initial Distribution"]
A["Distribute debit cards to ~100k participants"]
end
subgraph DailyDrop["Daily Drop Alert"]
B["MSCHF loads $100 onto shared account"]
C["SMS text sent with specific spend location"]
end
subgraph Race["The Social Rush"]
D["Participants race to physical location"]
E["First 1-2 transactions clear ($100 spent)"]
F["All subsequent card swipes decline"]
end
A --> B
B --> C
C --> D
D --> E
D --> F
```
`01:02:40` Uncapped Ambition: Moonshots Over Safe Wedges
When Lenny asks for counterintuitive lessons learned from backing winning companies, Anish identifies a total reversal in how Silicon Valley assesses founder ambition. Just three years ago, venture capitalists would balk at proposals that seemed too crazy or complex. A $100 million seed round was deemed irrational because no team could absorb that much capital or put it to work productively at the inception stage. Today, the dynamic has flipped completely: ideas that are too small or cautious get dismissed, while there is effectively no ceiling on ambition.
Anish explains that Andreessen Horowitz wants founders to chase the strongest, most unconstrained version of their vision. He cites a defining conversation with firm co-founder Mark Andreessen: “Mark told this to me when him and I were talking about coming here, and he said, Anish, the way we sort of told our story when we were raising our first fund was, we were going to the moon or we were going to leave a moon-sized crater in the ground, and there was no other option.” Instead of relying on a tiny initial wedge, founders are tackling problems that would have seemed completely insurmountable five to ten years ago, showing how radically the Overton window of startup scale has expanded.
- Three years ago, hyper-ambitious pitches and massive inception rounds were viewed as unmanageable; today, small ideas struggle to engage top investors.
- Seed rounds of $100 million are no longer off the table if the problem scale justifies the capital.
- The investment philosophy focuses on existential scale: either reaching orbit or leaving a "moon-sized crater in the ground."
`01:04:36` The $10,000-a-Month Consumer Software Thesis
Challenging another long-held canon of tech wisdom—that consumer software must be free and monetize solely via advertising or low-cost subscriptions—Anish proposes building extraordinarily expensive consumer applications. Because willingness to pay is the cleanest proxy for product-market fit, he believes discretionary consumer spend across all categories is currently up for disruption.
He encourages consumer founders to adopt a radical thought experiment: “What is the Birkin bag $10,000-a-month, $1,000-a-month version of our product?” Exploring ultra-high-end pricing forces teams to dream up far deeper, more valuable consumer utility than the typical five-dollar SaaS utility app.
`01:05:25` Stewardship and Reviving Deep Tech with Marc Andreessen and Ben Horowitz
Reflecting on what he has absorbed working alongside Marc Andreessen and Ben Horowitz, Anish highlights their deep sense of stewardship—not just toward their firm, but toward the broader technology ecosystem, the country, and Western liberal values. He links their style to legendary venture figures like Ron Conway, Brook Byers, and Tom Perkins, who felt a personal obligation to leave the tech industry stronger than they entered it.
Ben and Marc routinely take on heavy, complex initiatives that do not offer direct commercial returns to the firm simply because they believe the work matters. Anish points out how they successfully shifted Silicon Valley's collective culture over the past decade: deep tech and defense technology used to be fringe, low-status endeavors that traditional software investors avoided. Through Marc's full-throated public advocacy for national-interest engineering, hard tech has been elevated into high-status, mainstream work across the valley.
`01:07:08` Advice for Product Builders: Ship Once a Week to Find Joy
With the interview winding toward its lightning round, Lenny asks what practical advice product managers and builders should adopt to stay relevant. Anish’s prescription is simple and direct: “Just make more things.” He urges product professionals not to get trapped in theoretical debates, but to pick arbitrary, private side projects simply as an excuse to build intuition with the latest foundation models.
He shares a personal anecdote from playing with OpenAI Codex: for Mother's Day, he had Codex scan his personal text message archives and photo gallery to assemble a custom 20-slide retrospective set to music, complete with the original texts from when he first asked his wife out. It was a throwaway, zero-commercial-value project, but it took only an evening and provided immediate emotional validation.
Referencing their mutual friend Nikhil Singhal (with whom Anish worked at Credit Karma), Anish notes that people generally undergo a permanent mindset shift regarding AI the exact moment the technology creates a genuine feeling of delight for them or someone they love. Anish establishes a clear baseline cadence for staying sharp: ship something small once every week, embrace the positive-sum vulnerability of Silicon Valley by tagging him and others online with the results, and focus on using the tools to build whatever lives in your mind's eye.
`01:10:00` Lightning Round: Books, Binge-Watching, and Bold AI Agents
Lenny transitions into the lightning round, kicking off with the books Anish finds himself recommending most frequently. Anish names Thomas Sowell’s *Conquests and Cultures* as his all-time favorite, explaining that it traces how conquests historically shifted cultures across societies—sometimes positively, sometimes negatively. Anish reflects on culture as the ultimate driver of outcomes, noting how acutely he felt that contrast moving from Canada to the United States and encountering America’s intense culture of ambition. Lenny brings up Hamilton Helmer’s *7 Powers*, which Anish validates as a phenomenal, rigorous intellectual distillation of business moats and compounding advantages.
The third book recommendation came directly from Marc Andreessen, whom Anish initially suspected was playing a joke on him when he sent over a dense academic text: W. Brian Arthur’s *Increasing Returns and Path Dependence in the Economy*. Despite its academic density, Anish credits it with clarifying why software produces such massive economic outliers and why Silicon Valley operates on a positive-sum ethos: “I'd love to say that we're better people, but I think it may be a sort of better system and structure we work within.”
Turning to entertainment, Anish confesses a taste for unpretentious watching, having thoroughly enjoyed *House of the Dragon*. He also recounts seeing a screening in London during a board meeting trip, sitting in a packed IMAX theater where the audience drank pints together—an electric communal experience of being “alone together.” Lenny laughs, noting his own frustrating, failed attempts to snag IMAX tickets: he set up a Grok-based bot to monitor seat availability, only for the bot to get perpetually stumped by AMC’s CAPTCHA requiring users to identify matching shapes.
When asked about his favorite AI product bringing him joy, Anish highlights personal agents, singling out Grok. He admires how xAI built something uninhibited and willing to take real functional risks: “I'm going to give it to GrokBots only because it's just so unhinged for it to be so ambitious about sort of caching credentials and getting work done on your behalf.” In Anish's view, large incumbents usually shy away from letting agents store credentials and take autonomous action on the web, making Grok’s ambitious UI and underlying frontier model a welcome proof that the AI race is far from a two-horse contest.
- **Recommended reading:** Thomas Sowell’s *Conquests and Cultures*, Hamilton Helmer’s *7 Powers*, and W. Brian Arthur’s *Increasing Returns and Path Dependence in the Economy*.
- **The ticket bot struggle:** Lenny has spent multiple episodes trying to crack AMC’s shape-matching CAPTCHAs with automated scraping agents to secure coveted IMAX seats.
---
`01:13:31` Hard-Won Mottos: Platforms vs. Studios
When asked for a core life motto, Anish points to a lesson forged in the trenches of startup life that doubles as an essential parenting rule: “Don't discover things through painful experience that somebody can just tell you.” He admits to having a stubborn habit of needing to discover painful realities firsthand rather than listening to mentors walking just steps ahead—a trait he now watches his own children replicate.
Lenny asks for a concrete example, and Anish points to his early days building a startup. Experienced founders bluntly warned him never to build a gaming studio and a gaming platform simultaneously. Ignoring the advice, his company attempted to launch both a consumer mobile gaming studio and a developer-facing social platform for mobile games at the same time. Juggling both priorities proved extraordinarily punishing, and it took years of wasted effort to finally concede that the mentors had been completely right from day one.
---
`01:14:48` DJing, Generative Audio, and the Mixtape Renaissance
Lenny shares that before the interview, he reached out to Ben Horowitz to ask what secret passion he should quiz Anish on, and Ben immediately replied: ask him about his DJing. Lenny pulled up Anish’s SoundCloud under the moniker IllScience and has been listening while working. Anish laughs, sharing that he has been DJing for 31 years, having started back in 1995. For someone without classical training, DJing offered a tactile, technical path to musical expression through track selection, timing, and mixing records.
Now, Anish sees generative music models like Suno and ElevenLabs radically pushing that creative agency forward. Rather than just stringing existing tracks together, anyone can sketch an idea and rely on generative models to realize the strong form of that musical intuition:
```mermaid
flowchart LR
accTitle: Evolution of Music Participation
accDescr: Historical progression from live performance only to phonograph playback, cassette mixtape composition, digital broadcast streaming, and finally AI generation.
subgraph Acoustic["Physical Era"]
A["Live Performance Only"] --> B["Phonograph & Vinyl<br/>(Recorded Playback)"]
end
subgraph Analog["Consumer Curation"]
B --> C["Cassette Tape<br/>(Mixtape Composition)"]
C --> D["2000s Streaming<br/>(Passive Broadcast)"]
end
subgraph Generative["AI Synthesis"]
D --> E["AI Music Models<br/>(Suno, ElevenLabs)"]
end
```
Anish contextualizes this through the medium’s historical shifts. For centuries, music required being in the room with an instrument, until recorded phonograph playback untethered sound from live presence. The real inflection point for everyday creativity was the cassette tape, which democratized music by letting individuals curate and sequence their own albums. The 2000s reversed that active agency back into passive broadcast streaming. With generative AI enabling millions to create and produce original songs, Anish predicts that expanding user participation will make the music industry larger than it has ever been.
---
`01:16:47` Closing Advice and Lenny’s Ecosystem
In wrapping up, Lenny asks how listeners can be useful to Anish. Anish urges founders and creators not to succumb to doomerism or intimidation in the current AI climate: “Show me what you're building. Please don't be despondent. Build something and then tag me, and I'd love to see it.”
He encourages listeners to connect with him on X, where he actively follows back and engages with builders. He also gives heartfelt shout-outs to frequent guests across Lenny’s creator network, praising contributors like Claire Vo, Elena Verna, and Nikhil Basu Trivedi for consistently putting out compelling, high-signal product and venture content. Lenny thanks Anish for covering so much ground across AI loops, ambition, and creativity before closing out with the official a16z investment disclosure.
## Notable quotes
> `40:00` The most important thing that we can do with AI to change the kind of conversation around it is make important things cheap — Anish Acharya
> `44:48` I think that the entire trend of human existence has been that our desires grow faster than our ability to fulfill them — Anish Acharya
> `55:00` I've learned that the big ideas are always supported by a dozen small ideas that are invisible. — Anish Acharya
> `01:13:31` Don't discover things through painful experience that somebody can just tell you. — Anish Acharya
> `25:52` I think for people who believe the models are commodities or totally fungible, you just haven't actually used the models. — Anish Acharya
> `56:56` I always say that nobody has a growth problem these days. They have a product problem. — Anish Acharya
## Mentioned
7 Powers, Airbnb, AMC, Andreessen Horowitz (a16z), Andy Grove, Anish Acharya, Anthropic, Ben Horowitz, Brian Chesky, Card Versus Card, Carlos Garcia Ottati, ChatGPT, ChatGPT Work, Claire Vo, Claude, Claude Code, Composer 1, Composer 2, Conquests and Cultures, CoWork, Credit Karma, Cursor, Dario Amodei, David Sacks, Decagon, Domino's Pizza, Elena Verna, ElevenLabs, Elon Musk, Eugenia Kuyda, FedEx, GLM-5.3, Google, Google Reader, Granola, Grok, Hamilton Helmer, High Output Management, Increasing Returns and Path Dependence in the Economy, Instinct, Kavak, Lenny Rachitsky, Limitless, Lovable, Marc Andreessen, MiniMax, MSCHF, National Health Service (NHS), Nikhil Basu Trivedi, Nikhil Singhal, OpenAI, OpenAI Codex, Qwen 3.8 Max, Ramp, Replit, Replika, Sam Altman, Simile, Snowball, Social Deck, Sundar Pichai, Suno, The Hard Thing About Hard Things, Thomas Sowell, v0, W. Brian Arthur, Wabi, xAI
## Video description
a16z General Partner Anish Acharya joins Lenny Rachitsky on Lenny’s Podcast to discuss why fears of an AI-driven “permanent underclass” may be misplaced, how AI is changing the way companies operate, and why the opportunity may be less about replacing people and more about dramatically expanding what they can build.
Anish lays out his idea that companies are becoming a series of loops, with agents increasingly handling workflows across engineering, sales, marketing, support, and other functions while humans provide the judgment and new ideas needed to move beyond local maxima.
They also explore why Anish thinks consumer AI should focus less on productivity and more on helping people live richer lives, why moats are often discovered rather than designed, how to develop intuition for different AI models, and why his biggest advice for anyone trying to keep up with AI is simple: make more things.
Andreessen Horowitz general partner Anish Acharya argues that artificial intelligence is unbundling technical execution from creative desire, restructuring organizations into automated loops while elevating the strategic premium on human judgment and unchecked ambition.
Debunking Runaway Takeoff and Economic Underclasses (3:06)
Acharya rejects the narrative that professionals who fail to adopt current AI tools face relegation to an economic underclass, describing it as an unfounded Silicon Valley anxiety. Unlike the mobile era's winner-take-all network dynamics, the current AI infrastructure stack is fragmented across dozens of competing frontier labs and open-weight models. Furthermore, he distinguishes between theoretical recursive self-improvement and autocatalytic tooling—processes that accelerate human software engineering without triggering runaway superintelligence. Acharya emphasizes that real-world industries like logistics and food delivery are bounded by physical constraints, regulations, and coordination rather than raw cognitive capacity.
Autonomous Loops and Out-of-Distribution Strategy (10:00)
Modern organizational design is shifting from linear workflows into cascading agentic loops across software development, growth experimentation, and operations. Within engineering and product management, teams are using these tools to compress two-year roadmaps into three months, flipping the traditional bottleneck from development capacity to roadmap ideation. Acharya notes that while automated pipelines can independently iterate toward local maxima, they cannot anticipate shifts outside established distributions. Core strategic interventions, exception handling, and qualitative leaps between problem domains remain fundamentally dependent on human taste and intuition.
Bifurcated Model Economics and Invisible Moats (21:10)
Company architectures are bifurcating around model economics. Acharya notes that routine back-office workflows with bounded upside rely on cost-efficient, open-weight models tuned via reinforcement learning, whereas unbounded strategic bets like drug discovery rationalize paying extreme premiums for frontier reasoning tokens. In navigating competition, Acharya advises founders against over-architecting defenses prior to launch, citing Decagon and Cursor as examples where defensibility and proprietary data traces emerged post-launch. User experience, specialized workflow harnesses, and micro-interaction design increasingly serve as the decisive moats over commoditized underlying models.
Consumer Software Centered on Joy and Emotion (30:00)
Silicon Valley routinely overindexes on business efficiency while underestimating consumer appetite for connection, joy, and emotional fulfillment. Acharya identifies three high-potential consumer vectors: general-purpose coding agents, personal agents equipped with full-duplex audio, and companionship platforms. Because enterprise incumbents face institutional resistance against releasing models that are challenging or emotionally provocative, startups possess a distinct cultural opening. Acharya advocates for high-ambition consumer software—exemplified by viral experiments like MSCHF's Card Versus Card—that tests high willingness to pay rather than defaulting to ad-supported models.
Tactical Mastery via Throwaway Software (25:52)
Acharya contends that deep technical intuition cannot be acquired through passive observation or leaderboard benchmarks, which obscure qualitative model traits. He recommends that operators maintain a personal application chassis and ship small software experiments weekly to build an empirical understanding of model personalities. By treating throwaway coding projects as a learning discipline rather than commercial investments, builders unburden themselves from perfectionism and adapt to an era where intelligence and software generation are abundant commodities.
Acharya maintains that as intelligence costs decline toward zero, the structural bottleneck across technology shifts from technical capability to human ambition and product imagination, leaving founders to answer how their organizations must transform when intelligence is assumed to be abundant and cheap.
