Claire Vo Declares Product Roadmaps Obsolete as AI Enables Infinite Code

Lenny's Podcast

Because AI has eliminated engineering bottlenecks and made software execution nearly free, product teams must shift from tracking velocity to taking high-conviction experimental swings.

AI-assisted summary of Everyone’s shipping more. Does any of it matter? | Claire Vo — original source. Check the original for full context.

Source published: 2026-09-24

Key ideas

As AI agents drive pull requests up 3x and push engineering effort toward zero, the primary bottleneck in software development has shifted from building capacity to product conviction. (1:50)
Pairing near-limitless AI execution with traditional feature roadmaps traps teams in clearing low-value backlogs, copying competitor features, and churning through launches without compounding learning—turning feature lists into 'ammunition for a very powerful slop cannon.' (8:43)
When every backlog item becomes cheaply buildable ('roadmap zero'), effort-based prioritization frameworks like RICE collapse because engineering effort is no longer a valid proxy for strategic importance. (11:23)
Product teams must adopt a model of 'durable convictions, but disposable features,' classifying work into distinct tiers of Probes, Experiments, and Promises so they can trash mediocre code while protecting customer trust. (15:18)
The industry is shifting from the 'Velocity Game' to the 'Ambition Game,' where AI transformation OKRs should measure how many massive, multi-week experimental swings a team takes per month rather than PR volume. (19:53)

Detailed notes

## Background Published on YouTube by Lenny's Podcast on September 24, 2026, this keynote features Claire Vo speaking live at the Lenny and Friends Summit in San Francisco. According to the page description, Vo argues that while AI makes shipping software easier than ever, traditional feature roadmaps push teams into clearing backlogs and copying competitors instead of learning what matters, making the case for stronger convictions, bigger ambitions, and customer-grounded experiments.

## TL;DR Claire Vo argues that because modern AI tools have removed engineering capacity as the primary bottleneck, product teams now face a scarcity of conviction rather than execution speed. Cheap execution exposes organizations to the traps of burning through backlogs, building commoditized feature parity, and abandoning unvalidated features in a cycle that accelerates mediocrity. To move past obsolete date-and-feature roadmaps, Vo advocates for conviction-led development that pairs durable 1–2 year beliefs with disposable features categorized into probes, experiments, and promises. She urges leaders to pivot from a pure velocity game to an ambition game, measuring progress not by pull requests shipped, but by the scale and audacity of the experiments they run.

## Atmosphere & style Delivered with sharp candor, warmth, and self-deprecating humor, the presentation feels like an invigorating wake-up call for product leaders grappling with AI fatigue. Vo easily connects with the audience through candid confessions—admitting her own automated builds often produce 'code to nowhere'—before shifting into punchy, iconoclastic critiques of traditional roadmaps and OKRs. The emotional arc moves from shared, knowing laughs about industry pressures to an empowering, high-energy manifesto that rallies builders to ditch feature spreadsheets and take massive swings.

## Bulletin - [01:50](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=110s) As AI agents drive pull requests up 3x and push engineering effort toward zero, the primary bottleneck in software development has shifted from building capacity to product conviction. - [08:43](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=523s) Pairing near-limitless AI execution with traditional feature roadmaps traps teams in clearing low-value backlogs, copying competitor features, and churning through launches without compounding learning—turning feature lists into 'ammunition for a very powerful slop cannon.' - [11:23](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=683s) When every backlog item becomes cheaply buildable ('roadmap zero'), effort-based prioritization frameworks like RICE collapse because engineering effort is no longer a valid proxy for strategic importance. - [15:18](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=918s) Product teams must adopt a model of 'durable convictions, but disposable features,' classifying work into distinct tiers of Probes, Experiments, and Promises so they can trash mediocre code while protecting customer trust. - [19:53](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1193s) The industry is shifting from the 'Velocity Game' to the 'Ambition Game,' where AI transformation OKRs should measure how many massive, multi-week experimental swings a team takes per month rather than PR volume.

## Key takeaways - [01:50](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=110s) Traditional product roadmaps were designed for an era of engineering scarcity, where limited delivery capacity and cross-functional friction naturally filtered out bad ideas. - [01:50](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=110s) While AI coding agents have multiplied shipping velocity and PR volume by 3x, revenue is not growing proportionately because execution speed has outrun the discovery of commercializable ideas. - [06:02](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=362s) Automated AI factories can handle bug fixes, eliminate issue tracking, and re-architect codebases dozens of times, yet still produce 'code to nowhere' if the underlying product lacks differentiation. - [08:43](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=523s) The 'Backlog Trap' occurs when teams use AI to burn through every historical feature request on their list, mistaking queue clearance for meaningful business progress. - [08:43](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=523s) The 'Parity Trap' happens when competitors ingest the same customer insights and run identical AI polish workflows, collapsing competitive dynamics into homogeneous, undifferentiated software. - [11:23](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=683s) The 'Churn Trap' sees teams ship a feature, observe ambiguous initial signal, and immediately pivot to building the next cheap feature rather than staying to compound value—accelerating their 'path to mid.' - [11:23](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=683s) 'Roadmap zero' is the state where every visible feature becomes plausible and cheap to build, rendering effort-based prioritization frameworks like RICE obsolete as strategy. - [13:43](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=823s) Customer truth is the ultimate remaining constraint in software; no amount of internal LLM adversarial reviews can substitute for live market contact, real usage data, and repeated testing. - [15:18](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=918s) Conviction-led development replaces feature-and-date roadmaps with four steps: forming 1–2 year durable beliefs, defining upfront validation criteria, deploying a fast AI factory to intersect reality, and allocating tokens and attention based on evidence. - [15:18](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=918s) Teams must maintain 'durable convictions, but disposable features'—practicing 'good stubborn' commitment to the core problem rather than 'bad stubborn' abuse of cheap tokens to move goalposts on failing solutions. - [17:51](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1071s) Because code is abundant but customer trust is scarce, teams should categorize releases into low-friction Probes, durable Experiments, and hardened Promises that customers can safely build workflows upon. - [19:53](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1193s) The tech industry is transitioning from the 'Velocity Game' (optimizing PRs, prototyping, and tech debt cleanup) to the 'Ambition Game' (compressing year-long bets into two-to-three-week experiments). - [21:19](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1279s) Rather than tracking PRs or revenue per headcount, AI transformation OKRs should measure how many huge, high-ambition experiments a team runs each month. - [22:16](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1336s) Product leaders should abandon rigid spreadsheet roadmaps in favor of high-bar AI factories where teams gladly trash unconvincing code and embrace uncertainty about next year's exact feature set.