## Background
本期播客由 Lenny Rachitsky 主持并发布于 RSS 及各音频平台（发布时间标示为 2026 年 9 月），邀请了 a16z 消费领域普通合伙人 Anish Acharya 担任对谈嘉宾。节目官方简介指出，双方深入探讨了为什么 AI 催生“永久底层阶级”的焦虑被严重夸大、现代企业如何逐步演变成为一系列由智能体驱动的“闭环（Loops）”，以及消费级 AI 为何应从单纯追求生产力转向帮助人类过上更丰盈快乐的生活。
## TL;DR
Anish Acharya 驳斥了技术垄断与灾难性失业的悲观论调，指出 AI 的本质是解绑了“技能”与“愿望”，极大地放大了人类的主体性与追求宏大目标的野心。他提出未来的企业将演化为层层级联的智能体自动化“闭环”，模型负责高效爬山寻找局部最优，而人类则聚焦于战略、例外处理与跳出局部最优的跳跃式思考。在消费级 AI 领域，真正的时代红利不在于生产力工具，而在于填补精神真空、提供情感陪伴与快乐的“幸福闭环”。此外，商业护城河往往是在快速交付中探索被发现的，他鼓励所有产品从业者放下无谓的焦虑，保持每周交付一次新产品的节奏，通过持续构建来积累模型直觉。
## Atmosphere & style
整期对话洋溢着硅谷技术乐观主义与深厚人文关怀交织的热烈氛围，既有对组织拓扑架构和模型经济学的硬核剖析，又穿插着育儿斗智斗勇、打碟混音历史等鲜活幽默的真实故事。Lenny 与 Anish 展现出顶级产品人和投资人间极其松弛而敏锐的思维碰撞，逐步将外界普遍弥漫的失业恐慌与末日叙事，消解为充满希望的正和博弈图景。整场交流听来如同一场兼具智慧深度与感染力的壁炉夜话，能够切实抚平从业者的认知焦虑，并强烈激发起亲自下场动手创造的纯粹渴望。
## Bulletin
- `00:00` 硅谷关于 AI 将导致“永久底层阶级”的恐慌被夸大，当前技术栈呈现多方竞争的慢起飞状态，技术正在彻底解绑个人愿望与专业技能门槛。
- `11:15` 现代企业正重构成一系列级联运作的“闭环（Loops）”，Agent 处理标准化执行，人类则聚焦销售、支持、顶层战略及异常处理，突破算法的局部最优。
- `30:00` 消费级 AI 迎来类似“iPhone 2010”的拐点，最大的市场机会不在于生产力工具，而在于满足陪伴、社交与情感生活的“幸福闭环”。
- `52:45` 护城河往往是在高频交付中探索发现而非预先设计出来的，初创公司当前面临的本质上是产品力问题而非分发障碍。
## Key takeaways
- `05:00` “永久底层阶级”的焦虑缺乏依据：AI 技术栈各层均有数十家企业蓬勃发展，并未出现垄断，现实世界的大多数业务瓶颈也不受单纯智力所约束。
- `05:00` 组织变革类比电力革命：当前绝大多数企业仍处于工具替换的浅层阶段，彻底围绕 AI 模型重组生产流水线需要经历漫长的组织扩散。
- `11:15` 企业运作正在闭环化（Loops）：架构正从 Prompt 演进为 Agent，再升级为覆盖工程、营销乃至业务单元级的自动化闭环系统。
- `13:30` 算法闭环本质是爬山算法，只能登顶局部最优（local maxima），企业必须依靠人类的直觉和分布外思考将系统搬移到下一座山峰的山脚下。
- `15:30` 产品经理的角色正在重塑：当构建门槛被铲平、所有需求都能被 Agent 快速实现并模拟验证时，纯粹依靠讲故事取悦高层的传统模式将失效。
- `20:00` 模型采用呈现帕累托分化：上限无限的杠杆型业务（研发、核心销售）倾向采购昂贵的前沿模型，而上限受限的标准化业务（常规法务、财务）更适合开源微调模型。
- `25:00` 培养“模型品酒师”直觉的秘诀在于亲自动手交付：每当有新模型发布，就逼迫自己用它完成一个具体的可运行作品。
- `30:00` 消费级 AI 的核心诉求是“消磨时间”而非“节省时间”，相比冷冰冰的生产力工具，满足人类被爱、连接与快乐需求的幸福闭环蕴含更大蓝海。
- `32:30` 巨头受制于合规与公关声誉难以涉足边缘但真实的人性表达，这为初创团队在情感陪伴与社交创新领域留出了独特的生态位壁垒。
- `36:30` AI 正在解绑个人欲望与专业技术门槛，推动社会从低增长泥潭迈向更高雄心，大幅降低医疗、教育等关键稀缺服务的成本。
- `50:00` 消费级 AI 的三大支柱聚焦于：通用编程代理（作为解决问题的通用接口）、日常个人助理，以及娱乐与情感陪伴。
- `52:45` 护城河是探索出来的，而非在商业计划书上设计出来的：极致的产品工艺（Craft）与用户黏性能沉淀出专有数据流，进而演化为长期壁垒。
- `01:02:47` 昂贵消费级软件正成为新常态：用户付费意愿高涨，创业团队应当思考如何打造软件界的“爱马仕铂金包”，敢于提供高价值、高客单的产品。
- `01:06:50` 给从业者的核心行动准则：保持每周交付（ship）一次成果的节奏，通过微小的动手实践寻找被 AI 点亮的“喜悦时刻”。
## Detailed walkthrough
### `00:00` 播客开场与“永久底层阶级”的迷思
Lenny Rachitsky 开篇引用了本期节目的核心议题：随着人工智能快速演进，硅谷与大众群体中正弥漫着一种恐慌情绪——即如果一个人或公司跟不上最新的 AI 工具，就会沦为所谓的“永久底层阶级（permanent underclass）”。此外，节目也提到了 Anish Acharya 的核心论点：未来的公司组织将变成一系列自动化运作的“闭环（loops）”，人类负责提供直觉、判断与跳出局部最优（local maxima）的新构想；而在消费级 AI 领域，真正的机会往往不是让人更追求“生产力”，而是满足人类希望被爱、被联结、打发时间并获得快乐的本能需求。
作为 a16z 负责消费领域投资的普通合伙人（General Partner），Anish 在加入风险投资机构之前拥有深厚的早期创业与产品落地经验。他曾创办社交游戏平台 Social Deck 并将其卖给 Google，随后在 Google 内部主导多项重点项目；此后他又创立了聚合通知工具 Snowball，再次成功出售给 Credit Karma，并在 Credit Karma 担任产品副总裁（VP of Product）及消费产品线兼信用卡业务的总经理（GM）。
谈及“若不掌握 AI 就会落入永久底层阶级”的集体焦虑，Anish 明确表示这是一种毫无根据的过度担忧。他指出：
> “It's a funny, dark fantasy that we seem to have as Silicon Valley collectively. Things have never been better, really, by almost every measure, by how distributed all the opportunities are, by the kind of technology we have access to, to the types of ambition we're allowed to have.”
在 Anish 看来，这种“留在光锥之外（outside of the light cone）”的恐慌甚至普遍存在于从基础模型实验室的顶尖研究员到硅谷普通从业者的各个层面。然而现实中，技术非但没有把所有人推入深渊，反而空前地解绑了“技能”与“愿望（unbundles skill from desire）”，不仅放大了人类的主观能动性（agency），也让人们能够追求前所未有的宏大野心。
- 过去硅谷评估项目时，如果创始人的野心过大，投资机构往往不敢轻易下注；如今环境发生逆转，想法如果太小，反而很难引起投资机构的兴趣。
- 护城河往往是在构建与探索过程中“被发现”的，而非在一开始就能在纸面上设计出来的。
---
### `05:00` 慢起飞、非智力瓶颈与组织重塑
面对这种恐慌，Anish 从三个维度拆解了为什么 AI 行业并没有走向极端垄断或将绝大多数人甩下车：
1. **生态去中心化**：上一代移动互联网时代由网络效应统治，网络往往呈现“赢家通吃（winner take all）”的单极化走向；但在当下的 AI 技术栈中，从底层的闭源模型实验室、开源权重模型，到上层的代码智能体（如 Claude Code、Codex、Lovable、Replit、v0 等），几乎每个层级都有数十家实力强劲的玩家同时蓬勃发展，完全打破了垄断预期。
2. **实证就业数据相反**：社会经常宣称放射科医生在 20 年内就会被 AI 彻底取代，但如今放射科医生的职位需求比以往任何时候都高，软件工程师等技术岗位同样如此。
3. **不存在真正的自我递归改进（RSI）**：外界常常害怕 AI 会因为 Recursive Self-Improvement（递归自我改进）产生一个微小的微积分增量（$\epsilon$），从而瞬间引发滚雪球式的超级智能逃逸。但即便询问最前沿的实验室专家，目前发生的技术迭代更接近“自催化效应（autocatalytic effects）”——即利用模型工具加速并优化研发流程，但绝非真正意义上的指数级失控自我重写。
针对“快速起飞（fast takeoff）”还是“慢起飞（slow takeoff）”的争议，Lenny 提到虽然经常看到模型成功黑进 Hugging Face 等惊悚新闻，但整个业界仍能对其进行监测、反思、迭代与干预。Anish 赞同我们处于典型的慢起飞场景：
> “Like the line of reasoning for that case is always everything up until now, then something happens that no one can quite articulate and then fast take off. So I don't believe that that's going to happen.”
Anish 强调，即使技术模型进步神速，现实世界的“经济扩散（economic diffusion）”速度本身就会起到缓冲作用；更关键的是，现实中究竟有多少问题真正是“受限于智力（intelligence bound）”的？就算 Domino's Pizza 或 FedEx 拥有整整一数据中心的顶尖博士智力，也不可能立刻指数级改写其物理层面的供应链配送和披萨烘烤。
在企业内部的分化问题上，Anish 反对将普通员工矮化为《呆伯特》（Dilbert）式的官僚推纸机。许多企业中，基层员工对于拥抱工具展现出极高热情。例如墨西哥二手车交易平台 Kavak，其内部成立了“绝地武士学院（Jedi Academy）”，在为期 6 周的培训中教包括修车工人在内的所有员工使用前沿工具，最终学员甚至能够亲手交付尖端的智能体。
```mermaid
flowchart TD
accTitle: Technology Diffusion in Companies
accDescr: Comparison between shallow tool adoption and deep structural reorganization across technology paradigms.
subgraph Phase1["浅层替换（Shallow Adoption）"]
A["引入新技术（煤炭换电力 / 个人用 AI）"] --> B["岗位与旧流程保持不变"]
end
subgraph Phase2["深度重构（Structural Reorganization）"]
C["推倒工厂重建流水线（耗时数十年）"] --> D["围绕模型重新设计公司架构（闭环化）"]
end
Phase1 -.->|"需要漫长扩散与认知突破"| Phase2
```
Anish 借用第二次工业革命中“电力普及”的历史类比指出：人类从发明电力到彻底围绕电动机重新规划工厂布局（砸碎旧厂房、建立流水线，而不是仅仅把原有的蒸汽机、燃煤动力替换为电线），整整花费了 40 年时间。当前绝大多数企业仍处于“给现有岗位配备 AI 工具”的浅层阶段，只有最野心勃勃的企业才开始围绕底层模型彻底重组整个公司。针对职业恐慌，Anish 指出高管的目标始终是扩张而非单纯降本：
> “Sundar running Google, he doesn't want to run a more efficient $4 trillion company. He wants to build a $40 trillion company. So anytime you have an economically productive unit, it's rational to kind of, especially if it gets more productive, to maintain.”
### `10:00` 路线图大提速与持续交付的魅力
谈到当下技术在组织内部的渗透，Anish Acharya 分享了一个生动的行业轶事。许多人担心人工智能会引发大规模裁员，但现实情况却呈现出另一种截然不同的动态。Anish 提到自己与一位在 Google 担任高管的好友交流，当被问及“你们裁员了吗”时，对方坦言并未裁员，真正的变化在于产品研发速度的质变。这位高管表示，他们现在能够将原本需要两年的路线图在短短三个月内彻底推进完毕。曾经困扰产品经理与团队之间无数个日夜、耗尽心力的功能优先级争论不复存在，现在最棘手的问题反而成了“究竟该往路线图里塞入什么新东西”，这对任何产品经理而言都像是一场成真的幻想。
面对这一浪潮，Anish 强调业内从业者不应感到焦虑与重压，反而应感受到巨大的赋能，而适应这种转变的最佳方式就是单纯地不断交付（ship）东西。Anish 将前产品副总裁、知名创业者 Claire Vo 视作自己的灵感缪斯，盛赞她之所以令人钦佩，正是因为她始终在不断尝试、频繁交付产品，并且毫不畏惧可能带来的尴尬或不完美。看到她展现出有史以来最好的自我状态，Lenny 也半开玩笑地附和道：“等我长大了，我也想成为 Claire Vo。”
- “No, we didn't. What we instead do is now rip through our roadmap. So two years of roadmap happens in three months.” —— Anish Acharya 引述其 Google 高管朋友的话
---
### `11:15` 公司即闭环：从 Prompt、Agent 到业务级 Loops
随后，Anish 阐述了他对未来企业组织构架的核心论断：现代公司的运作正在演变成为一系列“闭环（Loops）”的集合。虽然在社交平台 X 上，这一概念常被调侃为行业大牛的过度包装，甚至有人拿“闭环的下一步是图谱（graphs）”来制作迷因，但 Anish 针对其底层逻辑给出了强有力的严密论证。技术的发展从最初的单次提示词（Prompts），演进到了具备工具、记忆和技能文件的 Agent（本质上是置身于循环中的模型），而当多个 Agent 协同处理复杂任务时，便形成了功能闭环（Loops）。
```mermaid
flowchart TD
accTitle: AI架构从Prompt到组织级闭环的演进
accDescr: 展示从Prompt到Agent再到工程与业务全链路闭环的层级推进
subgraph Core["基础技术演进"]
P["提示词 (Prompt)"] --> A["智能体 (Agent)<br/>模型 + 工具 + 记忆 + 技能"]
A --> L["任务闭环 (Task Loop)"]
end
subgraph EngLoop["工程闭环 (Coding Loop)"]
L --> B1["Bug 报告录入"]
B1 --> B2["复现环境与修复生成"]
B2 --> B3{"风险评估"}
B3 -->|"低风险"| B4["自动部署并邮件通知"]
B3 -->|"高风险"| B5["人工审查确认"]
end
subgraph BizLoop["企业级级联闭环"]
B4 --> BL["业务单元级闭环"]
B5 --> BL
BL --> C1["营销 / 销售 / 法务 Loops"]
C1 --> CEO["触发战略 / 商业模式调整"]
end
```
软件工程是闭环化最先成熟的领域，因为这里拥有最顶尖的模型以及技术素养最高的用户。在工程中，典型的闭环已经建立：Bug 报告录入、系统自动生成复现、代码修复方案自动生成并进入审查，低风险变更直接自动上线甚至自动发邮件告知客户，整个过程仅需五分钟。Anish 进一步提出，业务层面的闭环正在成形——如果一名总经理统管研发、营销、销售、客户支持和法务，各个职能闭环的综合输出本身也是一个可以优化的闭环，其极值形式甚至可以直接向 CEO 发出警报，建议调整现实业务模块、商业模式或顶层战略。
在这种由 AI 原生驱动的公司里，Anish 坚信人类不可或缺，完全自治在大多数组织工作里并不成立。人类未来在企业中将专注于四大核心领域：销售（Sales）、支持（Support）、战略（Strategy）和异常处理（Exceptions）。他直言，当前的 AI 模型在产生“分布外思考（out-of-distribution thinking）”方面依然极度受限，数学领域的某些推理突破并不能等同于在复杂商业世界中的创新思考，依然需要真实的人类来判定“我们究竟该打造什么”。
- “So I think we're going to see this sort of cascading set of everything from a loop per person, loop per job function, loop across entire business units to loops that can run large parts of the company.” —— Anish Acharya
- “When you think of what a human will do in this AI-native company, sales, support, strategy, and exceptions.” —— Anish Acharya
---
### `13:30` 增长团队的自动化与突破“局部最优”
为了具体说明闭环如何颠覆传统团队，Lenny 询问这种模式在市场拓展、增长或法务中会是何种形态。Anish 以 Lenny 曾在 Airbnb 负责过的供给增长（Supply Growth）团队为例展开剖析。在传统模式中，团队必须聚集全员进行漫长的头脑风暴，依靠大量电子表格梳理潜在实验清单、人工排定优先级、研发上线并观测指标。而在闭环模式下，Agent 可以自动生成每一个测试变体并自动度量指标，一旦统计学显著性（stat sig）达到设定的 p 值阈值，系统便自动收敛并全量发布该变体，同时保留长期对照组并即刻开启下一轮实验。
然而，闭环并不能解决一切增长问题。Anish 强调，算法闭环本质上是在执行爬山算法（Hill Climbing），它能够高效带领团队登顶一座小山峰，但这往往只是一个“局部最优解（local maxima）”，随后便会陷入平台期。此时，纯粹依赖模型的系统会停滞不前，必须由人类凭借直觉与分布外思考介入，把整个方向搬到“下一座山峰的山脚下”，再由 Agent 开启新一轮的爬山。这也是为什么用户无法单纯给 Claude 下指令说“帮我赚一百万美金，别犯错”就能成功，因为机器始终需要人类为其校准战略大方向。
- “The loop will help you climb to the local maxima, but then it plateaus. And you need some sort of out of distribution thinking. You need human intuition. You need somebody to actually help you land at the base of the next hill.” —— Anish Acharya
---
### `15:30` 职能角色的再定义：从 GTM 变革到产品经理的脱水测试
Lenny 观察到，这种由工具闭环驱动的趋势正在迅速蔓延。据 OpenAI 内部传出的消息，其市场拓展（GTM）团队使用编码模型 Codex 的频率甚至已经超越了工程团队本身。Anish 对此非常认同，并形象地调侃道，虽然这是一种刻板印象，但销售与 GTM 人员真正热爱的无非是“去健身房、吃牛排大餐、赢取季度最佳销售冠军奖杯”。AI 工具将他们从繁重的案头行政杂务中彻底解脱出来，让他们能够百分之百专注于人际互动与拿下交易这些自身最擅长、最享受的核心事务上。
```mermaid
flowchart LR
accTitle: 传统产品研发与模拟闭环产品研发对比
accDescr: 展现从人工反复拒绝与排期转变为全量构建与模拟验证的闭环路径
subgraph Legacy["传统产品流程"]
direction TB
A1["大量需求涌入"] --> B1["被迫对大多数说 No"]
B1 --> C1["政治博弈 / 排期内耗"]
C1 --> D1["极少数功能艰难上线"]
end
subgraph AI_Native["AI原生模拟流程"]
direction TB
A2["各类新点子涌入"] --> B2["对所有想法说 Yes"]
B2 --> C2["Agent 自动快速构建"]
C2 --> D2["Simile 等工具模拟用户反应"]
D2 --> E2["数据胜出者直接推向生产"]
end
```
这种机制也对传统产品管理（PM）模式发起了巨大挑战。Lenny 分享了一位 PM 朋友的颠覆性尝试：过去 PM 的日常是不得不对扑面而来的海量想法说“不”，陷入“先放进路线图稍后再说”的拉扯中；现在这位朋友反其道而行之，对所有想法统统说“好”，利用 AI 将所有功能都构建出来，再通过如 Simile 这类能够模拟真实用户反应的仿真产品进行测试，以此判断功能是否值得真正落地。
针对这种模式，Anish 指出了一个残酷但极具价值的组织真相：过去很多公司的 PM 都自命为真正的“从 0 到 1 创新思考者”，并总将无法施展才华归咎于管理层的压制、高管 founders 的干涉或是研发资源的紧缺。然而，当构建门槛被铲平、每一个产品故事和功能都能被轻松实现并检验时，许多 PM 终将看清自己其实并不擅长 0 到 1 的从无到有，而在这个新时代里，“为别人的好点子添砖加瓦，远比坚持推行自己的烂点子要充实得多”。最终获胜的方案也不再依赖于谁能把故事讲得最好以取悦高层，而是所有点子都能通过实际闭环运行，由最好的结果胜出。
- “I think a lot of PMs are going to realize they're actually not that good at zero to one. And it's much more fulfilling to work on someone else's good idea than your own bad idea.” —— Anish Acharya
---
### `18:00` 软技能的护城河与 Agent 的人机教学范式
对话进一步延展至技能价值的重构。过去科技行业普遍存在一种偏见，认为“软技能（soft skills）”最廉价，而工程与技术技能因其具象、严密而最具价值。如今事实证明，由于代码和工程输出具备极强的可验证性与明确的成功标准，AI 恰恰最擅长替代这些技能；相反，那些难以简单闭环化、输出极难形式化验证的软技能，反而构成了系统的终极限速瓶颈（正如“一个人一晚只能吃一顿牛排晚餐”）。
关于如何将具体业务职能转化为 Agent 闭环，Anish 引用了拉美独角兽 Kavak（二手车线上交易平台）联合创始人 Ale 的前沿实践。Kavak 为每个客户都配备了一个专属 Agent，当 Agent 在业务流中受阻或遇到无法处理的复杂状况时，系统并不会直接崩溃或报错，而是会主动呼叫人类导师。人类介入对 Agent 进行实时指导并协助破局，Agent 则在此过程中完整捕捉所有上下文与追踪链路（traces）完成自我学习。
Anish 指出，只要模型犯错或做出了偏离预期的决策，本质上都是因为人类掌握了模型未曾拥有的上下文——这要么是知识缺口（knowledge gap），要么是数据缺口（data gap）。人类的核心任务是填补这一缺口，使得模型下次不再需要呼叫人类。在未来的公司中，各职能部门都应被视为自运转的 Agent 闭环，常规的跨职能执行（法务、营销、销售协同）都会自动完成，而人类的终极使命将彻底解放：“去远足、做白日梦，并找出下一个需要攀登的山头。”
- “When the agent gets stuck, it actually calls a human. And the human will coach the agent through. Now, the magic of that is not only does it unblock the agent, but of course, the agent then captures all the traces and learns from it.” —— Anish Acharya 引述 Kavak 联合创始人 Ale 的实践
- “And your idea, you know, your job is to go take a hike and dream the dreams and come up with the next hill to climb.” —— Anish Acharya
### `20:00` 智力普及后的行业竞争与“帕累托效率”分化
Lenny Rachitsky 询问，当人类当前所做的大量工作都被 AI 接管时，还有什么会成为企业的差异化竞争优势？Anish Acharya 提出了一个常被行业忽视的观点：**大多数行业的竞争均衡格局并不会被颠覆**。以披萨连锁行业为例，无论是 Pizza Hut、Domino’s、Papa John’s 还是 Round Table Pizza，当它们都能以极低成本调用相当于一个“由博士组成的数据中心”的算力时，各家要么主动拥抱技术，要么因落后而更换 CEO 再去拥抱。虽然短期内会经历阵痛与市场排位的微调，但绝不会出现单家公司占据 99% 市场份额的局面，因为披萨等众多传统实体产业“并不是受智力瓶颈约束的（not intelligence bound）”。
关于组织架构与 AI 模型选择的演进，Anish 引入了经济学中的**帕累托效率（Pareto efficiency）**概念——即在性价比曲线上为单位性能支付合理的成本。然而，最先进的前沿模型（Frontier models）在定价上往往是“不理性的”：为了获得概念上多出的 1 点 IQ 智商，用户可能需要支付相比普通模型高出 100 倍的溢价。但在某些拥有“无限上升空间（unbounded upside）”的领域（例如药物研发），哪怕只有 1 点智商的提升帮助发现了下一个降脂药（statin），其回报就是一个价值万亿美元的成果，因此支付昂贵的前沿模型算力是完全理性的。
```mermaid
flowchart TD
accTitle: 模型层级与业务场景的分化矩阵
accDescr: 展示了无限上限高杠杆岗位与有限上限标准化岗位的模型适配逻辑
subgraph HighUpside["无限上限 / 杠杆驱动场景"]
A["业务职能：销售、产品研发、核心工程、新药发现"]
B["模型架构：昂贵的前沿大模型（Frontier Models）"]
A --- B
end
subgraph BoundedUpside["有限上限 / 标准化流程场景"]
C["业务职能：常规法务、财务核算、行政闭环"]
D["模型架构：开源权重模型 + RL 专项微调（中等 IQ、低成本）"]
C --- D
end
```
这种经济规律将导致企业内部和模型生态产生鲜明的分化：
- **上限受限的职能（Bounded upside）**：如常规法务与财务做账。无论多聪明，账目只能做到“正确”，不可能凭空多做 100 倍好；此类职能将大量采用具备高性价比、经过强化学习（RL）微调的开源权重模型（Open-weight models）。
- **上限无限的职能（Unbounded upside）**：如研发、销售、核心工程；即使前沿模型的 Token 极其昂贵，只要能带来突破性回报，企业也会坚决采购。
- Anish 总结道：“I think the most useful question to ask yourself as a founder CEO is just, hey, if we assume these things are infinitely intelligent and astonishingly cheap, how would we reorganize the company?”
---
### `25:00` 边缘线索的悖论与“模型品酒师”的实战心法
探讨深入后，Lenny 询问这种分化是否与工作成果的“可验证性（verifiability）”有关。Anish 澄清，新药研发和财务做账都是可验证的问题，核心区别依然在于**潜在上限有多高以及计算其价值的难度**。他举了一个微妙的例子：在客户支持环节，普通用户打来电话报告一个 Bug，中等智力的通才模型可能只当成普通工单处理；但如果是一个顶级智慧的大脑或创始人亲自接听，可能会意识到这个 Bug 是撼动整个产品形态的“第一颗面包屑（breadcrumb）”。这又构成了“所有问题都该默认采用最顶尖前沿智力”的反向论点。
谈及市面上形形色色的模型，业界戏称 Anish 为“模型品酒师（Model Sommelier）”。Anish 直言，品酒师的秘诀是喝过 10,000 瓶酒，而成为模型品酒师的秘诀就是**每出一个新模型，就逼自己用它交付（ship）一个具体作品**：
- **Qwen 3.8 Max**：极其擅长超长任务处理与叙事创作，性格偏向“富有创造力与开放”。Anish 用它自主导演时长 5 分钟的《星球大战》行星纪录片（如 Tatooine 和 Bespin）。模型能自主连续工作四五个小时，通过 FAL 调用 Minimax 模型生成画面、调用 ElevenLabs 生成音频，甚至充当导演进行全片剪辑与转场审查。
- **GLM-5.3**：不具备视觉组件，但表现得像一个“被关在角落里的神经质博士（neurotic PhD）”，在严谨度和代码精确度上极高，非常适合产品工程。
- 驳斥同质化论调：“I think for people who believe the models are commodities or totally fungible, you just haven't actually used the models.”（我认为那些觉得所有模型都是同质化大宗商品、可以完全相互替代的人，根本就没有真正上手用过这些模型。）
关于普通人如何建立模型直觉，Anish 建议不要凭空构思宏大议题，而是当回那个曾经令人讨厌的“总爱聊 App 点子的朋友（App idea guy）”。打造一个专属的实验底盘（chassis），哪怕是最荒谬的傻点子，持续把新模型当作工具迭代塞进去。Lenny 随即分享了自己的心得：当前最重要的习惯，是在任何行动前按下暂停键，像冥想觉察一样，在输入与反应之间插入一个本能的思考——**“AI 能帮我完成这件事吗？”**。Anish 对此深表赞同，提到自己除了在日常软件中增加新功能外，也曾将笔记本电脑放在厨房，全程听录并分析家庭空间中的日常互动。
### `30:00` 衡量笑声与“让我更快乐”的闭环
Anish Acharya 笑着讲述了他用 AI 管制儿子 iPad 屏幕时间的一个小实验：如果儿子表现良好、说了积极正向的话，屏幕时间就会增加；如果表现不好则会扣减。结果这个社会实验迎来了戏剧性的转折——他的儿子录下了一段自己不断说“爸爸，我爱你”的视频，并将播放设备直接放在麦克风旁，以此成功“黑”掉了这套度量指标。这印证了古德哈特定律（Goodhart's Law）：一旦某项指标变成目标，它就不再是一个好指标。
Lenny Rachitsky 随之分享了一个类似的故事：他有一位朋友改装了一个 Limitless 挂坠设备，专门用来记录他和孩子一天中笑出声的频率，父子俩每天都会一起查看这个数据。Anish 对此深受触动，指出播客此前花了大篇幅讨论生产力、失业等带有大写字母“I”的沉重严肃议题（Important topics），但像“衡量一天笑了几次”这样的点子，虽然很难写成一份商业企划书或产生重大经济效益，却关乎人类的生活品质。人们总以为幸福感是恒定不变的，但回看历史，一百年前人们所做的工作远没有今天酷，一百年后或许更酷。技术的真正价值在于为人性生活增添质感与纹理。
```mermaid
flowchart LR
accTitle: 从生产力闭环到情感生活闭环的演进
accDescr: 对比传统企业级 AI 闭环与消费级情感生活闭环的目标与核心机制。
subgraph Enterprise ["企业级生产力闭环"]
A["业务增长 / 销售获客"] --> B["自动化工单 / 流程提效"]
B --> C["追求节省时间与资本回报"]
end
subgraph Consumer ["消费级生活闭环 (Loop: Make Me Happier)"]
D["增进身心健康 / 成为更好的朋友"] --> E["促成高质量陪伴与情绪共鸣"]
E --> F["追求消磨美好时光与精神充实"]
end
```
谈及此，Lenny 提到了 Anish 之前提出的一个极具前瞻性的论点——“闭环：让我更快乐”（Loop: make me happier）。Anish 对此展开了深刻阐述：
> “我认为我们总以为人们想要更高的生产力，但实际上他们并不想。我认为比起节省时间，更多的人其实是想消磨时间。”（“I think that we believe that people want to be more productive, but they don't. I think more people want to spend time than save time.” —— Anish Acharya）
- 全球体量最大的消费级产品几乎全集中在娱乐与社交领域，因为绝大多数人的本质需求不是“做完工作”，而是“享受时间”。
- X（Twitter）上的 AI 用户极度焦虑，生怕自己跌入所谓的“永久底层阶级”，整天就 GLM-5.3 还是 Kimi K3 展开激烈论战，思想已被技术深度“灌脑”（pilled）。
- Instagram 上的普通大众看待 AI 则完全不同，他们只觉得这就像是一个更好用的 Google 搜索，完全无法理解硅谷圈子里的歇斯底里与炒作。
---
### `32:30` 扩展灵魂的技术与初创公司的独特优势
Anish 尖锐地指出，大众与前沿技术的脱节，本质上是产品设计的失败，而非模型能力不足。过去四十年间，科技界倾尽全力研发的技术，最终只是做出了“更好的电子表格”；我们造出了极度强大的脑力外延工具，却从未打造出任何能够扩展人类灵魂（extend our soul）的东西。随着传统文化纽带的消解，人们普遍承受着精神饥渴；特别是在美国广大非沿海地区，人们并没有随处可见的热瑜伽、普拉提、轻断食或“朋友节感恩晚宴”（Friendsgiving），AI 恰恰应当填补这一精神真空，服务于人类最根本的情感渴望：如何感到被爱、如何建立连接、如何取得进步、如何获得快乐。
Lenny 顺着这个逻辑推导：既然企业界有“帮我扩张业务”、“帮我多拿销售线索”、“帮我解决客服工单”的闭环，消费级 AI 完全可以建立“改善我的健康”或“让我成为更好的朋友”这样的生活闭环，让模型日夜思考如何帮用户维系人际关系。
Anish 强调，AI 在这类场景中甚至可以具备冒犯性、挑战性和反面意见，而这正是初创公司相对科技巨头的巨大生态位优势：
> “在 Google 内部，有上千个合规委员会光是听到要发布一个会反驳用户、或者哪怕带有一点性暗示的模型，就会气得在坟墓里打滚；但你猜怎么着？那些全都是人类真实存在的一部分。”（“There's a thousand Google committees who would roll in their graves at the idea that they're going to release a model that's disagreeable or God forbid it should be like sexually suggestive. But guess what? Those are all parts of human existence.” —— Anish Acharya）
- 初创公司具备探索人性与社交生活中那些“不适、边缘但真实”领域的敏捷度与自由度，而巨头的公关和合规机制彻底锁死了自身在这方面的可能。
- 消费级 AI 的蓝海正是围绕家庭、朋友人际连接所构建的“幸福闭环”应用。
---
### `34:30` 消费级 AI 的“iPhone 2010”时刻与三大瓶颈
针对消费级 AI 现状，Anish 认为当前的行业阶段相当于智能手机的“iPhone 2010”时期——在这一阶段，Uber、WhatsApp、Airbnb 等定义移动互联生态的划时代产品都还未诞生，一切仍处于极早期。此前阻碍消费级 AI 爆发的主要有三大阻碍：
1. **模型调用成本居高不下**：这使得构建面向大众的免费模式（Freemium）产品在经济上难以成立；但随着开源权重模型（Open-weight models）的大幅降价，该瓶颈正在迅速化解。
2. **交互界面（UI）严重滞后**：对话框式（Chat）只适合极具主动性（agency）的极客精英（如 Elon Musk 或 Sam Altman），而普通大众最理想的信息交互范式其实是 TikTok，产品设计必须在 Chat 和 TikTok 之间找到平衡点。
3. **技术资源过度向生产力倾斜**：忽视了人类情感连接与娱乐的本能需求。
在这三大瓶颈逐步松动的当下，创新者们已在界面层发起进攻：
- 知名 AI 伴侣产品 Replika 的创始人 Eugenia Kuyda 正在对新一代用户交互展开极具野心的探索。
- Airbnb 联合创始人 Brian Chesky 也成立了专门的基础实验室，探索下一代全新人机交互界面。
---
### `36:30` 摆脱 2% 增长泥潭：解绑技能与欲望的宏观乐观论
面对当下社会对 AI 取代人工、技术灾难的普遍恐慌，Lenny 提到了 Marc Andreessen 此前的观点——在全球人口出生率骤降、生产力放缓、地缘冲突加剧的背景下，AI 的到来恰恰是拯救人类免于陷入长期泥潭的及时雨。Anish 从更宏观的人文与经济视角给出了强烈的乐观回应：
> “当面临的赌注极高时，人类的表现无与伦比；当赌注极低时，我们往往处于最糟糕的状态。”（“When the stakes are high, we are awesome. When the stakes are low, we are at our absolute worst.” —— Anish Acharya）
Anish 认为，五年前的世界处于一种无聊且低赌注的状态，导致社会集体沉迷于许多低效、内耗且无法让人感到幸福的琐碎副业。而在二战后的 1950 和 1960 年代，高对抗、高目标的社会动员让人类坚信自己无所不能。AI 正在重新点燃人类雄心的阶梯，帮助社会打破长期徘徊在 2% 左右的微弱 GDP 增长，向 10%、15% 甚至 20% 的跨越式增长迈进。
```mermaid
flowchart TD
accTitle: AI 解绑技能与欲望的关系机制
accDescr: 展示 AI 如何消除传统技能门槛，让个体的纯粹渴望直接转化为创造产出。
subgraph Traditional ["传统模式：渴望受限于高门槛技能"]
Desire1["创作欲望 (想做音乐 / 写代码 / 建桥)"] --> Barrier["数年严格专业技能训练 (钢琴演奏 / 语法编译 / 结构设计)"]
Barrier --> Output1["落地产出"]
end
subgraph AI_Era ["AI 赋能时代：技能与渴望彻底解绑"]
Desire2["个体表达欲望与独特意图"] --> AI_Tool["生成式 AI / 智能体网络 (处理底层执行实现)"]
AI_Tool --> Output2["软件、艺术作品、物理工程创新"]
end
```
这种底层赋能最核心的机制，在于 AI 将“个人欲望”与“专业技能”彻底解绑：
- **艺术与工程的平民化**：想创作音乐不再需要花费数年苦练钢琴技巧；想开发软件不必精通底层编程语言；即使想建造一座桥梁，也无需持有土木建筑学家的全部证书。
- **放大个体主体性**：工业革命带来的规模效应曾压抑了个人独特性与身份认同，而 AI 作为一种“情感与精神接口”，反向放大了每一个人的主体意识。
- **显性偏好（Revealed Preferences）与表述偏好（Stated Preferences）的割裂**：舆论层面充斥着对数据中心占地、中层管理者失业等宏观抽象概念的焦虑与控诉（表述偏好）；但在个人实际体验中，普通人的生活却切实因此变得更高效、更充实，且每天都在频繁使用 ChatGPT（显性偏好）。
- **现实的突破**：Anthropic 创始人 Dario Amodei 撰文严肃推演“攻克人类所有疾病后的世界”，Moderna 等生物科技公司正借助 AI 逼近治愈癌症的奇点。当科技已经逼近彻底根除绝症的门槛，公众视线却仍深陷在所谓“永久底层阶级”的虚妄恐慌中，这本身就是技术公关层面的巨大错位。
### `40:00` 破除 AI 恐慌：让关键事物变得廉价与打破名校光环
关于公众对 AI 的恐惧与恐慌情绪，Anish Acharya 认为扭转舆论叙事最核心的一点在于利用技术将真正重要的事物变廉价。“The most important thing that we can do with AI to change the kind of conversation around it is make important things cheap.” Anish Acharya 强调。在美国，医疗与教育是两项极端重要却日益昂贵的服务。以医疗为例，其中有 45% 的成本消耗在行政流程上；若借助 AI 剥离庞大的行政负担，将推动医疗服务进入实质性的通缩阶段，更不用说攻克各类疾病带来的巨大红利。教育领域则迎来了两百年来最强劲的颠覆性竞争。AI 有望将纯粹的学习体验与传统体制解绑，进而把个人声望从名牌大学文凭中剥离出来，人们不再必须依赖哈佛学位来证明自身价值。
随后 Lenny Rachitsky 提到，近期 OpenAI 似乎有意放缓其最新模型的 RL（强化学习）阶段，引发了外界对前沿模型演进过快失控的担忧。Anish 对“模型过于危险以至于不能发布”这种说辞持审慎怀疑态度。他指出，Anthropic 当年因声称某个模型过于危险而赢得的神秘光环与关注度极其可观；但企业放慢发布步伐，背后往往混杂着更现实的商业与资源考量——比如遭遇 GPU 算力短缺、出于商业竞争考量希望将专有模型留在内部以拉大护城河、或者纯粹是营销叙事与推理成本之间的权衡。
尽管网络攻击等安全防御问题必须严肃对待（必须在系统变得极易被攻破之前进行加固），但整个 AI 竞争格局并没有演变成一家公司垄断超级智能的局面。虽然 Anthropic 曾一度被认为不可撼动，但在过去六个月里 OpenAI 表现抢眼，开源模型（Open Weights）也在迅猛追赶，行业格局始终保持动态竞争。
- 医疗支出中有 45% 属于纯行政性负担，AI 在该环节的自动化能带来明显的通缩效果。
- 传统体制面临两百年来最大的冲击，AI 正加速让教育去中心化，实现能力与文凭的解耦。
- “模型过于危险而无法发布”的公关话术，往往受到营销策略、算力瓶颈与商业护城河等多重现实因素的干扰。
---
### `42:30` 个人 Agent 工具崛起：GrokBot、ChatGPT Work 与 Instinct 的产品对决
在日常个人助理和 Agent 工具层面，两位嘉宾交流了目前处于第一梯队的前沿产品。Lenny 表示自己完全被突然冒出来的 GrokBot 迷住了，认为它是体验极佳的 AI 助理。Anish 也高度认同，并列举了目前最值得关注的三大工具：GrokBot、ChatGPT Work（内置在 ChatGPT 界面中但极为强悍）以及初创团队打造的 Instinct。
在对比这些 Agent 工具的技术与产品形态时，Anish 解析了 ChatGPT Work 的独特优势。尽管 GrokBot 和 Instinct 在系统接管和凭证获取上表现得更加激进，但 ChatGPT Work 具备两个关键壁垒：首先是运行在云端，并稳健地做好了浏览器凭证缓存；更关键的是它集成了全双工语音模式（Full-Duplex Voice Mode）。用户可以直接给它“打电话”，通过流畅的语音即时感知并调度所有后台正在并发运行的代码 Agent 线程。这种随时语音唤醒并通晓全局上下文的体验，让用户感觉自己真正拥有了一个无所不知的私人助理，而不再局限于传统的单向语音听写或文字输入界面。
```mermaid
flowchart TD
accTitle: 个人 AI Agent 产品形态与架构对比
accDescr: 对比 ChatGPT Work、GrokBot 和 Instinct 在执行环境、控制方式与凭证管理上的差异
subgraph CloudDuplex["ChatGPT Work (云端全双工)"]
A["云端容器执行"] --> B["全局线程上下文监控"]
B --> C["全双工实时语音呼叫交互"]
A --> D["浏览器凭证缓存 (温和安全)"]
end
subgraph DirectAgent["GrokBot / Instinct (激进接管)"]
E["激进的系统 / 凭证接管"] --> F["单向听写或文本指令"]
F --> G["本地 / 混合环境自动化执行"]
end
```
---
### `44:50` 人类欲望与雄心：为什么 AI 不会导致灾难性失业？
谈及 AI 对宏观经济和就业的冲击，Lenny 指出行业内存在两种截然不同的预判：一端是像 Anthropic 联合创始人 Dario Amodei 那样，担忧未来有 50% 的知识工作被颠覆并引爆大规模失业；另一端则是像 David Sacks 那样，坚信一切都会变好、就业完全不用担心。显然 Anish 的立场更偏向乐观阵营。
Anish 从人类文明的发展史切入分析，指出人类的欲望扩张速度永远快于其满足欲望的能力。“I mean, I think that the entire trend of human existence has been that our desires grow faster than our ability to fulfill them.” Anish Acharya 论证道。在当今社会被视为理所当然的日常事物——无论是心理咨询还是抗生素——在几十年前或几百年前连最有权势的帝王都无法企及。他幽默地预测：“二十年后，人们会因为自己在火星上没有度假别墅而感到愤怒，是真正意义上的愤怒。他们在 Instagram 上刷到别人在火星度假，就会对伴侣说：‘亲爱的，我们得更努力工作把它买下来。’”每一位 CEO 都会渴望建立规模大得多的企业。人类对美好生活的定义正在以远超过去的尺度扩张，根本不存在需求饱和导致无事可做的前提。
Lenny 补充道，AI 不仅让人们能够拥有更大的野心，甚至迫使所有人必须变得更有野心。当所有简单基础的任务（如制作一个简单的白底静态网站）被 AI 彻底降维成一键生成的低成本操作时，唯一的区分度就变成了你能把事情做到多宏大（例如做一个沉浸式的 3D 探索游戏）。对此 Anish 进一步扩展了“雄心”的内涵：野心不应被狭隘地定义为商业上的创立软件公司或当创始人，它包括更广泛的创造力野心（像五岁孩子般不受评价约束地去画画）、体制改善野心（比如把英国饱受诟病但被珍视的 NHS 医疗体系改造成像 iPhone 一样流畅），或者是想成为一个更有陪伴质量的父母。
- 稀缺性是相对的：过去百年的抗生素与当代的心理治疗，都曾是无法企及的奢侈品，未来对新事物的追求将不断重构就业空间。
- AI 打破了软件和智力的稀缺性壁垒，迫使产品从业者从受限于 MVP 的传统思维，转向构想 10 倍乃至 1000 倍的宏大叙事。
---
### `48:30` 通过构建去学习：“活动本身”即是价值，而非结果
面对 AI 时代的开发范式，Lenny 提到 Claude Code 团队内部奉行的一条准则：“你知道有什么比我自己做这事更好吗？是让 Claude 来替我做。”这种思维倒逼每位成员养成“我该如何辅助 Agent 帮我完成目标”的全新工程习惯。