## Detailed walkthrough ### [00:06](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=6s) Coming for the Product Roadmap

Claire Vo opens the morning session asking the crowd to hype her up after staying out past her usual 8:15 PM bedtime—since she has a little baby at home—at the previous night's party, which someone had texted her calling the "How I AI sauna." Looking back at the previous year's Lenny and Friends Summit, she laughs at her own habit of making grand statements just to be proved completely wrong by the market, which she says "keeps me really honest." Last year, she declared that "product management is dead." Obviously, it isn't dead—the room is filled with amazing product leaders, managers, and executives—but a show of hands confirms that the discipline has completely transformed in the last year.

Since going after PMs didn't quite work out, Claire announces her new target for this year: "This year I'm coming for the roadmap." She jokes to audience laughter that she'll come for OKRs next, "and then my job is over." Having spent over two decades in product, she notes that the roadmap has long been the defining artifact of the industry and of PM careers, meant to show teams where they are going, what is next, and what matters. Yet she believes—fingers crossed—that many people in the room are about to write their very last roadmap.

[01:50](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=110s) The Confession: Unlimited Execution, Zero Good Ideas

Before explaining why the roadmap is dying, Claire shares a candid personal confession. She is shipping more software than ever before, armed with the world's most intelligent coding agents, top-tier developer tools, super-smart models, and customer context piped in via API, CLI, and MCP ("all the letters"). She writes countless agent skills and, having counted just the day before, has 40 Grok bots running. She can build almost anything—which leads directly to her true confession: "I am out of good ideas, y'all. I'm like straight up out of good ideas of things to build. I'm out. I can build. I just don't think any of it's a good idea."

It isn't that she lacks plausible feature requests or things she *could* build; rather, her execution speed has completely outrun her ability to discover meaningful, meaty, and—her favorite word—"commercializable" products in the market. Historically, engineering capacity was the ultimate scarce resource, and the entire reason product management existed. Companies had far more ideas than people and far more demand than delivery capacity. Everyone's role revolved around that constraint:

- **Product Managers:** Spent all their time in meetings, Slack, and spreadsheets prioritizing, sequencing, stack-ranking, drawing aggressive quarterly cut lines, and saying "no." - **Engineering:** Pushed back with "no, not like that"—not with all those features or with perfect architecture, limiting scope to what was feasible. - **Design:** Pleaded, "no, wait for us, like please."

Because of this friction, ideas were narrowed down until only a sliver of the roadmap ever shipped. Today, the bottleneck has shifted from *building* ("what can I build?") to *believing* ("what do I believe is actually worth building?").

```mermaid flowchart LR accTitle: Shift in the Product Bottleneck accDescr: Comparison showing how the product development bottleneck moved from scarce engineering capacity to scarce product conviction. subgraph Past["Past: Engineering Scarcity"] A["Abundant Ideas & Demand"] --> B["PMs, Eng & Design Filter ('Say No')"] B --> C["Tiny Sliver of Roadmap Ships"] end subgraph Present["Present: Conviction Scarcity"] D["Near-Limitless AI Execution"] --> E["3x PRs & Automated Shipping"] E --> F["Bottleneck: What is Worth Building?"] end Past --> Present ```

Despite being "the biggest token maxer on the planet"—celebrating PRs being up 3x, agents everywhere, shipping constantly without PMs or specs—Claire warns that the gap between building capacity and true heart-of-hearts conviction ("does any of this code matter?") is a much bigger problem than anyone wants to admit. Teams face immense pressure from boards, timelines, peers, and bosses to move faster, prove they are AI-native, and get leverage. Yet when she polls the room on who is shipping more than ever versus whose revenue is growing proportionately with their pull requests, the disconnect is obvious: "that's the fundamental problem we're facing here."

[06:02](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=362s) The ChatPRD "Product Graph" Story and Code to Nowhere

To bring this problem home, Claire shares a story about building a feature called the "Product Graph" for her product, ChatPRD. It was a classic "product for product managers" concept that many competitors had already done: suck in everything a company knows about its customers and projects, parse it, slap Claude Opus on it, and make it accessible to agents so teams can make better decisions and write better PRDs.

She built the entire insights engine and a fancy semantic product graph that auto-generated a wiki consumable by agents. It was nearly a one-shot build, it worked, it looked good, and it matched competitors in the market feature-for-feature. Yet every time she worked on it, she thought, "This just belongs in the trash. All this work, all this amazing product belongs in the trash." Although it had an obvious market and customers called it interesting, it was merely competitive, not differentiated. Deep down, she felt the ROI was shaky, it wasn't worth her customers' attention, and its traditional web interface wasn't sufficiently "agent-first" or surprising in a way competitors couldn't imagine. The tokens and roadmap had executed her bet, but failed to test, validate, or strengthen her conviction.

Meanwhile, sitting on this pile of "code to nowhere," her automated AI factory kept humming along on its own:

- **Automated customer fixes:** Incoming bug reports were routed directly to one of her bots and fixed automatically. - **Obliterating tech debt:** She re-architected ChatPRD 70 times simply because she could. - **Death of issue tracking:** Her team stopped tracking issues altogether, deciding, "we're just going to ship PRs."

The self-running AI factory was magical, doing everything the industry dreamed of—yet she confesses, "I'm not sure any of it mattered."

[08:43](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=523s) Why Roadmaps Failed Scarcity and the Traps of Cheap Execution

This realization forced Claire to rethink the core purpose of product management, software development, decision-making, and how we allocate tokens. Roadmaps made sense under engineering scarcity: they laid out strategy, defined effort (back when effort meant much more than sending a voice note to Devin asking it to build something), and organized sequencing, milestones, and MVPs. Crucially, engineering scarcity *theoretically* filtered out bad ideas, because everything "below the cut line" never had enough engineers or organizational conviction to get built. Now, without that natural filter, "we can ship all our bad ideas—like, congratulations to us."

Cheap execution still requires human judgment. Combining a traditional roadmap with near-limitless execution capacity creates three major traps, the first two of which she details here:

1. **The Backlog Trap:** If you have a backlog, AI will build every single item on it—all it takes is typing `/go` to burn right through the list. However, clearing out a queue of feature requests or old ideas does not equate to meaningful progress for your business or against a real customer problem. 2. **The Parity Trap:** Illustrated by her Product Graph story, every competitor is talking to the same customers, ingesting the same insights, arriving at the same obvious conclusions, and building the exact same obvious product. Everyone is even running the exact same AI polish workflows: "run the taste skill, run impeccable, run whatever, make it look good, get rid of the em-dashes, get rid of the borders." Because everyone is token-maxing in the exact same way, the once-interesting competitive dynamic has collapsed into homogeneity.

[11:23](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=683s) The Churn Trap and "Roadmap Zero"

Claire Vo examines the third failure mode in software development: the "churn trap." Teams ship a feature, see ambiguous noise or initial pick-up, and because building the next thing is so cheap, immediately abandon what they just launched to ship something else. They never stick around long enough to learn or compound their progress. From the inside, an organization caught in these traps feels sensationally productive: the backlog is shrinking, competitor features are matched one by one, and shipping velocity is higher than ever. Yet this perceived momentum is an illusion: "what I think this does is it just like accelerates our path to mid."