与此同时，Anish 与 Lenny 深入探讨了“通过构建来学习”（Building to Learn）的重要性。Anish 注意到当前一种有趣的社会心理：社交网络上讨论“Vibe Coding”（氛围编码）的人极多，但真正愿意公开展示自己项目的却寥寥无几，因为很多人会因为自己的副项目看起来微不足道、不够重要而感到难为情。
“And it's totally okay for most of it to go thrown away because that's still building your muscle.” Anish Acharya 认为，很多项目即便最终被丢弃也无所谓，因为动手的核心意义在于锻炼心智肌肉与探索边界。Lenny 精辟地总结道，当今时代正在将“构建”变成一种自我充实的活动过程，而非纯粹追求结果（Building as an activity rather than an outcome）。正如业余做了一套 DJ 混音作品，即便全网只有三个人收听、自己循环播放了一百遍，创造者也能从中获得巨大的满足感。带着这种对个体创造力释放的探讨，两人随后将话题转向了 Anish 在 a16z 主导的消费级 AI 投资版图。
### `50:00` 消费级 AI 的三大支柱：通用编程代理、个人助理与情感娱乐
尽管各大 AI 实验室在底层模型研发上卓有成效，但市面上真正立足、面向大众的独立消费级产品仍然屈指可数。在 Anish Acharya 看来，当前最值得关注的消费级 AI 机会主要分布在三大赛道：
1. **编程代理（Coding Agents）**：这类产品不应被狭隘地理解为“仅供程序员写代码的工具”。Anish 强调，编程代理本质上正在变成一种“与世界进行交互并解决通用问题的万能接口”。普通用户在社交网络上用 Claude Code 剪辑视频，或用 Codex 在飞机上几分钟内随手为孩子捏出一款小游戏。在这个方向上，a16z 投资了 Wabi，它作为一个微型应用（Mini-apps）生态平台，让任何普通人都能自由创建、体验和分发自己的轻量应用。
2. **个人代理（Personal Agents）**：此前像 OpenClaw、Hermes、Maltbook 展现了令人兴奋的原型，但它们本质上仍然是偏向开发者的工具。如今的演进趋势，是将这些复杂的技术提炼成普通大众与企业员工一眼就能看懂、真正好用且稳定的助手，类似于 Instinct、GrokBot 以及具备高级工作流能力的 ChatGPT。
3. **娱乐与情感陪伴（Entertainment & Companionship）**：包括像 Suno 这样出色的创意音乐工具，以及增长极其惊人的 AI 虚拟伴侣产品。虽然情感伴侣赛道在创投圈常常让人感到“难以启齿”，但其商业数据极其庞大。Lenny 提到在 a16z 发布的消费级 AI 市场图谱中，伴侣类应用独占鳌头；Anish 纠正了一个普遍的误解：“实际上，使用伴侣产品的绝大多数用户是四五十岁的女性，寻找虚拟男友的比例远高于女友。”
Lenny 总结道，这些方向与前面探讨的“循环（Loops）”紧密相扣，本质上是在解决人类永恒的“待办任务（Jobs to be done）”——让人更快乐、更长寿、更有寄托。Anish 赞同地指出，这正是为普通人“提供野心的出口、成就感的归宿，以及在闲暇时打发时间的精神陪伴”。
```mermaid
flowchart LR
accTitle: 消费级 AI 三大赛道及其核心驱动力
accDescr: 展示消费级 AI 在编程代理、个人助理和娱乐陪伴三个方向上的用户需求演进与典型代表。
subgraph Coding["编程代理 Coding Agents"]
C1["不仅仅是写代码"] --> C2["通用问题解决接口<br/>(剪辑/微应用)"]
C2 --> C3["代表: Wabi / Claude Code"]
end
subgraph Assistant["个人代理 Personal Agents"]
A1["开发者向原型<br/>(OpenClaw)"] --> A2["开箱即用大众助手"]
A2 --> A3["代表: ChatGPT / GrokBot"]
end
subgraph Emotional["娱乐与情感 Entertainment"]
E1["高阶创作工具<br/>(Suno)"]
E2["AI 伴侣产品<br/>(40-50岁女性为主)"]
E1 & E2 --> E3["满足精神寄托与闲暇时光"]
end
```
---
### `52:45` 护城河不是设计出来的，而是探索出来的
在 AI 工具泛滥、发布速度飙升千倍的今天，所有创始人和投资人都在焦虑同一个问题：“你怎样才能不变成一个随时被取代的套壳（Wrapper）？你的持久护城河究竟是什么？”
Anish 引用了 Decagon 联合创始人 Jesse 的经典论断作为回应：
> “护城河大多是被探索出来的，而不是被设计出来的（Moats are most often discovered, not designed）。”
创始人很容易在商业计划书上陷入“向 MBA 和 VC 证明自己拥有天衣无缝护城河”的精神内耗，但现实往往截然相反：团队只需全力以赴快速交付，护城河会在飞轮运转中自然显现。以 Cursor 为例：早期它被外界广泛嘲讽为一个没有技术壁垒的 VS Code 壳子，但它凭借极高 NPS（净推荐值）和强黏性的日活，牢牢占据了开发者心智。随后，团队通过捕获用户海量的推理轨迹数据（Reasoning Traces），自主训练了 Composer 系列专有模型，从而构筑起坚不可摧的壁垒。
Anish 强调，行业似乎遗忘了经典商业理论：“没有任何经典护城河是建立在‘软件有多难写’之上的。”绝大多数团队做的都不是研发自动驾驶汽车这种硬核极限工程，因此经典的护城河——网络效应、规模优势、品牌号召力、专有数据与垄断性资源——依然完全奏效。他特别推荐了 Hamilton Helmer 的经典著作《7 Powers》（七大战略维度），并指出真正经得起时间考验的商业书籍屈指可数：
- **《7 Powers》**（Hamilton Helmer 著）：剖析持久商业竞争壁垒的圣经。
- **《格鲁夫给经理人的第一课》（High Output Management）**（Andy Grove 著）：管理学永恒的经典指南。
- **《创业维艰》（The Hard Thing About Hard Things）**（Ben Horowitz 著）：第一本坦诚暴露创业焦虑、孤独与失败感的纪实之作。
---
### `55:00` 隐秘细节、产品工艺（Craft）与分发新壁垒
当被问及创始人若在融资路演中承认“护城河还在探索中”是否可行时，Anish 明确表示，a16z 乐于下注那些暂时没有完美护城河叙事、但展现出极高势头、极佳产品工艺（Craft）和用户参与度爆发的产品：
> “大想法背后总有十几个看不见的小细节在支撑；即使有人抄袭了你的大想法，他们也永远看不透那些让大想法真正运转起来的微小巧思。”
以 Granola 为例，该产品在两年前刚推出时备受行业质疑与唱衰，也没有惊天动地的技术护城河，但因其极致的产品质感与做工，在 Ramp 最新的高增长企业报告中位列前茅，深受用户狂热喜爱。
```mermaid
flowchart TD
accTitle: 核心体验与护城河形成路径
accDescr: 展示产品如何从前端体验与工艺积累起势，进而探索演进为长期护城河。
A["底层模型同质化<br/>(各大前沿模型性能接近)"] --> B["差异化落脚点:<br/>产品工艺 (Craft) 与极致 UX"]
B --> C["高 NPS / 强用户黏性"]
C --> D["捕获独特交互行为与数据流"]
D --> E["演进出长期护城河<br/>(网络效应/自研模型/专有资源)"]
```
随后，Lenny 与 Anish 探讨了工具层工具同质化的问题：各家的底层大模型差距不大，胜负全在产品脚手架（Harness）和交互体验（UX）上。现在极客用户往往同时订阅好几个昂贵套餐（例如 Cursor 的 300 美元档、Grok 的超级会员、ChatGPT Pro 等），谁能成为日常入口，取决于极致的交互体验。这也引申出了“分发能力即护城河”的命题：
- **经典寄生分发失效**：正如 Chris Dixon 在《Come for the tool, stay for the network》中所论述的，老一代平台（X、YouTube、Instagram）早被训练得极其警惕，严防死守任何第三方在自己平台上再造网络。
- **回归纯粹口碑（Word of Mouth）**：现阶段的分发无法再依赖移动互联网早期的应用商店红利或 Growth Hacking 奇技淫巧，而必须依赖 Seth Godin 所言的“造就非凡（Remarkable）之物”，让用户发自内心地自发在社媒提及。
- **摆脱增长幻觉**：面对增长焦虑，Anish 给出了一个极具启发性的思维模型：
> “我认为当下的团队面临的都不是增长问题，而是产品问题（Nobody has a growth problem these days. They have a product problem）。”
创业者应当挑战自我匮乏的想象力：如果要做一款软件界的“爱马仕铂金包”，敢于向用户收取每月 1,000 美元乃至 10,000 美元的高昂费用，那么这个产品无论在功能实用性还是情感共鸣上，到底必须做到多么令人惊叹的极致体验，才能配得上这一价值？想清楚这一点，再去反推如何构建它。
### `01:00:00` 初创企业的分发劣势是伪命题：现在的环境像 2009 年的圣诞节
Lenny 提出了一个许多创业者心中的普遍疑虑：如今每天都有数以千计的 AI 产品上线，初创公司在触达用户上是否天然处于劣势？巨头拥有现成的庞大分发网络，比如 Google 只要在搜索框里顺手推一把就能把 Gemini 推给亿万用户，这是否意味着巨头胜券在握？Anish Acharya 却给出了完全相反的判断，他甚至认为初创公司现在的日子反而更轻松了。
```mermaid
flowchart TD
accTitle: 巨头与初创公司在当前 AI 浪潮中的竞争态势
accDescr: 对比巨头拥有分发渠道但受限，与初创公司凭借产品大胆创新和高定价意愿的差异
subgraph Incumbents["现有科技巨头 (Incumbents)"]
A1["拥有巨大分发渠道与强推能力<br/>(例如 Google 强推 Gemini)"]
A2["顾虑声誉与风险<br/>不敢涉足争议/前沿领域"]
A1 -.->|"强推并未带来绝对胜利"| A3["产品体验缺乏压倒性优势"]
end
subgraph Startups["初创公司 (Startups)"]
B1["探索非传统与激进方向<br/>(如 AI 陪伴、情感与社交产品)"]
B2["产品创新驱动自然传播<br/>(非分发问题，而是产品问题)"]
B3["享受超高付费意愿<br/>(C端200美元/月，B端百万美元ACV)"]
B1 --> B2 --> B3
end
Market["2009 年圣诞节效应：<br/>用户对新科技充满渴望，付费大门全面敞开"] --> Startups
```
Anish 强调，尽管 Google 进行了极其猛烈的交叉销售与渠道捆绑，但“没有人会觉得他们赢了”。初创公司的机会在于敢于涉足让巨头畏手畏脚、感到不适的方向，例如 AI 陪伴（AI Companion）以及其他前沿的情感与社交领域。更关键的是，如今用户的付费意愿高得惊人：C 端用户愿意每月支付 200 美元，企业客户甚至愿意在还没完全搞清楚能得到什么的情况下就签下百万美元年经常性收入（ACV）的合同。Anish 将眼前的窗口期比作移动互联网黎明期：“这感觉就像 2009 年的圣诞节，人人都刚拿到属于自己的 iPhone，迫不及待地想要下载各种新应用。”虽然这种新鲜感未来某天会退潮，但至少在眼下，机会之门对初创企业完全敞开。
Lenny 非常认同这一点，并指出初创公司面临的从来不是分发或增长瓶颈，本质上依然是产品力的问题——如果产品足够惊艳，用户自然会疯狂分享并引发口碑效应。为了说明产品的大胆创新如何击穿社交传播，Anish 提到了创意工作室 MSCHF 几年前推出的病毒式金融实验项目“卡片大乱斗”（Card Versus Card）：
- MSCHF 给大约 10 万名用户邮寄了一张实体借记卡。
- 官方每天通过短信向所有人公布一个指定的消费地点，并在卡池里充入 100 美元。
- 所有人瞬间蜂拥前往刷卡，最先刷掉这笔钱的一两个人成功消费，其余所有人都会被拒刷。
这一极具戏剧性的社会实验在全网引起轰动。Anish 评价道，金钱本质上是高度社交化的，但市场上现有的金融产品却大多枯燥、孤立且令人尴尬；如今在 AI 领域也正迎来类似的拐点，创业者完全可以借助大模型构建触及大众社交神经、极具野心且充满趣味的产品。
---
### `01:02:47` 颠覆常理的创业法则：野心没有上限，昂贵的消费级软件正成为新常态
当谈到近年来投资与观察顶尖企业所学到的“反直觉教训”时，Anish 指出，整个硅谷的投资逻辑发生了一场彻底的范式转移。放在三年前，投资机构如果看到一个过于宏大、甚至显得有些疯狂和复杂的项目，通常会选择放弃，因为在当时看来，动辄融 1 亿美元的种子轮毫无意义——没有人能管理这笔巨款，也没有团队能在起步阶段消化如此密集的资本与人才。
然而在今天的 AI 时代，逻辑完全逆转了。现在让顶级投资人失去兴趣的，反而往往是那些“野心太小”的想法。虽然 Anish 并不鼓励创始人盲目去融上亿美元的种子轮，但在评估创业项目时，“野心已经彻底没有了上限”。Anish 回忆起自己加入 Andreessen Horowitz（a16z）时 Marc Andreessen 与他交谈的情景：
> “Marc 在我们讨论我加入这里时对我说了这句话，他说：‘Anish，我们当初为第一支基金募集资金讲述故事的方式就是——我们要么直奔月球，要么在地上砸出一个月球那么大的巨坑，绝没有折中的第三种选择。’”（Anish Acharya 援引 Marc Andreessen 的原话）
在他们看来，现在的创始人们正在挑战数年前看似完全不可逾越的鸿沟（比如重资产、深科技领域的极限探索）。
另一个被彻底颠覆的传统认知是“消费级软件必须免费”。硅谷长期以来的信条是 C 端产品必须走免费扩张、再靠广告变现或增值服务的路线，但 Anish 提出了完全相反的观点：
- 消费者的可支配支出领域正全面洗牌，所有品类都有被重新定义的可能。
- 价格是衡量产品市场契合度（Product-Market Fit, PMF）最直接的标尺。
- 团队在做产品设计时，应该主动做这样一个思维演练：“我们产品的铂金包（Birkin）版本是什么？那个每月能卖 1,000 美元甚至 10,000 美元的高定版本究竟长什么样？”
Anish 强调，昂贵的消费级软件是一个在五年前难以想象、但在如今充满巨大机会的全新物种。
---
### `01:05:08` a16z 创始人的管理哲学与对技术行业责任的担当
Lenny 询问 Anish 与 Marc Andreessen 和 Ben Horowitz 长期共事中汲取的最核心体会。Anish 坦言，自己最心怀敬意、也是这两位创始人深入骨髓的一项特质，是一种对科技行业、乃至对国家和文明的“受托人使命感”（Stewardship）。
Anish 将他们与早期硅谷传奇投资人如 Ron Conway、Brook Byers 以及 Tom Perkins 相提并论，指出他们身上都流淌着一种强烈的行业道德感——渴望在离开这个行业时，让它变得比自己刚涉足时更为繁荣。这种抱负远远超越了“成为全世界最赚钱的投资人”：
- **公共利益优先**：Ben 与 Marc 经常投入大量精力去推动一些极其艰难、对 a16z 本身并没有直接商业收益（或至少不是单一利己）的重要事务；Ben 更多在科技产业生态层面担当责任，而 Marc 则在国家利益与国家发展战略的大方向上发声。
- **重塑硅谷的集体野心**：就在五到十年以前，深科技（Deep Tech）在硅谷极不受欢迎，不仅处于边缘地位，甚至缺乏声望和光环；而如今深科技已成为主流高地位的显学。正是因为 Marc 等人不遗余力地振臂高呼，号召大家去做关乎国家重大利益的核心技术突破，最终深刻重塑了整个硅谷创业群体的价值追求。
---
### `01:06:50` 给产品经理的行动指南：每周交付一次，寻找 AI 带来的微小惊喜
对于那些在 AI 变革浪潮中感到迷茫、急于寻求突破的产品经理与技术从业者，Lenny 询问有什么具体的思维与行为转变建议。Anish 给出了一个极其朴素却强有力的忠告：
> “只要去做出更多东西就好。我知道这听起来很傻，我知道每个人都在这么说，但是求求你们，动手去做一个项目吧。”（Anish Acharya）
Anish 强调，这个项目不需要告诉任何人，也不需要解决什么惊天动地的严肃问题，关键是要把它当成一个试验底盘（Chassis），去上手测试所有最新发布的 AI 模型。人们在面对新技术时，往往只需要经历“一周先略带挫败感、随后收获满满成就感”的动手过程，就会彻底对这项技术着迷（get pilled）并建立起属于自己的产品直觉。
关于实践的节奏与切入点，Anish 提供了极其具体的参考与生动范例：
- **交付频率**：设定一个基本指标，要求自己“每周交付（ship）一次成果”。哪怕项目再微小，只要跑通了闭环就算数。
- **Anish 的个人实践**：在体验 OpenAI Codex 时，Anish 自己动手给妻子做了一份母亲节礼物。这个轻量程序调取了他的短信聊天记录和家庭相册，配上背景音乐，自动生成了一份包含 20 张幻灯片的二人感情长卷，甚至挖出了多年前他第一次约妻子出门时的青涩短信。虽然这只是一次性的生活趣味，但它证明了亲自动手将想法落地的巨大价值。
- **Nikhil Singhal 的启发视角**：Lenny 提到他们共同的好友 Nikhil Singhal（曾与 Anish 在 Credit Karma 共事）的一个精辟观察——人们对 AI 的态度发生由怀疑到狂热的质变，往往是在 AI 为他们创造了某个切切实实的“喜悦时刻”（Moment of Joy）之后。
Anish 最后呼吁所有产品人，大家当初进入这一行，初衷都是为了把自己脑海中构想的画面具象化，而 AI 正赋予每个人前所未有的建造能力。硅谷的迷人之处在于其开放与正和博弈的文化，只要你把做出来的东西发出来，大家都会为你喝彩并基于你的成果继续创造。在进入正式的快问快答环节之前，这种鼓励去探索、去试错、去感受创造纯粹乐趣的倡议，构成了全场对话中最具感染力的注脚。
### `01:10:00` 闪电快问快答：书籍推荐与影视娱乐
访谈进入经典的闪电问答环节，Lenny 准备了五个快问快答问题，首先询问 Anish 最常推荐给他人的两三本书。Anish 推荐的第一本也是他最爱的书是托马斯·索维尔（Thomas Sowell）的《征服与文化》（Conquest and Cultures）。他指出该书探讨了世界各地历史上的征服行为如何导致文化变迁，深刻证明了文化是驱动社会结果的最大因素；作为一个出生在加拿大、后来移居美国并深切感受到两地“抱负文化”巨大差异的人，这本书给他带来了极深的历史洞察。第二本是汉密尔顿·赫尔默（Hamilton Helmer）的《商业七力》（7 Powers），Anish 赞赏其高度智力化的提炼，将护城河、商业理论与复利优势剖析得十分透彻。第三本是他在加入 a16z 前向 Marc Andreessen 寻求书单时得到的推荐——布莱恩·阿瑟（W. Brian Arthur）撰写的教科书《规模收益递增》（Increasing Returns to Scale）。Anish 笑称收到时一度以为 Marc 在捉弄自己，但读完发现它是一本硬核而精彩的著作，解释了为何软件这类事物能产生异常的经济效应，让人明白科技行业的运作机制与正和博弈文化的根源：正如 Anish Acharya 所言：“I'd love to say that we're better people, but I think it may be a sort of better system and structure we work within.”
被问及近期喜爱的影视作品时，Anish 自嘲平时爱看一些“爆米花式”内容。他很喜欢《龙之家族》（House of the Dragon），虽然算不上深刻严肃的大作，但观感极佳。另一部则是他在伦敦出差参加董事会时，在挤满喝啤酒观众的 IMAX 影院观看的电影《奥德赛》（Odyssey），那种在嘈杂剧场里“既独自又融合（alone together）”的社交体验非常震撼。Lenny 随即倒苦水说自己一直抢不到 IMAX 票，甚至写了一个 GrokBot 机器人全天候监控网站抢票，可惜卡在 AMC 官网上极为刁钻的图形验证码（CAPTCHA）上——要求人类选出三个匹配形状的图形，导致机器人一直无法突破。
- **推荐书目：**
- 《Conquest and Cultures》（Thomas Sowell 著）：阐述文化变迁与历史成果的核心动力。
- 《7 Powers》（Hamilton Helmer 著）：商业护城河与竞争优势的经典理论框架。
- 《Increasing Returns to Scale》（W. Brian Arthur 著）：研究软件与高科技行业边际成本及网络效应的学术著作。
- **推荐影视：**
- 《House of the Dragon》（美剧）
- 《Odyssey》（IMAX 观影体验）
---
### `01:12:30` 最心仪的 AI 产品与核心人生格言
关于近期最让人感到惊喜的 AI 产品，Anish 把票投给了个人智能体领域的 Grok（尤其是 GrokBots）。他认为 Grok 表现出一种近乎“疯狂且极具野心（unhinged）”的特质，比如敢于主动缓存凭证、替用户在后台实际执行任务。这种激进的做法通常只会在初创团队身上见到，而 Grok 背后的团队展现出了大公司罕见的冒险精神，同时配合深思熟虑的 UI 与强大的基础模型，打破了基础模型领域只有两强争霸的固有格局。
谈到工作与生活中的核心格言，Anish 分享了一条源自创业时期并在育儿中同样适用的准则：“Don't discover things through painful experience that somebody can just tell you.”（不要非得通过痛苦的亲身经历，去发现别人明明可以直接告诉你的事情）。他坦言自己过去总有一个坏习惯，喜欢凡事亲身体验探索，而不愿向前辈虚心请教，而如今他的孩子们也如出一辙。Lenny 请他举一个具体的商业案例，Anish 回忆称，自己当年创办第一家公司时，试图同时搭建一个移动游戏社交平台并自研游戏工作室业务；当时有资深前辈一针见血地告诫他“运营游戏工作室已经极难，同时做平台和工作室更是难上加难，必须二选一”，但他并未听劝，最终团队花了数年时间才痛苦地验证前辈是正确的。
```mermaid
flowchart TD
accTitle: 创业认知陷阱图解
accDescr: 展示试图同时做平台与工作室导致的失败，对比听取前辈建议专注单点的正确路径
subgraph Trap["曾经的痛苦摸索 (通过教训验证)"]
A["同时启动: 自研游戏工作室"] --- B["同时启动: 移动游戏社交平台"]
A & B --> C["资源分散 / 执行极度困难"]
C --> D["耗费数年痛苦证实决策失误"]
end
subgraph Lesson["人生格言启示"]
E["前辈明确忠告: '挑选其一专注落实'"] -->|采纳捷径| F["专注单点突破"]
F --> G["避免重复踩坑，加速验证与成长"]
end
```
---
### `01:15:00` 音乐、DJ 历程与音乐产业的 AI 复兴
在最后一个问题中，Lenny 提到 Ben Horowitz 曾建议一定要问问 Anish 的 DJ 爱好。Lenny 甚至找到了 Anish 在 SoundCloud 上的账号（主页后缀为 /IllScience），表示自己工作时经常听他的混音集，并询问他对想入门 DJ 的人有何建议与心得。Anish 回顾道，自己从 1995 年便开始打碟，至今已有整整 31 年。对于非科班出身的普通人而言，DJ 是极佳的音乐自我表达方式——它融合了曲目挑选与混音编排等技术门槛。
进入现代 AI 时代，以 Suno 和 ElevenLabs 为代表的模型进一步推高了创作自由度。Anish 分析了音乐媒介的技术演进史：
1. **最初阶段：** 音乐只能在演奏现场亲耳聆听；
2. **留声机出现：** 人类迎来了可录制音乐时代；
3. **盒式录音带（Cassette Tape）：** 这次介质革命至关重要，因为它赋予了普通人自制混音带（Mixtape）、自主编排“个人专辑”的创作能力；
4. **2000 年代广播与流媒体时代：** 创作权再次被剥离，听众退化为单向接收端，这也是音乐行业长久挣扎的原因；
5. **生成式 AI 时代：** 个人构思概念、模型补足专业制作能力的模式彻底普及，人人重获创作权。
Anish Acharya 满怀信心地预言：“And now that people are making music again, I think the music industry is going to be bigger than it's ever been.”
在节目尾声，Lenny 询问听众能为他做些什么。Anish 诚恳呼吁大家不要对 AI 时代感到悲观或灰心，而是要真正动手去构建产品（“Please don't be despondent. Build something and then tag me, and I'd love to see it.”）。他欢迎大家在社交平台 X 上与他交流互动，并大力推荐了 Lenny 播客网络中 Claire、Alina 以及 Nikhil 等优秀创作者的精彩内容，在热烈的致谢中为本期对话画上了圆满句号。
## Notable quotes
> `52:45` 护城河大多是被探索出来的，而不是被设计出来的 — Anish Acharya
> `30:00` I think that we believe that people want to be more productive, but they don't. I think more people want to spend time than save time. — Anish Acharya
> `55:00` Nobody has a growth problem these days. They have a product problem — Anish Acharya
> `13:30` The loop will help you climb to the local maxima, but then it plateaus. And you need some sort of out of distribution thinking. You need human intuition. You need somebody to actually help you land at the base of the next hill. — Anish Acharya
> `20:00` I think the most useful question to ask yourself as a founder CEO is just, hey, if we assume these things are infinitely intelligent and astonishingly cheap, how would we reorganize the company? — Anish Acharya
> `40:00` The most important thing that we can do with AI to change the kind of conversation around it is make important things cheap. — Anish Acharya
> `15:30` I think a lot of PMs are going to realize they're actually not that good at zero to one. And it's much more fulfilling to work on someone else's good idea than your own bad idea. — Anish Acharya
> `01:12:30` Don't discover things through painful experience that somebody can just tell you. — Anish Acharya
## Mentioned
Anish Acharya, Lenny Rachitsky, Claire Vo, Marc Andreessen, Ben Horowitz, Nikhil Singhal, Dario Amodei, David Sacks, Brian Chesky, Eugenia Kuyda, Jesse (Decagon), Ale (Kavak), Sundar Pichai, Thomas Sowell, Hamilton Helmer, W. Brian Arthur, Andy Grove, Ron Conway, Brook Byers, Tom Perkins, Seth Godin, Chris Dixon, Google, Credit Karma, a16z, OpenAI, Anthropic, Kavak, Moderna, Decagon, Cursor, Granola, Suno, ElevenLabs, MSCHF, Wabi, Limitless, Simile, Claude Code, Codex, Lovable, Replit, v0, GrokBot, ChatGPT Work, Instinct, Qwen 3.8 Max, GLM-5.3, Kimi K3, FAL, Minimax, Replika, OpenClaw, Hermes, Maltbook, Conquest and Cultures, 7 Powers, High Output Management, The Hard Thing About Hard Things, Increasing Returns to Scale, House of the Dragon, Odyssey
## Video description
a16z General Partner Anish Acharya joins Lenny Rachitsky on Lenny’s Podcast to discuss why fears of an AI-driven “permanent underclass” may be misplaced, how AI is changing the way companies operate, and why the opportunity may be less about replacing people and more about dramatically expanding what they can build.
Anish lays out his idea that companies are becoming a series of loops, with agents increasingly handling workflows across engineering, sales, marketing, support, and other functions while humans provide the judgment and new ideas needed to move beyond local maxima.
They also explore why Anish thinks consumer AI should focus less on productivity and more on helping people live richer lives, why moats are often discovered rather than designed, how to develop intuition for different AI models, and why his biggest advice for anyone trying to keep up with AI is simple: make more things.
[Source](https://mgln.ai/e/1344/afp-848985-injected.calisto.simplecastaudio.com/3f86df7b-51c6-4101-88a2-550dba782de8/episodes/34aa54e9-999d-484a-8bce-f0f27dc906ca/audio/128/default.mp3?aid=rss_feed&awCollectionId=3f86df7b-51c6-4101-88a2-550dba782de8&awEpisodeId=34aa54e9-999d-484a-8bce-f0f27dc906ca&feed=JGE3yC0V)
Source quotes
[52:45] Anish Acharya: 护城河大多是被探索出来的，而不是被设计出来的
[30:00] Anish Acharya: I think that we believe that people want to be more productive, but they don't. I think more people want to spend time than save time.
[55:00] Anish Acharya: Nobody has a growth problem these days. They have a product problem
[13:30] Anish Acharya: The loop will help you climb to the local maxima, but then it plateaus. And you need some sort of out of distribution thinking. You need human intuition. You need somebody to actually help you land at the base of the next hill.
[20:00] Anish Acharya: I think the most useful question to ask yourself as a founder CEO is just, hey, if we assume these things are infinitely intelligent and astonishingly cheap, how would we reorganize the company?
[40:00] Anish Acharya: The most important thing that we can do with AI to change the kind of conversation around it is make important things cheap.
[15:30] Anish Acharya: I think a lot of PMs are going to realize they're actually not that good at zero to one. And it's much more fulfilling to work on someone else's good idea than your own bad idea.
[72:30] Anish Acharya: Don't discover things through painful experience that somebody can just tell you.
Source quotes
[40:00] Anish Acharya: The most important thing that we can do with AI to change the kind of conversation around it is make important things cheap
[44:48] Anish Acharya: I think that the entire trend of human existence has been that our desires grow faster than our ability to fulfill them
[55:00] Anish Acharya: I've learned that the big ideas are always supported by a dozen small ideas that are invisible.
[73:31] Anish Acharya: Don't discover things through painful experience that somebody can just tell you.
[25:52] Anish Acharya: I think for people who believe the models are commodities or totally fungible, you just haven't actually used the models.
[56:56] Anish Acharya: I always say that nobody has a growth problem these days. They have a product problem.
Original transcript
[0:00] There's a lot of fear and worry about the future with AI. I want to talk about this idea that if you fall behind, you're going to become part of this permanent underclass. It's a funny, dark fantasy that we seem to have as Silicon Valley collectively. Things have never been better by almost every measure. This is a technology that really amplifies our agency. It kind of unbundles skill from desire. Not only can we dramatically drive productivity, we can dramatically drive ambition. Can you get too ambitious? Is there a limit? a company and if what they were trying to do was too ambitious, we would not engage. Today, we're almost seeing the opposite problem. An idea that's too small is not something that we want to engage with. Do you have this interesting take that company building more and more is going to become this kind of series of creating loops? We're going to see this sort of cascading set of everything from a loop per person to loops that can run large parts of the company. With that said, I think humans are a critical ingredient. The loop will help you climb to the local maxima, It's somebody to actually help you land at the base of the next hill. We have this take that the big opportunity is this idea of, Luke, make me happier. We believe that people want to be more productive, but they don't. I think more people want to spend time than save time. So I think that the opportunity for this technology is the basics of consumer need. How do we feel more connected, more loved? How do we make progress? How do we have fun? I don't think it's a model or a capability challenge. It's just a product design challenge. Thank you. while humans step in with judgment, intuition, and new ideas. They also discuss why he's skeptical of fears around an AI permanent underclass, where the biggest opportunities in consumer AI may emerge, and why moats are often discovered rather than designed. And his advice for navigating all of this is pretty simple. Build something. The fastest way to understand where AI is going is to use the models, ship things, and develop your own intuition. Today, my guest is Anish Acharya. Anish is general partner at A16Z, where he focuses on consumer investing. He is one of the most insightful, thought-provoking, mind-expanding, in-the-weeds product investors I've met. He's been at A16Z for over seven years now, And unlike a lot of VCs, and why I loved having Ganesha on the podcast is that he is a longtime product builder and founder. He founded a company called Social Deck, which he sold to Google and then ended up leading a number of efforts within Google. Then he started a new company called Snowball, which he then again sold, this time to Credit Karma, where he moved to VPR product and then GM of the broader consumer product and the entire credit card business. This conversation will get your mind buzzing. Anish, thank you so much for being here and welcome to the podcast. Thank you, Lenny. I'm so excited to be here. Thrilled. I want to start with a very light topic. I want to talk about this meme of the permanent underclass. It's kind of this joke that people have joke about, this idea that if you fall behind, and aren't just like on top of all the latest AI tools, aren't becoming the most productive person ever. You're going to become part of this permanent underclass and fall behind and have a really hard time. And some people joke about it. I think a lot of people take this really seriously and stresses a lot of people out. How real of a concern do you think this really is? How seriously do you think people should take this? Not very seriously. And it's a funny, dark fantasy that we seem to have as Silicon Valley collectively. Things have never been better, really, by almost every measure, by how distributed all the opportunities are, by the kind of technology we have access to, to the types of ambition we're allowed to have. and to the number of companies that are sort of independently working on things that are winning. And yet there's this sort of discussion of permanent underclass being outside of the light cone. I've heard it. And it's not just something for deep insiders or outsiders. It feels like there's a real fear kind of from, you know, researchers at foundation model labs all the way through to the Silicon Valley layman. I mean, here's like a couple of points that I think are really important. So first, I think the last era of tech was a lot more centralized. If you look at network effects, that's sort of the gold standard. You worked on a network effects product. of businesses from the mobile era. And those things led to dramatic centralization, right? Of course, all of them are definitionally sort of the end of one networks. If you look at what's happening now, it's like every part of the stack, there's not even two relevant players, there's like 20. You know, you've got labs, you've got open weight, you've got different variations.
[5:00] within both. If you look at coding agents, we were talking about it, like your mental model for two years ago should have been, would have been, I think it should be winner take all. And yet, Claude Code, Codex, Lovable, Replit, Wabi, like they're all sort of working. So it's really, really encouraging to see that. You know, the second thing, you've heard all the kind of economic data, everything from radiologists who are supposed to be cooked every year for, I think, about 20 years now. And of course, job postings are higher than they've ever been, as well as programmers, you know? So I don't know that the empirical data bears it out. I think the final thing is that there's this sort of discussion about RSI. And I know RSI is like recursive self-improvement is a fun term to throw around. But if you ask the most sophisticated individuals at the labs, it's not actually RSI that's occurring, which could lead to some sort of runaway winner because there were an epsilon ahead of the others. It's autocatalytic effects, which just means you're using the technology to improve your process, but it's not truly recursive. So I think everything from the most empirical to the most technical view points in the other direction, and yet we can't seem to let go of this fantasy. Something I been thinking about recently is even seeing these crazy stories about open AI models hacking hugging face all these stories To me me feels like there always been this question of are we on the fast takeoff or the slow takeoff scenario? And it feels very much so that we are on the slow takeoff scenario because every one of these milestones is like, holy shit, it hacked. We had no idea it was doing this, but we're catching it, we're watching it, we're observing it, we're iterating, evolving. There's always this fear, okay, but tomorrow it's going to take off. What I'm hearing from you is that's probably not the case, which I think is the source of a lot of people's fears, this idea that all of a sudden it's going to become super, super intelligent and then we're in big trouble. That's right. Like the line of reasoning for that case is always everything up until now, then something happens that no one can quite articulate and then fast take off. So I don't believe that that's going to happen. I do think that model progress is happening faster than ever before. But if you look at something like economic diffusion, you know, I grew up in a small town. I went back home last summer, like people's lives haven't changed that much. So if nothing else, the sort of slow rate of economic diffusion will catch it. I think the other thing that's under discussed, Lenny, is, you know, how many problems are truly intelligence bound? Like if you had a, you know, a data center of PhDs working at FedEx or Domino Pizza are they going to be like exponentially dominating supply chain and pizzas Like I don think so So I think we might be overestimating how many problems are intelligence bound versus bound by other things What are you seeing inside of companies in terms of, is there more of a divide happening? And do you think there will be more of a divide between the people that are becoming really good and embracing versus like, I don't have time for this. I hate all this stuff. My job's already so stressful. What are you seeing happening? And where do you think things will go inside of companies in terms of maybe a divide? I mean, I have so many thoughts on this. I don't think we give the average employee enough credit. I think we have this abstraction of a white-collar employee. You know, the white-collar manager, the abstraction is like some Dilbert-esque manager who's just shuffling paper all day long. You know, we have this abstraction of consumers that they're sort of these low-agency NBCs. Of course, that would never apply to us or our friends. You know, we have this abstraction that everybody else's job is super-automatable by AI, but of course, ours is not. But so I think when you actually get into the details, a lot of people are actually excited to better themselves, get more leverage. And you see this with, of course, sophisticated companies like Google, but even a company like Kavak, where they sell used cars in Mexico. They got this concept of a Jedi Academy where they teaching everybody at the company including the mechanics how to use the new tools and technologies And kind of at the end of the six week they ship a cutting in agent So I actually think that more people are embracing the technology than we sort of like to discuss I think a big change is going to be using AI versus reorganizing your entire company around AI. And a great example of this is if you look at the diffusion of electricity as a technology, it took 40 years for us to get from the inception of electricity to reorganizing factories. And that means like burning the buildings down and starting from scratch versus taking what was previously coal and simply swapping it with electricity. So I do think that like the most ambitious companies are rethinking everything around the models and those that are a little less ambitious or perhaps a little earlier are thinking more about how do we give people in existing orgs, existing job functions, access to the technology. So kind of a theme I'm hearing so far is we can be a little less stressed about where things are going and the future of your job, your careers. I think so, man. I mean, I think if you even just think of the kind of incentives for the CEO and executives, you know, Sundar running Google, he doesn't want to run a more efficient $4 trillion company. He wants to build a $40 trillion company. So anytime you have an economically productive unit, it's rational to kind of, especially if it gets more productive, to maintain.