```mermaid flowchart LR accTitle: The Churn Trap Cycle accDescr: Diagram illustrating how rapid feature production without compounding leads to mediocre products. subgraph S1["Activity"] A["Ship Feature"] --> B["See Signal / Noise"] end subgraph S2["Abandonment"] B --> C["Move to Next Feature"] C --> D["Never Compound or Learn"] end subgraph S3["Outcome"] D --> E["Path to 'Mid'"] end ```

This phenomenon leads directly to what Vo terms "roadmap zero." She clarifies that this does not mean an empty backlog where every task is finished; rather, it describes a state where every visible feature has suddenly become both plausible and buildable. When everything in the backlog can be implemented easily, the backlog itself loses its organizing purpose. Historically, buildability and engineering effort served as crude proxies for strategic importance, giving rise to frameworks like RICE prioritization. When effort effectively drops toward zero, standard prioritization ceases to function as strategy:

- "Buildability and effort stop being like this meaningful proxy for what matters." - When several variables in prioritization formulas no longer mean anything, prioritizing is no longer an effective way to guide product direction. - Speeding along an old roadmap with an AI engine simply pushes companies into strategic traps at machine speed, because an AI factory cannot differentiate between an idea and a genuinely consequential bet.

[13:43](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=823s) Truth as the Remaining Constraint

Because AI accelerates delivery, the fallout from weak judgment arrives at the doorstep far quicker than in the past. Bad ideas turn into operational headaches almost instantly. For this reason, Vo declares traditional roadmaps dangerous and obsolete. Instead of asking what can be built, teams must ask: "What do I believe strongly enough to go out and try and prove?" The scarce constraint in software is no longer engineering capacity, code, or the capability to build features—it is customer truth.

No matter how many internal adversarial reviews a team runs with language models, an untested assumption cannot be turned into truth without real market contact. Teams need live customers, real usage data, repeated tests, a distinct perspective, and an exceptionally high quality bar. While AI enables significantly faster contact with reality, that velocity imposes a much stricter obligation to confront reality honestly and accept what the market reports back. Historically, product leaders claimed to favor "outcomes over outputs," yet preserved roadmaps fixated purely on features and calendar dates. In an era of roadmap zero, shipping off a feature backlog turns that list into "ammunition for a very powerful slop cannon."

[15:18](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=918s) Conviction-Led Development and Disposable Features