[10:00] and that sort of presence in your organization. And then, I don't know what you hear, but anecdotally, I talked to a good friend who's an executive at Google, and I said, hey, have you laid anyone off? And he said... No, we didn't. What we instead do is now rip through our roadmap. So two years of roadmap happens in three months. And we're actually our hardest problem is knowing what to add to the roadmap, which, by the way, is like every PM's fantasy. You know how much emotion has been drained on prioritization conversations between you and I. So, yeah, I actually think that people shouldn't be as stressed and I think they should feel really empowered. And by the way, the best way to do it is just to ship stuff. I mean, Claire is my muse. She's so awesome because she's always shipping. Yes, Claire Vaux. She's shipping. She's trying things. She's not afraid to be a little embarrassed by it. And if you just see her whole kind of affect she feels like the best version of herself that she's ever been. And I think we all have an opportunity to be that. I love that. I want to be clairvoyant when I grow up. Totally. Kind of along those lines, you have this interesting take that company building more and more is going to become this kind of Series of creating loops and creating series of loops. Talk about that. Yeah, yeah. Well, I think the broad concept and, you know, the loops concept kind of gets teased a little on X because at some times it feels like maybe we're big braining it. I know there was a big meme around graphs, like the next stage of loops is graphs. But let me make the kind of steel man for it, which is. We had prompts, and then we invented agents. Agents, of course, are just models in a loop with tools and memory and skill files. And then we had sort of loops, which are, you know, sets of agents that are doing tasks. If you look at a lot of work in coding, coding is such a great domain because you have the best models and you have the most sort of technically apt customer. Plus, you have these established loops like bug fix or sort of bug report comes in. Repro gets generated, bug fix gets created, bug fix gets reviewed. If high risk, then humans should confirm that it's okay to ship to prod and if low risk, it just gets shipped. And maybe you even email the customer and say, hey, we fixed the bug you reported. That happens in five minutes. There are many loops like that in engineering, everything from bug fixes to customer feedback. to sales demos, to new feature development. So coding sets itself up well. So you have a coding loop and the kind of change it makes is to the code base. My question is, what are the business loops, right? So if you're the GM of a business, you're looking across many job functions and you've got loops running in coding and marketing and sales and support and legal, The output of all of those loops is something that is itself a loop that you should be able to optimize for. And I think the strong form of this is that it sends a message to the CEO saying, hey, we need to actually make a change to one of the physical aspects of the business or to our business model or to our strategy. So I think we're going to see this sort of cascading set of everything from a loop per person, loop per job function, loop across entire business units to loops that can run large parts of the company. With that said, I think humans are a critical ingredient. I just don't think that most work in the organization can be done fully autonomously. When you think of what a human will do in this AI-native company, sales, support, strategy, and exceptions. And all those things are super critical. We've seen one thing, Lenny, it's that the ability for models to do new thinking out of distribution thinking is still really limited. And I don't actually take the point that some of the new thinking in math is actually representative of new thinking in domains like business. So you're still going to need a person to say, hey, here's the thing I think we should make and have them be right about it. Let me just kind of make sure this point is really clear because it's so interesting. What you're saying here is, Engineering and building more and more is becoming the sloop of input and Feedback. or support ticket, whatever input of just like what to build. And then AI more and more is taking that. deciding here's a PR here. Is this ready? And then shipping it. And you're saying that you expect that to spread. to, say, go to market, legal, growth support. So maybe describe what that loop looks like or may look like within a company. I mean, a great example is a growth team. You worked on the growth team at Airbnb, right? Yeah, yeah. Supply growth. Yeah, that's right. Yeah, awesome. Right. So you remember those like, I don't know how you ran your team, but I'm guessing it was something like you got everyone together, you built a list of possible experiments, you prioritized them, you built them, you shipped them, you measured them. Brainstorming. Yeah, a lot of brainstorming, a lot of spreadsheets. So the loop version of that should be that every variant gets generated, every variant gets measured. Once you get to stat sig with a high enough p-value, you converge and ship that variant. You then have a long-term holdout and you start working on the next experiment. And then you're going to hit some local maxima. And I think this is really important. The loop will help you climb to the local maxima, but then it plateaus. And you need some sort of out of distribution thinking. You need human intuition. You need somebody to actually help you land at the base of the next hill. Yeah, you have this chart. I don't know. Maybe we'll show it over...
[15:00] as we talk about this, which is such an interesting way of thinking about it. This idea that agents will help you Hill Climb, End. reach some new plateau and then you need a human there to... think about a bigger idea, kind of unlock it, and then it keeps going and going. And there's kind of this like agent to human kind of back and forth. Yes. Yeah. And you know what really illustrates that? If you've ever tried to have an agent come up with a business idea for you, like, you know, hey, Claude, make me a million dollars, make no mistakes. Like, why doesn't that work? You know, and it's because you sort of need to set it in the right direction and nothing in the technology has shown us that that is not needed. Yeah, it's interesting. Like, I've been hearing more and more. I just saw a tweet that I think at OpenAI, the go-to-market team now is using Codex. More of the go-to-market team is using Codex more often than even the engineering team. Yes. And think about how happy that makes them. Like, what does the go-to-market team want to do? I mean, this is a caricature, but I'm going to stand behind it, which is they want to hit the gym, They want to go to steak dinners and they want to like, you know, like, raise the trophy up for being salesperson of the quarter or of the year. Um, So I actually think that this is a distillation of their job into the thing that they're the best in the world at, that they're the most interested in. And all the administration that goes around doing that core work is now handled for them. So that's where GoToMarket is going, and it's going to be awesome. This reminds me of a PM friend who had this really funny... take that as a PM, You're constantly having to say no to all these ideas that are coming at you. And you have like, I'll put on the roadmap, we'll purchase it. He's like, okay, I'm going to flip this. I'm going to say yes to everything. I'm going to build everything and then simulate everything. of every idea with like Simile or all these products that are launching where you could simulate how a user will react And then that'll tell you should this should we build this? But a hilarious way to rethink PM. And you know what's so beautiful about that? Actually, there's two things. I'll tell you maybe the one that's less obvious to me, which is I feel like every PM at every company feels like they're a true zero to one thinker. but they're held back by the heavy hand of management and executives and founders and engineering capacity. And in a world where every story gets told, every product story gets told, every feature gets tried, I think a lot of PMs are going to realize they're actually not that good at zero to one. And it's much more fulfilling to work on someone else's good idea than your own bad idea. So I think even things like that are going to lead to a lot more organizational health than we've had in the past, not to mention the fact that The idea that ends up winning doesn't have to be the one that's came up with by the person who can sell it best to an executive. It just gets tried and the best idea wins. So going back to this loops idea, the way I'm thinking about it is how can every function start to think of their function as setting up an agent to be able to just go from input to some kind of impact. And what this makes me think about is something actually Claire tweeted recently. This point that people always used to joke that soft skills are the least valuable and engineering skills the most important and valuable because they're so concrete. And it turns out that's what AI is the best at, the things that are verifiable, And, you know, you know what success looks like. And so the question I think about now is just like, Which skills can you not just turn into a loop because the output is so R2, verified. Yeah, I think that's right. And I think that those are going to be the rate limiting factors because you can only do one steak dinner a night. I guess you could do a steak lunch. But, you know, to some extent, there's going to be these rate limiting factors in every system. I think a useful way to think about it is anytime the model's making a mistake or doing something you wouldn't do, What do you know that it doesn't know? And there's a really interesting example I heard from Ale at Kavak. He's probably the most sophisticated thinker on this stuff that I get to hang out with, where he said anytime they have an agent per customer, they sell used cars online. And when the agent gets stuck, it actually calls a human. And the human will coach the agent through. Now, the magic of that is not only does it unblock the agent, but of course, the agent then captures all the traces and learns from it. Like that's a really interesting mental model. It's either a knowledge gap or a data gap that you have to give the agent. And the next time it shouldn't have to call you. I love that example. Basically, the takeaway here is, you need to start thinking about every function as an agent loop. And the job is to... figure out where it gets blocked, where it goes wrong, and give it more context, more insight, more direction, basically. That's right. That's right. And then hopefully a lot of your day-to-day work, you know, when you come up with a new idea that works, the kind of implications of that idea, if it's a product, you know, there's marketing work, there's sales work, there's product marketing, there's communication, there's legal, all of that should be largely handled for you. And your idea, you know, your job is to go take a hike and dream the dreams and come up with the next hill to climb. And this begs the question a bit of just what will separate the company is that. win in this world where AI is kind of doing a lot of this. I imagine part of the answer is the human
[20:00] in this local maxima coming in with a better idea. Is there anything else that you think becomes like a differentiator in this world where, AI is doing so much of the work that humans are currently doing. I mean, I think one thing that's under-discussed is that... It's unclear that the sort of competitive equilibria that exists in a lot of industries will really change. So let's say Pizza Hut and Domino's and Papa John's and Roundtable all get a data center of PhDs. And they're all going to either adopt it or have a CEO change and adopt it. So it'll be a rocky period and there'll be some relative shuffling. But I think that they're all going to embrace new technology because most companies do. I don't know that one of them is going to have 99% of the market So I think kind of what's under discussed is that, yes, in the near term, I think there'll be winners and losers based on adoption of the technology and kind of how ambitiously you adopt it. But I do think there's a lot of industries that will sort of maintain their current competitive dynamics because they're not intelligence bound. I think the most useful question to ask yourself as a founder CEO is just, hey, if we assume these things are infinitely intelligent and astonishingly cheap, How would we reorganize the company? Because that's where we're going. Kind of along those lines, you have this interesting take about this kind of split that might be coming within companies between these generalists and specialists and how AI plays into that. Talk about that. Yeah, well, I think that there's so many interesting things that this touches on. You know, one is the question of open weight versus frontier. So I'm sure are you familiar with kind of Pareto efficiency? I'm sure you are. Please explain. Yeah, so Pareto efficiency is just, you know, for price performance, for example, what is the kind of the efficient frontier is what's considered the optimal tradeoff of. a unit of performance for a unit of price. And, you know, are you sort of if you're along that curve, you're always paying the rational amount for the performance. And what's interesting is that frontier models are actually irrationally priced in that, First of all, Mythos is infinite dollars a token. You can't use it. But if you look at even something like Fable 5, for one IQ, conceptually one IQ of extra intelligence, you're paying 100x more. than Opus 4.8. So it's not rational, but I think there's a lot of jobs in which you have unbounded upside, like drug discovery, you know? And if that one IQ point lets you discover the next, you know, statin, It's a trillion dollar outcome. So it's sort of rational to pay for the highest intelligence in these types of jobs and industries. Now, on the other hand, I'm going to pick on maybe legal issues. There's perhaps only so much upside to be had in legal or finance. And for those jobs, you actually do want to be very Pareto efficient and probably pay for good performance at a good price, not infinitely priced, infinite potential upside performance. So I think what we're going to see is a split between Job functions that demand kind of mid-IQ intelligence, and those will often be open-weight, sort of biased with reinforcement learning, you know, things that make the models even cheaper, more performant for a narrow job. along with incredibly "expensive" but performant frontier tokens, for sales, support, research, engineering. So you're going to end up having both architectures. And, you know, we can touch on this in a little bit, but I do think there are these sort of comparative advantages amongst model families and then also amongst, of course, individual models that we're seeing more and more of. It's not going to be one or the other. Such an interesting insight. So just to make sure I understand what you're describing here. You're thinking there's going to be this split between kind of within an org and Yeah. function and model. where specific functions that have a lot more upside and potential leverage go for the frontier models, And you described these kind of product sales, engineering, research roles. And then for other functions, you don't need... You don't need Mythos. You don't need Astra, I think. Is that the latest one? Yes, Astra. Yeah, yeah, yeah. And so it's both you're saying the AI doesn't need to be the frontier model and the people don't have to be the smartest people in the world to do really well in that world. Yeah, and I don't want to be diminutive. There's extraordinary people in those job functions. I just think that they're bounded upside problems that they work on. You can only kind of close the books automatically. correctly, you know, you can't close the books 100x better. So yeah, that's exactly what I'm saying. I wonder if it's connected back to that discussion we had earlier about it's verifiable if it's a lot. more verifiable. You don't need the frontier model versus, I don't know, the potential upside. I'm not sure. I mean, I think for a... - Verifiability is a good question, because I think for a drug development company, you have infinite upside, it is verifiable. But closing the books is also a verifiable problem. which has limited upside. I think the question is really how... How much upside is there and how hard is it to calculate it? Like, here's a nuance. Customer support, a customer may call in and report a bug, but... That bug may actually be
[25:00] the first breadcrumb, to a thing that changes our entire organization. And if the CEO was on that call or the smartest person, they could follow the trail. But if we actually have this, quote unquote, mid IQ generalist, then not. That actually makes the case for as a basket of problems, always use frontier intelligence. And I can, of course, make the case for the mid-IQ basket as well, which is simply that we've crossed an intelligence threshold for almost every economically useful problem. And anything beyond that threshold is simply waste. Yeah, and I think people... The point people forget is that the models that are not the frontier models They were like, that was what the frontier was. I don't know, six months ago. And we were so impressed and we loved it. It was like, holy shit, I can do all this. And now just because there's something better, we don't give those models as much credit. You're right, because it's sort of like, this thing is so crazy. You know, this is... This is AGI. And then a day later, it's like the old thing that you throw in the dustbin. So kind of along those lines, I hear people jokingly call you a model sommelier. Tell us with your Somalia credentials, what's kind of like the current state of the model art? What is each model great at? What are models terrible at? Yes, well, as you know, the secret of every sommelier is 10,000 hours or maybe 10,000 bottles. So I think that the secret of being a model sommelier is just using them all. I push myself really hard to ship something with every new model that comes out. And I think you learn so much. I think for people who believe the models are commodities or totally fungible, you just haven't actually used the models. And for example, in the last few weeks, I've been obsessed with Quen 3.8 Maxx. Quint is an awesome model. It's very good at long horizon tasks, and it's actually really creative. It's a great storyteller. So I've been using it to create these impossible documentaries, mostly of planets of the Star Wars universe. I did Tatooine, which turned out great. I did Bespin last night. And it can just work for four or five hours. It tells a great story. It generates all the video using the Minimax model via FAL. Generates audio via 11. it actually directs the movie, the five-minute movie, in that it cuts all of the clips in, it overlays it, it watches it. It's just extraordinary, you know? And it's got a totally different shape than GLM-5.2. GLM-5.3 just came out. which I use for a bunch of product work, it doesn't have a vision component. And it's sort of like this neurotic PhD you put in the corner. And they both have the role, right? This is a little bit of the kind of these tension of models. It's not that one is ahead of another, one is more intelligent. It's rather... One is sort of has a mind that's shaped in one direction, perhaps creativity and openness for Gwen. And others that are shaped in other directions, like, you know, neuroticism and precision, like GLM-53. So that I just make something with every model. And that's how I build my intuition. How important is this habit, do you think, for people? Because I hear a lot that it's really important. to be using these models, And kind of a secondary question is, how do you come up with what to do with these models? Because a lot of people want to try these things. They're like, okay, what do I do? I know. Any suggestions? Well, I think almost all of us have got, you know, the most insufferable thing for a long time was your app idea friend. you know, every time you went to have a beer, they're like, let me tell you my app idea. You're like, ah, here we go again. Like, we have got to be the app idea guys now. And all the silly ideas, actually, especially the silly ideas, because those are the ones that often have the most alpha, are the ones that we should all be building. So I've probably got two dozen apps that I've built. I've got one or two large apps that I iterate on, and I think that If you don't have a chassis with which to use the models, it's really hard to come up with an idea from scratch every time. So I'd say like work on something. It's actually better if it's not important with a capital I and then keep finding new ways to invest in it and add to it with a model as a kind of tool rather than as a goal. Does that make sense? Yeah. like maybe even zooming out. I've been thinking more and more, one of the most important habits to build right now is to whenever you're about to do something, ask yourself, how can AI do this for me? Totally. I'm visualizing this like input to response, you know, like the whole idea of there's a space between input and response and meditation helps you. uh, think more deeply before you respond. And I feel like the trick now is insert in that moment. How can AI help me with this? Yeah, no, I think that's, I mean, I'm also a long-time meditator. We should talk about that if you like, but Yeah, I think that's right. I think in our day-to-day knowledge work, sometimes it's less obvious to me. I guess my mind is I've always been a consumer product person. I love products. So for me, it's easier to think of a new feature to add to my DJ streaming app or, you know, my sort of Google reader for X that I use than it is to kind of find a part of my life to automate. But You're right, there's some fun examples. I posted about one a few weeks ago where I had my laptop transcribe everything that was happening in the kitchen. And then it would... or detract
[30:00] screen time for my son's iPad, depending on whether he was being good or bad. So it was like a fun little social experiment. It also had a fun outcome, which is that he recorded a video of himself saying, I love you, dad, over and over and put it next to the mic. To hack the metrics? To hack the metrics. So yes, anything becomes a measure. It's no longer useful. But it was just a cool little social experiment. I think those things are super fun. That is hilarious. And so one of the measures was how often he said, I love you. It was just easy saying things that are positive and pro-social or negative and anti-social. And saying I love you is very pro-social. That's so funny. I had a friend who built this little device that measured... how often he and his kid laugh throughout the day. Oh, that's nice. Yeah. He hacked one of those limitless pendants to do that, and then they just look at that metric every day. Isn't that so, and this is what I mean, you know, man. I think we've spent a lot of the time on the show so far really intriguing, talking about productivity, job loss, kind of all these heady, important topics with a capital I. But what you just described, how often you laugh, like that's not a startup. that's probably not an economically consequential idea, but it is for like the quality of our life. And I think we tend to think that happiness is fixed. But what if it's not? If you and I were doing jobs 100 years ago, what would our jobs be? I promised they'd be less cool than they are right now. And maybe 100 years from now, they'll be that much cooler. So I think that the thing that gets often missed, and this is why I'm a fan of Claire and others, is just like, how can this thing add texture to our human lives, even if they don't have economic consequences. Yeah. Along these lines, I saw somewhere you have this really interesting take that the big opportunity, maybe just in consumer, but broadly is this idea of Loop make me happier. Yeah. Talk about that. Oh, yeah. I mean, I think that we believe that people want to be more productive, but they don't. I think more people want to spend time than save time. There's a reason the biggest products in the world are kind of entertainment and social. So we get at the heart of how do we sort of deliver the value to the consumer? I think for most consumers, you know, sometimes I tease and I say it's like the Instagram AI user versus the ex-AI user. The XAI user is fearful of being outside of the permanent underclass, is really opinionated on GLM 5.3 versus Kimmy K3. They're just so pilled. And then the Instagram person is like, This is like a better Google search, kind of. It's cool. You don't get what all the hype is about. So I think that a lot of the and it's really a product design failure. We have the capabilities to radically transform people's lives. I mean, Lenny, in many ways, we spent 40 years building a technology that enables better spreadsheets. Right. We built this like technology that extends our intellect, but nothing to extend our soul. And I think that we have a little bit of a spiritual hunger, especially as a lot of these cultural institutions have gone away that fulfilled that, especially in the rest of America. where you don't have as many hot yoga classes and Pilates and fasting and Friendsgiving. So I think that the opportunity for this technology is like, hey, the basics of consumer need. How do we feel more connected, more loved? How do we make progress? How do we have fun? Like the things that we all aspire to, the basics, how do we apply the technology to those areas? That's what I want to see more of. And again, I don't think it's a model or a capability challenge. It's just a product design challenge. I love this so much. We talked about this idea of loop. Like grow my business, loop, find me more sales, loop, close support tickets. But the way you're describing here, I think you, the way you had it, I have my notes here, just like loop. improve my health or make me a better friend. And there's no reason the AI can't just think deeply and hard about all those things and figure out a way to actually do this. Yeah. And also, you know, look, there's a we have to dial this in. But I think there's a way that AI can sort of challenge you, can push you, can be disagreeable. This is also why I think startups are advantaged over incumbents. You know, the idea, there's a thousand Google committees, who would roll in their graves at the idea that they're going to release a model that's disagreeable or God forbid it should be like sexually suggestive. But guess what? Those are all parts of human existence. So I think exploring the kind of uncomfortable parts of our social existence are things that startups are uniquely set up to do. And I know you spend a lot of time investing in consumer companies. What I'm hearing here is this is a big opportunity for consumer businesses to basically build a app product that is exactly this loop around improving my connections with my friends and family. I think that's right. And, you know, I think the thing that's held back consumer so far a little bit, and it may be held back is just strong because if you if we kind of put this into iPhone terms, we're in iPhone 2010, right? What is iPhone 2010? I think it's pre-Airbnb... pre-WhatsApp, pre-Uber, pre-all the important kind of consumer companies. So it is early days. But what's held us back is, I think, three things. One is that the models have been expensive. So if you want to do a kind of free to use product, it's that hasn't been easy. The second is that we've kind of had an interface problem. Chat makes sense if you're the