Instead of relying on dated artifact roadmaps, Vo outlines an upstream, conviction-driven process focused on what must be validated rather than what needs to be coded.

```mermaid flowchart TD accTitle: The Conviction-Led Process accDescr: Four steps for replacing traditional feature roadmaps with conviction-driven product strategy. subgraph Stage1["1. Conviction"] A["Form Durable Beliefs<br/>(1-2 year horizon)"] end subgraph Stage2["2. Criteria"] B["Define Upfront Evidence<br/>(What proves or disproves it)"] end subgraph Stage3["3. Velocity Engine"] C["Deploy Fast AI Factory<br/>(Intersect reality quickly)"] end subgraph Stage4["4. Allocation"] D["Allocate Resources<br/>(Invest tokens, effort, attention)"] end A --> B B --> C C --> D ```

This operating model hinges on an essential asymmetry: teams must hold **durable convictions, but disposable features**. When discussing this with Mara (a summit speaker), the objection arose that enterprise buyers rely heavily on explicit feature roadmaps. Vo responds that consumers and markets have already learned to absorb high levels of feature churn, shifting their trust toward teams, visions, and spaces rather than single fixed features. Success requires demonstrable progress against core convictions paired with zero ego regarding the specific solutions:

- **Good stubborn vs. bad stubborn:** "Good stubborn" means staying committed to solving the core problem while endlessly refining or throwing away solutions, regardless of passing trends. "Bad stubborn" means abusing cheap AI generation ("tokens") to move the goalposts—continually shipping minor iterations without facing up to poor results. - **Systemic resilience:** Teams must design their workflows to absorb disposable features without lowering the bar on quality or learning rigor.

[17:51](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1071s) Probes, Experiments, and Promises

Not everything pushed to production should be treated as a binding promise. Vo recalls older enterprise playbooks where product managers would go "hand on heart to go to market" and swear on their children that an exact set of features would ship by an agreed calendar date so sales could write them into contracts. Modern product teams need to classify their work into three distinct levels of commitment:

| Commitment Tier | Definition & Role | |:--- |:--- | | **Probe** | An exploratory bet or low-friction probe designed to gather early signal in the problem space. | | **Experiment** | A deeper, durable test run against a core conviction; teams commit to sticking with the problem until they determine if they can get it right. | | **Promise** | A hardened, dependable capability that customers can safely build business workflows upon, and where the product team commits to compounding value over time. |

Vo reminds the audience that "code is abundant, customer trust is not." Citing her own experience building out a product graph feature, she kept it guarded behind feature flags, rewriting it repeatedly until she possessed absolute conviction. Shipping too early would risk customers building on half-baked foundations, squandering hard-earned trust.

[19:53](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1193s) Shifting from the Velocity Game to the Ambition Game

Roadmaps should not disappear, but they must be completely reframed around scale and ambition instead of minor incremental deliverables. A roadmap should articulate the future world the team believes must exist, the evidence required to validate progress, the criteria that would prove the hypothesis false, and the milestones that justify investing further time, tokens, and customer attention.

Vo summarizes the transition occurring across the technology industry: - **The last 12–18 months (The "Velocity Game"):** Coinciding with the explosive arrival of modern LLMs ("the year of our Claude"), product teams focused obsessively on pure velocity—PMs writing PRs, widespread prototyping, and training agents to clean up tech debt and spit out features to accelerate feedback loops. - **The upcoming year (The "Ambition Game"):** Pure speed is table stakes. The real differentiator moving forward is using that acceleration to take massive, unprecedented swings—running ambitious experiments in two to three weeks that previously would have demanded a full year of engineering effort.

[21:19](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1279s) Redefining OKRs Around Big Swings Instead of Raw Velocity

Claire argues in favor of making very large investments in very ambitious builds, noting that this is far more important than raw feature velocity—and that the traditional roadmap as it currently exists simply does not serve that goal. Recalling a recent conversation, she jokes about returning to the topic of OKRs, which are "unfortunately not dead yet." A peer had asked her what their OKRs should be relative to their AI transformation, wondering if they should track metrics like PRs or revenue per headcount.

Claire's response cuts straight to the heart of her philosophy, complete with a quick snort of amusement: the metric should simply be "how many huge experiments are you running a month?" She does not care or even know what the specific experiments should be ahead of time; what matters is how many big swings a team is taking every month under the presumption that most of them will not work out. She frames this as a completely new way to think about building a roadmap and doing product work—one that is "big on the ambition and very fuzzy on the specifics."

[22:16](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1336s) Build Your Last Roadmap: A Call for Conviction and a High Bar

Closing out her talk, Claire offers a final "ask slash gift" to the audience: "Build your last roadmap." She quickly clarifies that she does not mean teams should stop planning altogether. Rather, she is calling for an end to the era of maintaining spreadsheets filled with lists of ideas and features, guessing their impact, slapping a date on them, and then locking arms to declare, "we'll never change anything because this is how we work." That day, she insists, is over.

Instead, she urges product leaders to raise their conviction and ambition, come up with big ideas, and define what success looks like in a year or two—even though that requires making some big guesses at a time when she cannot honestly tell what is going to happen in a week or two. The goal is to build "a factory that can surprise you," which requires the willingness to discard software and hold that factory to a very high bar.

```mermaid flowchart LR accTitle: From Traditional Feature Roadmaps to an AI Product Factory accDescr: Contrasts the old spreadsheet-based roadmap approach with Claire Vo's proposed high-ambition, high-bar experiment factory. subgraph Old["The Old Way (Over)"] A["Feature lists in spreadsheets"] --> B["Guessed impact & fixed dates"] B --> C["Refusal to change the plan"] end subgraph New["The New Way (The Factory)"] D["High conviction & 1-2 year vision"] --> E["Huge monthly experiments (big swings)"] E --> F["Trash bad code & hold AI to a high bar"] end Old -.->|"Build your last roadmap"| New ```

Claire walks through a rapid-fire series of hard truths about this new reality, punctuating each one with a decisive "Good":

- **Throwing away code:** "You're going to write a lot of code and then you're going to trash it. Good. Your customers do not need trash products." - **Being out-ideated by AI:** "Your AI is going to have better ideas than your team. Good. We need good ideas. And you need to hold that AI to a very, very high bar." - **Embracing uncertainty:** "You're not going to be able to promise your team, your customers, what next year looks like. Good. Because it's probably going to be better than you can imagine right now."

She leaves the room with one final encouragement to write their last roadmap, promising that "it is extremely fun on this other side," before wishing everyone a great rest of the summit and inviting them to come say hi later that afternoon to warm applause.

## Notable quotes > [00:06](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=6s) This year I'm coming for the roadmap. — Claire Vo

> [01:50](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=110s) I am out of good ideas, y'all. I'm like straight up out of good ideas of things to build. I'm out. I can build. I just don't think any of it's a good idea. — Claire Vo

> [06:02](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=362s) This just belongs in the trash. All this work, all this amazing product belongs in the trash. — Claire Vo

> [08:43](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=523s) we can ship all our bad ideas—like, congratulations to us. — Claire Vo

> [11:23](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=683s) what I think this does is it just like accelerates our path to mid. — Claire Vo

> [11:23](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=683s) Buildability and effort stop being like this meaningful proxy for what matters. — Claire Vo

> [13:43](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=823s) What do I believe strongly enough to go out and try and prove? — Claire Vo

> [17:51](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1071s) code is abundant, customer trust is not. — Claire Vo

> [21:19](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1279s) how many huge experiments are you running a month? — Claire Vo

> [22:16](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1336s) You're going to write a lot of code and then you're going to trash it. Good. Your customers do not need trash products. — Claire Vo