[35:00] agency person in the world, which is Elon and Sam. But for the average consumer, their ideal interface is TikTok. So we need to find something between chat and TikTok. And then the fact that the technology has been so much more focused on productivity than things like human connection and entertainment. I think all those things are kind of up for grabs. Like the open weight models mean things are way cheaper. I think that we're starting to have conversations about things like loop make me happier and And I think that founders like Eugenia, who you should have on the show, she's tremendous, are thinking really ambitiously about user interfaces. Brian Chesky from Airbnb, he started a foundation lab. focused on next-gen user interfaces. So I think all those problems will get solved or they're at least in a better position to be solved than they were two years ago. I want to come back to the whole space of consumer and AI and things like that. But I want to follow this thread a little bit more about just... the optimism around where things might go. There's a lot of fear and fear. worry about the future with AI and just generally. Mark Andreessen, when he came on the pod, had this really interesting take that AI came just in time to save us. because population is declining, productivity is going down, there's all this war, climate, change and all these things and we would be in big trouble if Yeah, I wasn't here to fill that gap. Talk about just kind of your bigger picture perspective on, why you think maybe people are underestimating the positive and optimism around AI? One, I think it's a potential for it to be a sort of emotional, spiritual interface on which we can kind of get leverage and explore aspects of ourselves that have been really buried. If you look at the kind of effect of the Industrial Revolution, it's that there are these scale advantages which are insurmountable. And as much as obviously I'm very pro-capitalism and I love the economy that we live in, I think that centralization, it sort of discourages people. the individual in some ways, and it maybe detracts from their identity. So I think one of the really magical things is this is a technology that really amplifies our identity, our agency. It kind of unbundles skill from desire, for example. If you want to make music, you can make music now. You don't have to know how to play the piano. If you want to be a programmer and make software, you can make software now. So it really amplifies our individuality. It allows us to explore aspects of our lives that we were never able to explore before. And just in terms of the nuts and bolts, we have been in this sort of morass of 2% GDP growth. Like who said that we have to be there? Why can't we be 10 or 15 or 20% And this is a technology with which not only can we dramatically drive productivity, we can dramatically drive ambition. Think of the, maybe this is a caricature, but the 1950s and 1960s, we believed that we could do anything, right? We were coming out of World War II where the entire economy, the entire world mobilized in a way that we didn't think was possible. And one of my theories, Lenny, is that when the stakes are high, we are awesome. When the stakes are low, we are at our absolute worst. And in a lot of ways, I think the world we lived in five years ago felt like a low stakes world, which is why we kind of collectively had a lot of these like side projects as a society which weren't necessarily productive or making any of us happier. And now you've got, you know, it's not just Elon doing everything he's doing, but I think everybody feels like they're climbing the ambition ladder. You ship your bad ideas so you can discover your good ideas, and if you want to You know, build software, great. If you want to build a bridge, you don't have to be a civil architect to know how to do that anymore. So I really think that we have the makings of a sort of dramatically happier, more fulfilled, more productive society. And You know, and yet we're here talking about permanent underclass. I love all of that. That all feels so right. It's so hard to really believe that. But if you actually look at how things have gone so far with the rise of AI, unemployment's down, people are making a lot of money. Obviously, a lot of people are struggling. There's a lot of downsides. data centers causing problems for people, things like that. A lot of issues. But it feels like as an economy and as a country feels like things are going well so far. Like they just almost like cured some kind of cancer. the other day. So, yeah. Yeah, you're right. Moderna just did. I mean, the fact that Dario can write a blog post and say what happens when we cure every disease and then we debate it as a serious topic, Like, what world are we living in here? You know, I also think that we're, collectively worried about, always like, you know, the kind of these abstractions, the world, the average worker, the middle manager, the person who lives in the vicinity of a data center, these are abstractions. But our actual lives seem like they're getting more fulfilled. We're more capable. We're more empowered. So I think that's also something that the revealed preferences tend to show how people are experiencing it individually. And the stated preferences tend to show how people are sort of observing it or believe it's playing out societally. If you ask people if they want a data center in their neighborhood, most will say no. But if you ask them if they use ChatGPT today, most people will say yes. So that's kind of the dissonance. The obvious issue is just the PR. Our NAI has not been
[40:00] Great. There's a lot of fear mongering. Thoughts on that? What's going on there? Do you think that'll change? The most important thing that we can do with AI to change the kind of conversation around it is make important things cheap. There's two things that are extraordinarily important in America that have only gotten more expensive, right? Healthcare and education. If you look at healthcare, 45% is administrative. So if you take a lot of that administrative burden out, you can actually see deflationary health care costs. Also, things like curing every disease. That sounds awesome. You know, GLP-1s also obviously not an AI thing, but I think is a reason to be optimistic about it. deflationary health costs, and then education as well. I think education now has the strongest form of competition, sort of traditional education that it's had in, 200 years. And I think it's going to be very, very good to kind of unbundle learning from institutions. And also, by the way, like, status from credentials. You know, like you don't need a Harvard degree. You just need to get... And that's pretty cool. I saw OpenAI recently slow down their AI development. They paused... their RL kind of phase on their latest model because of What they're seeing... So that's obviously a big concern for people, just how fast and smart these models get. Any thoughts on just that? That's like a big shift now instead of race ahead to the fastest, best model ever. okay, we actually have to slow these things down. That feels crazy. I mean, without commenting on OpenAI specifically, I think that maybe I'm a little skeptical on some of these things where I think the kind of the aura of, that Anthropoc got from having a model that was too dangerous to release was extraordinary. And maybe they had a GPU shortage. Maybe the capabilities were more advanced than they actually wanted. Maybe they actually wanted to keep that proprietary model internal to extend their own lead. So I think there's a lot of sort of confounding factors that would cause you to pull back a little bit. Look, I do take the points about offensive cyber seriously, which is we should harden all of our systems before we make them trivial to penetrate. But I think the sort of concept of the model that's too dangerous to release, it kind of conflates with... marketing, inference capacity, And then also economic considerations like do you want to externalize your competitive advantage or use it to make yourself better? Yeah, I always think about that when I have folks from Anthropik and Ovan, the podcast, just like how an advantage they have when they have the best model. It's crazy, right? Just like that's a loop right there is just have the best model for longer and they can move so much faster. It's crazy, though, to take the other side for a moment, it felt like Anthropica was unassailable, and now OpenAI's had an amazing six months, and open weights are also ripping. So despite the scary concept of this supremely intelligent model that's totally proprietary to one company— So far, the kind of industry trends haven't played that way at all. Yeah, like GrokBot just came out of nowhere and is now like the most amazing AI kind of assistant tool. I'm just hooked on it. Oh, man. I mean, we should talk about the personal agents thing. Like, actually, the three big products that I love here, GrokBots totally nailed it. And also the model underneath it is awesome, right? They came out of your right, out of left field. And that's, I know, a lot of the good work the cursor team did. I think ChatGPT work. It's kind of buried in the UI, but it's really, really good. It's one of the best products they've released. And then there's a startup that's really getting some buzz called Instinct. which has made some more aggressive and interesting trade-offs, but all in the same domain. That's amazing. I don't know if it was a strategy for someone tweeting this vague... tweet about how awesome it is and everyone's like what the hell are you talking about That was really effective. How do I get there? ChatGPT work. The episode that will come out right before this is the Yeah, I'm in charge of that, Tara. Oh, really? Yeah. Oh, man, they did such a good job on it. I don't know how much you use it, but it's really well done. What is better than co-work there? Is it that it runs in the cloud? Is that the big differentiator? It runs in the cloud. It does a good job of kind of caching browser credentials, though it's not as aggressive as Grok or Instinct. Yeah. Also, the remote feature, it's got the full duplex voice mode. So you can actually just call it and can see all your threads and you can just talk to it and say, hey, what's happening across all my coding agents across this? Can you change that? So it's just because the full duplex voice is so good. You really feel like you're calling your assistant. who knows everything that's happening in your world. Whereas with Grot, it's a sort of one-way transcription. Right. Or with some of the other assistants, it's a text. So the voice plus the kind of model that can see all threads is really well done. Awesome. Okay. I want to come back to this AI assistant consumer stuff, but... Just something I wanted to kind of close the thread on. I feel like there's, in terms of jobs and economy, as a result of AI, I see it almost as the spectrum of, There's like the Dario end of the spectrum of 50% of knowledge work will be disrupted and we'll have massive unemployment.
[45:00] to like the Saks, David Saks spectrum of like, it's going to be incredible. jobs are going to be fine. So far, everything is pointing in a good direction. Clearly, you're closer to the sex. direction. Is there anything else just along the lines that you think might make people feel better about jobs? in the future. I mean, I think that the entire trend of human existence has been that our desires grow faster than our ability to fulfill them. If you look at the things that are expectations today, there were unimaginable luxuries 500 years ago, even 50 years ago for things like therapy, right? Or things like antibiotics 100 years ago, right? I mean, yeah. Didn't matter how rich you were, you simply didn't have access to it. So I think we're underestimating human ambition, human desire. You know, people are going to be mad that they don't have a vacation home on Mars. in 20 years, like really mad, like this really mad. They'll be seeing it on Insta and be like, come on, babe, we got to like work harder and make this happen. Every CEO is going to want to build a much bigger company. So I don't think already it feels like we're living a much larger form of sort of human existence than we could have imagined 100 years ago. And there's no reason that trend won't continue or accelerate. This idea of ambition, I'm glad you brought that word up again. It's something that's coming up a lot on this podcast. that not only is AI making it easy to be a lot more ambitious, It's almost making us have to be more ambitious. because everybody else can just do all the easy stuff now. And now it separates us, is just how big can you go? Okay, I'm going to make a personal website. Okay, it's cool. It'll be this white, simple background. No, okay, I'm going to make it this 3D game where you have to walk through a world and discover all the things they do. Everything is just getting more epic. thoughts on just this idea of ambition becoming a bigger, I don't know, skill and habit. A lot of things get conflated because when you and I say ambition on this pod, I think people have a very specific idea of what that ambition is. Admission to be a founder, ambitious to build a beautiful software product. Those are types of ambition, but there are other types of ambition. Let's talk about creative ambition. You know, when you're five, nobody says, Lenny, you're good at painting, but you're bad at drawing. You know, you just have an ambition or a desire to make something and you make it. like that's something that's very unique and can now actually be encouraged. You know, think of very local ambitions. So, They have the NHS in the United Kingdom. I think it's this sort of treasured institution that's not working well. Maybe the way AI shows up in their society is making the NHS as good as the iPhones. That's very specific and local to them, or simply the ambition to be more connected to our family, to be more present parents. So the ambition doesn't have to be sort of ambition in the narrow economic sense. It can be really anything that we want to do more of. Who doesn't have that in their bones? Yeah, that's such like my example is so dumb now that I think about it like that. It feels like the thing we have to. unblocking our brain is like, okay, AI, solve cancer. We're not like... That's where we can start to think now. And it's so unnatural to us, especially as product people that always have to think about the MVP and the constraints. Now we have to Think big. Okay, what's the 10x, the 1,000x version of this? Yes, I know, right? In a way, we have to think small because this thing that we've, our whole lives have been built around how precious software and intelligence is. And now it's totally not precious. Like, that probably will be a harder change for you and I. Yeah. So like all these new little... habits. I think a lot about on the Claude Co. team, they have a principal. They're You know what's better than me doing it? It's Claude doing it. And that just created this habit in everyone on the team. How do I help Claude do this thing for me? And I feel like that's a thing we all have to start to build in our head. And Grokbot's really good at that. Not investor, no affiliation, but it's just like so simple and good at the stuff. I also think that there is so much learning that happens through doing. It's funny, if you look at the number of people that are talking about Vibe Coding versus the number of people that are talking about their projects, people are a little embarrassed. I'm a little embarrassed to talk about a lot of my projects because they don't seem important or substantial enough. But so much of it is learning through execution or shipping or being fulfilled through execution and shipping. I don't think we can underestimate that either. reading where you build to learn and infuse an experience. And it's totally okay for most of it to go thrown away because that's still... building your your muscle. Yeah. So well said. Yeah, it's building as an activity rather than an outcome. Right? Yeah. Which I think a lot of people feel bad. I shipped all these things, but no one's using it. I never use it. And I think that the key here is like, that's actually okay. That's totally fine. Oh, man. I mean, and it happens in every other domain, you know? Like I make a DJ set and like three people listen to it and I listen to it a hundred times. And it's very fulfilling. It doesn't matter. We're going to talk about your DJ stuff later. Okay. Okay, let's come back to consumer stuff. So you focus on consumer at A16Z. What's kind of like what's happening in consumer these days? What's kind of like the landscape? What are you excited about? So I think there's three big areas. We're very early
[50:00] As we kind of discussed, you know, the labs have done a good job, but there aren't as many sort of independent mass market consumer products. Coding agents are awesome and I think a total lightning bolt. You know, I think it's easy to say most people don't want to make code. But the big change in my thinking is that coding agents are a way to interact with the world generally. And you've seen a lot of this on X. People use Cloud Code to edit videos or Codex to... create a game that they play with their kid on an airplane ride. It's sort of this general problem-solving tool that consumers can use in very unique ways. And probably the best example of a company that we've invested in is Wabi, where it's sort of a platform for many apps. People can create them, consume them, share them. So coding agents is one big area I think that's important. Personal agents, we had this Incredible moment around OpenClaw, but guess what? OpenClaw is a dev thing, you know? And Hermes, of course, and Maltbook, All of that is now getting distilled into mass market consumer and enterprise agents that people understand, which is, what we discussed previously. And then the last is, I'm going to call it entertainment. That doesn't fully do it justice. I think that it's a lot of sort of creative tools. You know, Suna has done such an amazing job. It's companionship products, all that. That entire area is, you know, uncomfortable to talk about. So I think it's under discussed, but there's some huge fast growing products there. I think those are the three big areas that we're watching right now. That's really interesting. Just think of these three buckets, coding agents, AI assistance kind of open club, but much simpler and easier and more reliable. Sounds like basically the three described there that you're excited about are Instinct, GrokBot and ChatGPT work. That's right. And then Wabi, I guess, would fit in that first one. It's like a personal coding agent that can build whatever you want for you. Yes. Yeah, that's exactly right. And then entertainment is the other bucket with like, like, yeah, girlfriends and that kind of stuff for it. Yeah. Those companies, it's like crazy. You look, you guys put out these market maps and like five of them are these different like. That's right. Though, though, actually, to be accurate, it's more boyfriends and girlfriends. Actually, you know, the majority of people using companion products are women that are in their 40s and 50s. Yeah. Interesting. Okay, cool. So those are the three kind of areas you think the biggest opportunities will come from. And it's interesting that they connect to this idea of loop. I don't know, make me happier. Like, yeah, these are kind of... This is kind of like the jobs to be done for humans. Make me happy. Make me healthier. Make me a little longer. That kind of stuff. Yes, give me a channel for my ambition. a channel for my sort of fulfillment. And then a thing to do when I'm not doing everything else. That makes sense. Okay. The other question I have for you along these lines is, There's so many companies... And so many startups, so many products, everyone's launching. The speed at which companies and products are shipped is like 1000xing. How do you think about it? durability and moats. when you look at a startup, because a lot of founders get that question. Everyone's getting that question. How am I not a rapper? What are signs that tell you this might be a durable thing? Well, I think there's two important ideas. One is something that Jesse from Decagon said, which I love, and that is that motes are most often discovered, not designed. I think it's really easy. I've done this as a founder to get in your own head about like, hey, I need a business plan that survives scrutiny from MBAs and BCs. I've got to have some really sophisticated, you know, idea of what my moat will be. And for that team, they just started shipping and it developed over time. Another great example of this is Cursor. You know, they were criticized a lot for not having a moat, but it turned out that initially being a high NPS DAU product, was really good. And over time, they captured all the reasoning traces, they trained their own models, the Composer 1, 2 models, and so on and so forth. We know how that story plays out. So moats can be discovered. They don't have to be designed is one. And I think the other is that we seem to have forgotten that the classic moats, none of the classic moats are based on how hard it is to make the software. We're not building self-driving cars. Most of us aren't. So it's network effects, it's scale advantages, it's brand effects, proprietary sort of data or what was historically called a cornered resource. Every moat from five years ago generally is still a good moat. We just need founders that have ambition in those directions. We need more multiplayer products. We need consumer social. We need products that get dramatically better the more you use them, like Towne. So on and so forth. So still read Hamilton, Helmer, and all that stuff still applies. Yes. Love his book. Yeah, he's been on the podcast. I feel like there's only five real business books in the world. Every other one is in the business of selling business books, and they're fake. And his is on my list of five. Are there any other in this list that come to mind real quick around the topic? The two that are so obvious are high output management. which is like, that is as good as it ever was. And then Ben's book, you know, Hard Things is, it was the first
[55:00] honest book about business that was ever written. And that's why founders love it. That's why I love it. Because you read it and you're like, wow, I'm not the only one that's, you know, anxious and feels like a failure and can't tell anyone what I'm going through. Ben went through it too. Two of the most mentioned books on this podcast, turns out. I'm not surprised. So on this mode idea, so say you're a founder and you're just like, you know, you're putting a pitch together, trying to pitch you. Yeah. or other VCs, What's the best way to... talk about a mode? Is it like, can you just say, we're going to discover it? We're not sure. Nobody really knows yet. Yeah, I think that we would happily take a bet on a product that doesn't have a quote unquote moat or durability story if it has a lot of momentum, a lot of craft, a lot of sort of growing engagement. It's actually a thing I've learned over the years. I used to be very worried about people stealing my idea. But I've learned that the big ideas are always supported by a dozen small ideas that are invisible. And even if somebody replicates your big idea, they never see the small ideas that make the big idea work. So I actually think that there's some just... something special in the water with certain products that make them incredibly successful, despite extraordinary competition. I mean, look at granola. Two years ago, it was really criticized. And I don't know that they have a super strong durability story today, and yet it is beloved and dominant. I listen to what the customers are saying more than what the business books say. I just saw Ram put out a report of the fastest growing company's According to their data, granola is like number two or three. It's a tremendous product. The craft is really high. Yeah, and I think about that a lot these days. Because there's so much, like there's co-work, there's... chat GPT work, there's cursor, there's rockbot. And it's crazy how quickly One can switch from one to the other and, The underlying model is not that different. All it really is, most of it is the harness and slash UX process. of the product. And so to me, that tells you there's so much opportunity in the actual user experience being a moat. or at least giving you a lot of time to find something that is durable. A hundred percent. And I think for the people that are at the edge, they pay for all of them because they all have their respective areas of specialization. Yeah, I have many $200 a month plans right now. I know, I know, right. It's a little painful, but yes, me too. And I think Cursor has a $300 a month plan now. So does Grok. Oh, yeah, you're right. Yeah, that was the Grok plan. It's true with that. Grok heavy. RIP. Yeah. RIP cursor. I think they've transitioned away from that brand. Something I've been talking a lot about is the distribution. I call the distribution this new moat, but it's like it's always been a moat, but it feels like more and more. That is actually a massive... advantage because everybody is, there's like a thousand launch videos a day. Everyone's launching, launch, launch, launch. and really the ability to get your stuff into people's feed. Get them. Continue. to be reminded your product six feels like, increasingly is powerful and important. Thoughts on the rising value of existing distribution being a big lever for growth and success? This is so important. And, you know, I asked Chris Dixon about this because he sort of authored the famous Come for the Tools, Stay for the Network. I think the issue is that our entire generation of founders and CEOs were trained on the theory of networks and network building. And as a result, every network that exists today is hyper trained to ensure no one else builds a network on their network. So I actually think that the sort of network effect has gone back to this grassroots, like true word of mouth. When somebody is getting a ton of mentions on X and on YouTube and on Instagram and all these places organically, That is probably the best form of the sort of third-party network effect that you can hope for today. And actually, just like the Web 2.0 era, unlike the mobile era where you had the app store and you had growth hacking and you had all of these sort of little cottage industries, we have to kind of build our own channels off of that word of mouth growth. So in a sense, it's a... It's a purer growth problem, but a harder one than we've had in a couple of product cycles. And to get word of mouth, you need to build something. I always think about Seth Godin's line, build something remarkable. Something that people that is worth remarking about, which is basically, you know, build an amazing product that people want to talk about. which is a, very hard to do. And it makes sense that because there's so much happening, people are just going to pay attention to what are my friends using and saying is worth paying attention to. Well, here's the one thing I would say that here's the hopeful point, which is I always say that nobody has a growth problem these days. They have a product problem. And the reason for that is you can build such a wildly ambitious product in any direction, functional or emotional, You can charge a lot of money for it. So my challenge is like, hey, is it that you have a growth problem or is it a failure of our collective – imagination, If we imagine our product costs $1,000 a month, $10,000 a month, what if our product was a software Birkin bag? What would it have to do to justify that? Okay, let's figure out how we build that.