> [22:16](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1336s) Your AI is going to have better ideas than your team. Good. We need good ideas. And you need to hold that AI to a very, very high bar. — Claire Vo

> [22:16](https://www.youtube.com/watch?v=VM5kuvWgwDY&t=1336s) You're not going to be able to promise your team, your customers, what next year looks like. Good. Because it's probably going to be better than you can imagine right now. — Claire Vo

## Mentioned Claire Vo, Lenny's Podcast, Lenny and Friends Summit, How I AI, Grok, Slack, ChatPRD, Claude Opus, Claude, Devin, Mara

## Video description AI has made it easier than ever to ship software, but how do you know what’s worth building? In her talk at the Lenny and Friends Summit, Claire Vo argues that traditional feature roadmaps can push teams to clear backlogs, copy competitors, and keep shipping without learning what matters. She makes the case for bigger ambitions, stronger convictions, and experiments grounded in real customer evidence.

Recorded live at Lenny and Friends Summit on September 10, 2026, in San Francisco.

[Source](https://www.youtube.com/watch?v=VM5kuvWgwDY)

中文简报

Claire Vo：传统产品路线图在AI时代已死由于AI工具已使软件执行几乎零成本，产品团队必须从追踪功能交付速度转向建立长期信念并开展高风险实验。 Claire Vo认为，现代AI开发者工具已经彻底颠覆了软件开发模式：工程产能目前近乎无限，而产品信念和高价值构想反而成为了真正的稀缺瓶颈。在执行成本极低的新时代，依赖传统功能路线图会使团队陷入危险陷阱——清理毫无意义的积压工作、盲目追赶竞品功能对齐，并以机器般的速度疯狂产出低影响力的代码。面对“零路线图”（Roadmap Zero）以及传统优先级评估框架的瓦解，团队必须从追踪交付时间表转向系统性验证需要对照现实去证明的假设。Vo呼吁产品领导者拥抱“持久的信念与可抛弃的功能”，将工作划分为明确的承诺层级（探索、实验与承诺），并从单纯追求交付速度转向大胆、高抱负的重大尝试。 Claire Vo认为，传统的“功能+交付日期”型产品路线图已经过时，因为AI代码智能体与前沿模型消除了工程稀缺性，将核心产品瓶颈从执行能力转移到了战略信念上。 (3:14) 当构建软件的成本近乎为零时，团队会失去开发者带宽这一天然的优先级过滤器，并陷入“零路线图”（Roadmap Zero）困境——此时RICE评分机制失效，团队容易跌入积压工作、功能对齐和低效空转的陷阱。 (8:44) 即使自主智能体对ChatPRD进行了70次架构重构，并一次性构建出由Claude 3 Opus驱动的产品图谱，Vo依然通过功能开关将其关闭，因为廉价的执行无法替代市场差异化或真实的客户真相。 (6:05) 产品团队必须用“持久的信念与可抛弃的功能”模式替代静态路线图，将面向客户的工作划分为探索（Probes）、实验（Experiments）和承诺（Promises），从而避免将每次发布都视为永久承诺。 (15:18) 领导者不应通过合并的PR数量或人均营收来衡量AI转型，而应通过追踪团队每月开展了多少项大规模、高风险的实验，从而从“追求速度”转向“追求野心”。 (21:19)

Reading summary

Claire Vo argues that modern AI coding tools have eliminated engineering capacity as the primary bottleneck in software development, shifting the core constraint from building features to establishing true product conviction. As a result, traditional feature-and-date roadmaps have become obsolete, frequently trapping teams in low-differentiation cycles rather than driving business growth. The Bottleneck Shifts from Execution to Conviction (1:50) Vo contends that while developers and product leaders can now ship code at unprecedented rates using AI coding agents, model APIs, and automated tools, execution speed has outpaced the ability to discover commercially viable ideas. Historically, engineering capacity served as the software industry's defining constraint: product managers, engineers, and designers filtered ideas down to a tiny fraction of proposed roadmaps by routinely saying no to unfeasible or lower-priority requests. With building capacity effectively unbounded, Vo notes that pull requests have surged—tripling across many teams—yet revenue growth rarely scales in tandem. She claims the fundamental challenge of modern software development is no longer determining what can technically be built, but finding concepts teams genuinely believe are worth building. The Three Traps of Low-Cost Software Production (8:43) When near-limitless execution is paired with traditional roadmap thinking, Vo observes that teams consistently stumble into three failure modes: the Backlog Trap, where autonomous agents clear out feature backlogs without solving meaningful customer problems; the Parity Trap, where competitors consult identical customer signals, apply standard AI workflows, and build identical features; and the Churn Trap, where teams abandon newly launched features at the first ambiguous signal to chase the next cheap build. Vo illustrates the Parity Trap using her own experience building the "Product Graph" for ChatPRD. Although the insights engine functioned smoothly and matched competing products, it lacked meaningful differentiation, resulting in what she describes as "code to nowhere." Vo warns that unguided velocity creates an illusion of high productivity while merely accelerating a company's descent into mediocrity. Roadmap Zero and Market Truth as Constraints (11:23) Vo introduces the concept of "roadmap zero" to describe an operational state where every backlog item becomes both plausible and buildable. Under this condition, traditional prioritization frameworks like RICE collapse because engineering effort drops toward zero, removing effort as a meaningful proxy for strategic value. Relying on an automated factory to burn through feature queues merely generates unvalidated software at machine speed, turning backlogs into ammunition for a "slop cannon." Vo asserts that customer truth is the only remaining constraint in software development. While language models can run internal simulations, they cannot manufacture truth without live customer contact and usage data. Because flawed decisions turn into operational liabilities much faster than in the past, organizations face a stricter obligation to test ideas against reality rather than relying on internal assumptions. Commitment Tiers and Disposable Features Replace Roadmaps (15:18) Rather than committing to static feature lists, Vo recommends an operating model characterized by durable convictions on a one- to two-year horizon paired with disposable features. She distinguishes "good stubborn"—persisting in solving a core problem while ruthlessly discarding failed solutions—from "bad stubborn," which misuses cheap token generation to iteratively polish weak ideas without confronting poor market response. To manage external expectations, Vo outlines three tiers of commitment: Probes, which are low-friction tests designed to gather early signals; Experiments, which are sustained tests run against a core conviction; and Promises, which are hardened, dependable capabilities that customers can safely build business workflows upon. Because code is abundant while customer trust is scarce, Vo emphasizes that speculative work should remain behind feature flags until conviction is fully proven. Trading Tactical Velocity for Large-Scale Ambition (19:53) Vo characterizes the previous 12 to 18 months as the "Velocity Game," during which teams focused heavily on tactical acceleration, rapid prototyping, and training agents to resolve bugs and tech debt. Looking ahead, she argues that the competitive differentiator must become the "Ambition Game," using AI capabilities to run bold, two- to three-week experiments that previously would have required a full year of engineering effort. To align teams with this shift, Vo suggests replacing velocity-based OKRs—such as pull request counts or revenue per headcount—with metrics tracking the volume of large, high-risk experiments attempted each month. She advises product leaders to abandon spreadsheets filled with estimated dates and fixed feature lists, urging them to set high-conviction goals while accepting that significant amounts of written code must be discarded to keep quality high. Vo concludes that while AI factories make building software effortless, they cannot provide strategic judgment or guarantee market resonance. The unresolved operational challenge for product organizations is whether leadership can resist the temptation of friction-free feature production to pursue high-risk, high-conviction bets where code is treated as disposable and customer trust is preserved.