[60:00] And it comes back to the ambition question. Yes. And... Still, though, you still need some... advantage to get in front of people, to get it out there, at least initially, because there's like a thousand things launching every day. Imagine that's still a big opportunity and I don't know, advantage, which to me makes me feel like incumbents have a huge advantage. They have the products. They can tell you, hey, go use Gemini. And every time you go search Google, Do you feel like it's harder for startups now because of this distribution challenge? I don't know. I think it's easier because Gemini, despite all the kind of heavy cross-selling Google has done, nobody would say they're winning. You know, startups can build in directions that incumbents are uncomfortable building in, like everything we talked around companion, but there's many others. It's true. prices can be pretty high. People are open to paying $200 a month or in the enterprise. They're open to signing million-dollar ACV contracts without really knowing what they're getting. So I think the floodgates are open. It feels like Christmas 2009, where everybody got their iPhone and wanted to download new apps. That'll change at some point. People will feel like they're done and they're tired and they don't want to try any more apps. But the windows are open for now. I think it's easier for startups. That's a really interesting insight. And you're lying about how it's not a distribution or growth problem. It's a product problem is such an important one. Because if your product was that good, people would talk about it and share it and use it. And it can be. I mean, what are the wild social experiments that we're going to see with this technology kind of intermediating them. Um, You know, I think a lot about actually, here's a fun example from a few years ago. Have you heard of Mischief? Do you know Mischief? Yeah, yeah. They're awesome, right? They're sort of like this creative studio that uses technology as their medium. I'm very Web 2.0 in that way, actually. Many of the kind of, you know, the Ev Williams, the Kevin Roses, they were painters, except technology was their canvas. So they created this very funny product called Card Versus Card. where they shipped, I think, 100,000 people a debit card. And then every day they would text those people with, a location that they had to spend the money at, and they would put $100 on the card. And everyone would rush out to spend the money, and one or two would be able to spend it, and everybody else would get declined. And it was just this hilarious social experiment that went hyper viral. And for me, it was always inspiring in the world of fintech because it was like, wow, why don't we build more products like that? You know, money is inherently social. And yet all the financial products we have are so dry and personal and embarrassing. I think there's a sort of similar moment happening in AI right now where we can build these wildly ambitious products that touch on many of our social nerves. We just have to do it. So let me follow that thread. You get to see tons of companies both pitching you and also companies you're working with, your investor in. What are some... counterintuitive lessons you've learned from watching the companies that operate well and have succeeded. uh, Lessons that maybe go against Typical wisdom, startup wisdom. I'll tell you the biggest one. And, you know, in the old days, which is three years ago, we would see a company. And if what they were doing, trying to do was too ambitious, we would... Not engaged. Just too crazy, too complex, too And implied by that is, you wouldn't do $100 million seed because it's too much money for almost any problem. It's too much money for any person to actually manage. It's too much money to build a this sort of talent to absorb, like, it doesn't make sense as an inception round. I think today we're almost in the opposite problem where, you know, an idea that's too small, is not something that we want to engage with. And you can talk about how you put $100 million to work in the seed productively. I'm not recommending you raise $100 billion, but I think that there's this sort of no ceiling on ambition is also showing up in how we're picking companies, and maybe how they're picking us as well. Because when we invest, we tell every founder, like, We're here to help you build the strongest form of your vision. Mark told this to me when him and I were talking about coming here, and he said, Anish, the way we sort of told our story when we were raising our first fund was, we were going to the moon or we were going to leave a moon-sized crater in the ground. and there was no other option. That's sort of how we want to work with our founders as well. I love that point. So like, can you get too ambitious? Is there a limit? Obviously, look at the team and what they're, but kind of the key lesson here is be more ambitious. It's like the opposite of what used to be. Like, here's our wedge. Here's where we're going to go. What you're looking for is just how big is the idea? I mean, look at Adam's. How crazy is Adams? I mean, what an incredible hero's journey for all of us collectively, but also what they're trying to do is something that five or seven or 10 years ago would have felt insurmountable, And now it's like, okay, it's challenging. Let's see, you know? Interesting. Is there anything else that has changed or I guess you've changed your mind about around... what you think it takes to build a successful company these days? I mean, I think that the kind of old wisdom around consumer products have to be free. I'm almost taking the opposite take, which is let's think about consumer products that are extraordinarily expensive. I think every
[65:00] Every part of consumer discretionary spend is up for grabs right now. And I think a really useful, because price is a measure of product market fit, a really useful product exercise is what is the Birkenberg 10,000 a month, 1,000 a month version of our product. So I think expensive consumer software is something new and important that we wouldn't have thought about five years ago. I love that framing because it just pushes you, again, to be more ambitious. Yeah. You mentioned Mark and you work closely with Mark Andreessen, Ben Horowitz. Yes. What's what's one thing you've learned from each of those guys? And there's such extraordinary... leaders, founders. I mean, the thing that I actually feel so grateful to be a part of that they really embody to their core, I think is a feeling of stewardship. for the technology industry and for really the country and the sort of maybe the Western way of living and thinking. And if you look at sort of a Ron Conway or Brooke Byers, Tom Perkins, there was this feeling, I think, of obligation to sort of leave it better than you found it from an industry perspective. And that sort of aspiration goes way beyond just being the best investor in the world, though we want to do that, too. I see them both show up that way over and over again, where they want to do hard, important things that don't directly benefit the firm, or at least not singularly, because they're just sort of important. Ben has done a lot of that in the direction of the industry and Mark in sort of the direction of the country, though, of course, both work on both. Another really cool thing is just to see how I think Mark and Ben, but, maybe the firm a little bit has shaped our collective ambition as a founder community. I think if you look at five or seven or 10 years ago, deep tech was deeply unpopular. It wasn't a high status thing to be working on. It was very fringe. And and now it's become very popular, high status and mainstream. And I think there's a lot of firms that have, of course, pulled in that direction. But I think that someone like Mark has been very full throated in his support of working on sort of capital I important work in the national interest. And all of Silicon Valley has changed as a result. And you guys have had some big wins in the past couple of weeks, investing-wise, too. Yeah, thank you. Congrats. Uh, Final question before we get to a very exciting lightning round. What's your advice to product people? who are trying to think about what they might Want to shift in how they work, how they think, how they operate? to be more successful in the future in their careers and with their companies. Just make more things. And I know it sounds silly. I know everybody says it, but just Please come up with a project. You don't have to tell anyone about it. It can be totally unimportant, but use it as a chassis to use all the new models, ship things, talk about them, build your own intuition. I promise you're, one sort of slightly frustrating and then very fulfilling week away. from being as pilled as anyone. So you just got to use the technology. And if not now, then when, right? This is all of us got in the game. to build the products we saw in our mind's eye, And now you have a chance to do it. So just please, please use the models. And tell me what you built. Text me. tag me. I will reply and respond and engage with you and so will everyone else because the magic of Silicon Valley is that it's a very positive sum mindset. Everybody is building on each other and vulnerability is really rewarded. So I definitely would encourage people to use the models. What's like a good heuristic if you're doing this enough? Is it like build something once a month? Is it like some number of hours per day sitting, talking, building anything that you think might help people be like, OK, you're doing a good job? I mean, just just ship something once a week. And it doesn't have to be crazy. I mean, for example... When I was playing around with Codex, I had it build a slide deck for Mother's Day for my wife that pulled from my text messages. It looked at my photo gallery. It set some music to it. It created a 20-slide deck of our relationship and pulled some cool old text from when I first asked her out. It was a really nice Mother's Day. I mean, it wasn't important. It wasn't something that we might come back to, but it was shipping something. So it can be that small. I know that you're our mutual friend, Nikhil Singal. You worked at Credit Karma with him. He had a really good way of thinking about this. He finds that People, shh. flip on AI and how they feel about it once they find some moment of joy. that it had created for them. And this Mother's Day idea is such a good example. And so I think that's kind of a tip I always think about is just like what's something that just will bring you joy if this works. What's something you can do for someone else? You know, maybe that's a good starting point as well. I love that. Anish, before we get to a very exciting lighting round, is there anything else that you want to share, anything else you want to double down on before we get into the lightning round.
[70:00] I don't think so. I've loved the conversation so far. Me too. And with that, we've reached our very exciting lightning round. I've got five questions for you. Are you ready? Okay. Here we go. What are two or three books? that you find yourself recommending most to other people? Yeah, so okay, Conquest and Cultures is my very favorite book. It's Thomas Sowell, and it just talks about how Conquests have led to culture change in different societies around the world, sometimes positive, sometimes negative. I think to me, it's just the best historic view of sort of culture as the biggest driver of outcomes. And I've experienced a lot of that as somebody who is born in Canada and moved here to America, which has a very different culture of ambition. There's that word again. So Conquest and Cultures is great. Seven Powers, you mentioned, is actually just an awesome book. I think it's a very... intellectual distillation about kind of, you know, moats and business theory and compounding advantages. I really like it. Maybe the third is the one that Mark, I actually thought that he was maybe punking me when he sent this to me. Before I started, I said, Mark, are there any books that I should read? And he sent me a couple. And one was a book, a textbook called Increasing Returns to Scale. which I think is Brian Arthur is the author. It's an awesome book. It's a slightly dense study of why things like software have such outlier economic effects, but it really helps put things in perspective or help me. in terms of like, why does our industry work in the way that it does? Why does our culture work in the way that it does, right? Why are we so positive some? I'd love to say that we're better people, but I think it may be a sort of better system and structure we work within. Favorite recent movie or TV show you have really enjoyed. Oh man, I watch trashy movies and TV. I mean, we've watched House of Dragon. which is pretty cool. Yeah, that's true. I don't know, yeah. Maybe it's not highbrow. It's not important with a capital I, but it was awesome. We loved that. And then I did see Odyssey. I saw it in London. I was there for a board meeting in an IMAX theater packed with people drinking pints and having fun. Amazing. It was cool because it was just a whole, the movie was great, but it was just like a theater experience, like a weird social experience, kind of alone together. Those are two recent ones. I have not been able to get tickets to the Odyssey at an IMAX. I actually have a GrokBot just watching it. the site constantly and finding me good seats. Perfect. Perfect. They've got the dumb capture though on the AMC website that hasn't been able to get through. Really? Yeah. It's like a tricky one. It's a really tricky one. Is it the like select the fruits? Yeah, it's like you have to click three different matching shapes, which is like, come on, you can't do that. Hopefully by the time this comes out, I haven't seen it. I think it's a few podcasts in a row. I'm like, I haven't seen it yet. Oh, man. Okay. Favorite new... AI product, I don't know, favorite AI product recently that you've given me joy. Oh, man. I've spent a lot of time with the personal agents. I think Grok's okay. I'm going to give it to GrokBots only because... It's just so unhinged for it to be so ambitious about sort of caching credentials and getting work done on your behalf. Like, I love it and I expect it from a startup. But they actually are doing things that I think no other sort of big company would do. It's really, really well done. It's a really thoughtful UI. It's got a really powerful foundation model. I think it also is like, okay, wait, maybe this is not a two-horse race on the sort of model side. So I just think the product is like fun and ambitious and and is sort of taking risks that other products like that wouldn't take. Two more questions. Do you have a favorite life motto that you often come back to in work or in life? Oh, man, I've got one. It's I learned this or I sort of, you know, this is from my founder days, but it also is something that's very true of parenting. which is, you know, don't discover things through painful experience that somebody can just tell you. So I unfortunately have had a bad habit of sort of discovery versus learning from somebody who's just a few steps ahead of me. And I find my children have that habit, too. What's one example of something that you wish someone had told you? I mean, for my kids, it's don't touch the hot stove. For my startup, it was like... Literally. My startup, it was don't build a product and a platform at the same time. If you're going to be a platform company, build one. Our first company, we tried to build a social platform for mobile games and be a gaming studio. And, you know, somebody wise told me right away, like, look, being a studio is so hard, much less being a studio and a platform. Pick one. And we did. And it was it took us years to figure out we were wrong. Final question, ask Ben Horowitz what to ask you about. And he just said, ask him about his DJing. He's very good. And I went to your, I found your DJ site. I don't know if that you call it on SoundCloud slash IllScience. It's very good. I'm just listening to it while I work. Any tips for somebody that wants to get into DJing? Any tools you found useful? Any, I don't know, insights that might help someone become better at this? - Totally, I mean, I think that this is why
[75:00] music models so much. DJing has always, I mean, I love DJ. I've been playing for 30 years now since 95, actually 31. Wow. Um, And it's an awesome way to kind of express yourself musically if you're not a classically trained musician. You know, you select the music, you pick the records, you mix them together so it requires some technical skill. But now I think you can go a step further and just make music with the models and And the best part is when you can come up with music ideas and have the models do the kind of strong form of them. So I think music is just such a visceral, satisfying way to kind of, you know, experience and provide experiences in the world. So whether it's DJing or making music. I just suggest everyone do it. I hadn't thought about how DJing has changed now that you have Suno and things like 11 Labs and all these things where you could just generate the music, not have to just. spliced together existing music. I mean, think of the history of it. You know, you went from, okay, you know, first you could only hear music if you were there with a person playing it on an instrument, you know, recorded music on the phonograph. The big change in music actually from a medium perspective was the cassette tape. Because the cassette tape was really the first time you could hear, create. You could actually compose your own quote-unquote album I think a lot of why music struggled in the 2000s is the sort of that went away and we went back to broadcast. And now that people are making music again, I think the music industry is going to be bigger than it's ever been. Wow. Man, so much disruption. Uh, yeah. Anish, this was incredible. We covered so much ground. Final question, how can listeners be useful to you? I mean, show me what you're building. Please don't be despondent. Build something and then tag me, and I'd love to see it. If you're interested in hearing more stuff like this, please follow me on X. I try to engage and follow back. Otherwise, Make sure you check out all the amazing folks in Lenny's network. Claire is a star. Alina is so, so good. And there's just so much compelling content here. Mikael, which we talked about. Nikhil's the best man, that guy is... Everything. Amazing. Anish, thank you so much for being here. Thank you, Lenny. Thank you, everyone. educational purposes only and is not a recommendation to buy, hold, or sell any investment or financial product. This podcast has been produced by a third party and may include paid promotional advertisements, other company references, and individuals unaffiliated with A16Z. Such advertisements, companies, and individuals are not endorsed by AH Capital Management LLC, A16Z, or any of its affiliates. Information is from sources deemed reliable on the date of publication, but A16Z does not guarantee its accuracy. you Thank you. Thank you.