Source quotes

[0:06] Claire Vo: This year I'm coming for the roadmap. [1:50] Claire Vo: I am out of good ideas, y'all. I'm like straight up out of good ideas of things to build. I'm out. I can build. I just don't think any of it's a good idea. [6:02] Claire Vo: This just belongs in the trash. All this work, all this amazing product belongs in the trash. [8:43] Claire Vo: we can ship all our bad ideas—like, congratulations to us. [11:23] Claire Vo: what I think this does is it just like accelerates our path to mid. [11:23] Claire Vo: Buildability and effort stop being like this meaningful proxy for what matters. [13:43] Claire Vo: What do I believe strongly enough to go out and try and prove? [17:51] Claire Vo: code is abundant, customer trust is not. [21:19] Claire Vo: how many huge experiments are you running a month? [22:16] Claire Vo: You're going to write a lot of code and then you're going to trash it. Good. Your customers do not need trash products. [22:16] Claire Vo: Your AI is going to have better ideas than your team. Good. We need good ideas. And you need to hold that AI to a very, very high bar. [22:16] Claire Vo: You're not going to be able to promise your team, your customers, what next year looks like. Good. Because it's probably going to be better than you can imagine right now.

Original transcript

[0:06] Good morning everyone. How's it going? [0:09] Everybody excited to be here? [0:12] >> Okay, for those of you that were in the [0:14] Somebody texted me, they called it the [0:16] How I AI sauna last night, the party we [0:19] had. I told you all you need to hype me [0:21] up this morning because I was out way [0:23] past my bedtime, which is 8:15. I've got [0:26] a little baby. Um but I am here prepared [0:30] to do the thing that I do. So last year, [0:34] last year I came here and I said product [0:36] management is dead. I said that because [0:38] as a product leader I like to make these [0:41] big grand statements and then be proved [0:43] completely wrong in the market. Just [0:45] keeps me really honest. So product [0:47] management is obviously not dead. There [0:50] are so many amazing product leaders, [0:51] product managers, product executives [0:53] here in this room. But who's with me [0:56] that between the last time we had [0:57] Lenny's summit and now product [1:00] management is completely different. [1:03] Okay, got some hands. It's really [1:05] different. And so last year I came for [1:08] PMS that didn't work. Let's do it again. [1:10] This year I'm coming for the road map. [1:14] The Yeah, you're you're welcome. OKR's [1:19] next and then my job is over. [1:23] >> [laughter] [1:23] >> So the road map has been the defining [1:26] artifact of our industry, of our [1:28] careers. It's supposed to tell us and [1:31] our teams where we're going, what's [1:34] next, and what matters. And I have been [1:38] in product for over two decades. And I [1:42] believe, fingers crossed, a lot of us in [1:45] our room in this room are about to write [1:47] our last road map. Before I go into why [1:50] I think that, I want to tell you all I [1:53] have a confession. [1:56] I am shipping more than ever. I have [1:59] access to the most intelligent coding [2:03] agents in the world. I have the best [2:05] developer tools anybody can ask for. I [2:09] have super smart models. I've got [2:13] customer context via API, via CLI, via [2:17] MCP. all the letters I have. I write so [2:22] many skills. And y'all, I counted [2:24] yesterday. I have 40 Grock bots. I can [2:28] build almost anything. And here is my [2:32] true true confession. [2:35] I am out of good ideas, y'all. I'm like [2:37] straight up out of good ideas of things [2:42] to build. I'm out. I can build. I just [2:45] don't think any of it's a good idea. So, [2:47] and and I'm not saying I don't have [2:50] plausible requests. It's not that I'm [2:52] saying I don't have things I could [2:54] build. It's just that execution has [2:58] outrun my ability to discover [3:02] meaningful, meaty, my favorite word, [3:06] commercializable [3:08] products in the market. And this is a [3:12] very different situation that I have [3:14] been in than before. You know, it used [3:18] to be that engineering capacity was the [3:22] scarce resource. That was like I mean [3:24] that's kind of why we exist honestly is [3:26] like engineering capacity used to be the [3:29] really scarce resource. We had way more [3:33] ideas than people and we had way more [3:36] demand than our ability to deliver [3:39] against that demand. And the product [3:42] manager, the products leaders job was to [3:46] prioritize all those ideas, sequence [3:49] them, and then spend like all our time [3:52] in meetings and in Slack and in [3:55] spreadsheets saying no. Saying no. PMs [3:59] should say no more. we should have cut [4:01] lines. It was like very aggressive like [4:03] here's my cut line for the quarter. This [4:05] is my priority list. This is my stack [4:06] rank. And while the PM's job was to say [4:09] no, it felt like engineering's job was [4:12] to say like no, not like that. Not with [4:15] all those features, not with perfect [4:17] architecture. We can only do this much. [4:20] And then it seemed like design's job was [4:22] say to like no, wait for us, like [4:23] please. And so everybody's like just [4:27] constrained on this like precious [4:28] engineering capacity. And so things got [4:31] narrow narrow narrowed down till you [4:33] only shipped a little bit of your road [4:35] map. And now I feel like I have more [4:37] [clears throat] execution capacity than [4:40] like true conviction about what to [4:43] build. So my bottleneck has moved from [4:48] like building what what can I build to [4:52] believing like what do I believe is [4:55] actually worth building. And look, I am [4:59] like the biggest token maxer on the [5:01] planet. I'm like PR's up 3x. Like do it [5:03] all. Ship, ship, ship. Agents [5:06] everywhere. I don't need a PM. I don't [5:07] need any specs. But like I think this [5:10] gap between our building capacity and [5:12] like true in our heart of hearts [5:16] conviction like does any of this code [5:18] matter is a much bigger problem than any [5:22] of us are willing to admit. I think [5:24] we're all getting a lot of pressure from [5:26] the board, from the timeline, from each [5:29] other, from our teams, from our bosses [5:32] to move faster, to get leverage with AI, [5:34] to like really lean in and show we're AI [5:37] native. We can build in this new way. [5:38] Claire said product management is dead. [5:40] She showed all these scary circles. She [5:42] said like ship to the moon. Do it, do [5:43] it, do it. But like I do think none of [5:47] us are I know this is a question I asked [5:48] one who's who's shipping more than ever [5:52] in the last year at their companies [5:54] whose revenue is going up proportionate [5:56] with the amount of PRs [5:58] like that's the fundamental problem [6:00] we're facing here and so I'm going to [6:03] tell a story and like bring this home [6:05] for you this like I have no good ideas [6:07] and I don't know if they matter. Um so I [6:09] had this idea for this thing called the [6:10] product graph for chat PRD. It's like a [6:13] It's a real product for product managers [6:15] idea and it's not special because [6:16] everybody's done it. But it's like we're [6:18] going to suck in everything a company [6:19] knows about customers, about what you're [6:21] working on, blah blah blah blah blah. [6:23] We're going to parse it. We're going to [6:24] like slap, you know, opus on it and then [6:27] it's all going to be available through [6:29] agents and it's going to be awesome. [6:31] It's going to make you be able to make [6:32] these decisions better and build better [6:33] products and of course write better [6:35] PRDs. And so I built it. I built this [6:38] like insights engine. I built this like [6:40] very fancy semantic product graph. It [6:42] like generates this auto wiki. Agents [6:45] can consume it. It was like almost a [6:48] oneshot. Not quite, but it was a pretty [6:50] close oneshot. It worked. It looks [6:53] pretty good. It was like feature [6:55] forfeature matching competitors in the [6:58] market. And I kind of just like every [7:01] time I would build it, I would be like, [7:03] "This just belongs in the trash. It [7:06] belongs in the trash. All this work, all [7:07] this amazing product belongs in the [7:09] trash." And why? Because to me it was [7:12] competitive, it wasn't differentiated. [7:15] And so, you know, even though I had [7:17] built this thing that like obviously had [7:19] a market, people would probably want it, [7:21] customers said it was interesting, like [7:23] in my core, I felt like it wasn't worth [7:26] my customers attention. I like the ROI [7:29] was really shaky to me. I wasn't [7:31] convinced on the interface. I was like, [7:32] should this it it still had a web [7:34] interface. I was like, should it be [7:35] agent first? Is this the right thing? I [7:38] wanted to build something surprising [7:40] that competitors couldn't even imagine. [7:43] And my big problem was that these tokens [7:46] and this like road map that I had [7:49] executed my bet, but it didn't really [7:52] test or validate or strengthen my [7:55] conviction. Meanwhile, I'm sitting on [7:57] all this kind of like code to nowhere. I [8:00] kept shipping more than ever. So, [8:03] customer fixes were like automated. They [8:06] would come in, they go to like one of my [8:08] bots, and then they get fixed. Goals [8:10] were like obliterating tech debt. I [8:12] mean, I re rearchite rearchitected chat [8:14] purity 70 times because I could. Um, we [8:17] stopped tracking issues. We're like, we [8:19] don't need issue tracking anymore. We're [8:20] just going to ship PRs. This AI factory [8:23] like manifested and started running [8:27] itself. It did the thing that we all [8:29] want it to do. It's magical. And here's [8:33] the secret. [8:35] I'm not sure any of it mattered. Like [8:38] I'm not yet convinced that any of that [8:41] mattered. And that really got me [8:45] thinking. It got me thinking about what [8:46] is product management? What is the [8:48] purpose of building software? How do we [8:50] make decisions? How do we prioritize? [8:52] How do we allocate all these tokens? [8:55] And I just kept going back to this idea [8:58] of the road map. road maps made sense [9:03] under engineering scarcity. They like [9:06] laid out your strategy. They told you [9:09] what features were you going to build. [9:11] We had this thing called effort which [9:13] was like more than me like voice noting [9:15] you know Devon saying can you please [9:16] build this and we thought a lot about [9:19] sequencing and milestones and MVPs and [9:24] scarcity helped us theoretically let's [9:27] be honest it helped us theoretically [9:29] filter out our bad ideas because the [9:32] bottom of the list just never shipped. [9:35] It never could. There were not enough [9:36] engineers. There was not enough [9:38] conviction across the organization. and [9:40] said this like below the cut that's what [9:42] we called it below the cut line never [9:44] shipped and now we can ship all our bad [9:48] ideas like congratulations [9:51] to us cheap execution I think still [9:56] needs judgment and I think we're seeing [9:59] this moment where the road map plus like [10:01] limitless or feeling like limitless [10:03] execution capacity is creating three [10:06] traps that I want to warn you about the [10:08] first is the backlog track. So this is [10:12] if you have a backlog, AI will build it. [10:15] They will build every single thing. It [10:18] will build every single thing on your [10:20] list. And so I hear a lot about like, [10:23] you know, I we're burned through our [10:24] backlog. It's because all you have to do [10:27] is slashgo build the backlog and it gets [10:29] done. But the problem is clearing out [10:32] requests or ideas does not mean [10:35] meaningful progress against your [10:36] business or even meaningful progress [10:38] against a customer problem. The second [10:41] thing is the parity trap. This is [10:44] something I really felt in that story [10:45] about the product graph. All your [10:48] competitors are copying each other and [10:51] they all are talking to the same [10:54] customers and they're all sucking in the [10:56] same insights and they're all arriving [10:58] at the same obvious conclusion and [11:00] building the same obvious product with [11:03] the same obvious like run the taste [11:05] skill. Run impeccable. Run whatever. [11:07] Make it look good. Get rid of the m [11:09] dashes. Get rid of the borders. Like [11:10] we're all doing Can you tell I took max? [11:13] Yeah. So we're all doing the exact same [11:16] thing. And so this like parody trap [11:18] where we used to have this like really [11:21] interesting dynamic with competitors now [11:23] it feels like it's evening out and again [11:25] like this question about what Moes are [11:27] and what differentiators are is a big [11:29] one in my mind. And then the final kind [11:31] of trap I see is the churn trap which is [11:34] we ship something, we see noise and pick [11:38] up or not because we can always ship [11:40] something else, we abandon it and then [11:43] we never really learn or compound on top [11:46] of what we're shipping. And so all of [11:49] these things from the inside feel very [11:52] productive. Like my backlog's going [11:54] down, my competitive features are going [11:56] up. like we're shipping more than ever, [11:59] but what I think this does is it just [12:02] like accelerates our path to mid. And [12:06] that's the thing I want to warn you [12:08] about. And so this brings me to [12:10] something that I'm really observing, [12:12] which is this concept of road map zero. [12:15] I don't mean like we've gone through the [12:18] road map, although some people truly [12:20] have road map zero, nothing on the road [12:22] map anymore. What I mean is that every [12:26] visible feature becomes both plausible [12:29] and buildable. You don't have like an [12:32] empty backlog. That's not what I'm [12:33] talking about. It's just like everything [12:34] on your backlog you can kind of do. And [12:37] when everything on your backlog you can [12:38] kind of do, what is the point of the [12:41] backlog? I think buildability and effort [12:44] stop being like this meaningful proxy [12:46] for what matters and prioritization [12:49] again kind of like why PMs exist as [12:52] we've practiced it I think stops being [12:55] strategy right like we've been doing [12:56] these like rice prioritizations but when [12:59] like many of those letters mean nothing [13:01] anymore especially effort like why are [13:04] we still pretending like prioritization [13:06] is the right way to think about our [13:09] products and this is why I think you [13:11] know, road maps are over. They're dead. [13:14] It's because I think they're really [13:15] dangerous right now. I think an AI [13:18] factory plus an old road map will get [13:20] you to those three traps at machine [13:23] speed. They will get you there fast [13:25] because this factory really can't [13:27] distinguish a truly consequential bet [13:30] from an idea. And AI will make the [13:34] consequences of this weak judgment show [13:36] up at your front door faster. And so [13:38] your bad ideas will become your problems [13:40] quicker than ever. And so I don't think [13:43] the road map is the right thing. I think [13:44] the right thing is to ask yourself what [13:47] do I believe strongly enough to go out [13:51] and try and prove [13:53] because the remaining constraint is not [13:55] code. It's not building. It's not [13:57] features. It is truth. It is like on the [14:00] ground with our customers truth. AI [14:02] cannot take an untested assumption into [14:05] fact. No matter how many adversarial [14:07] reviews you do, and I know you're doing [14:09] them, you need real customers. You need [14:12] real data. You need repeated tests. And [14:15] then you need a very unique point of [14:17] view and a very high unique quality bar. [14:21] And what AI allows us to do is make [14:25] faster contact with reality, which is [14:28] great. We all want to make faster [14:29] contact with reality. But that means you [14:32] have a higher obligation to encounter [14:35] reality in your minds. You have to both [14:37] get into the market and accept what it [14:39] is telling you. And this is going to [14:41] feel really retro to people like, you [14:43] know, me and they're a little older [14:45] because we've been saying outcomes over [14:46] outputs all the time, but yet this like [14:49] central artifact, the road map still [14:52] lists features and dates. And again, [14:55] engineering scarcity once made that [14:57] workable. But like I think road map zero [15:01] makes those features and dates very [15:02] dangerous for the reasons that I [15:04] outlined. This feature list I say [15:07] becomes ammunition for a very powerful [15:09] slob cannon. And so what I think we need [15:11] to do is move upstream from what we need [15:14] to build to what we actually need to [15:16] prove. And so here's how I would think [15:18] about what we do instead of a road map. [15:21] First we build our convictions. So what [15:24] is the direction that we want to go and [15:26] what do we believe the future looks [15:28] like? Not in 3 months, 6 months, 9 [15:31] months, in in a year, in two years. [15:33] Where do we think all of this is going? [15:35] I do think you need to determine what [15:38] evidence would show you that your [15:40] conviction is true. And you need to [15:41] define that upfront. What would I need [15:43] to see to prove that I'm doing the right [15:45] thing? And what would I need to see to [15:47] stop? You do need a factory. So I do [15:50] love the factory. I'm saying that's [15:51] dangerous, but I love something [15:52] dangerous. You need your factory. You [15:54] need the ability to build quickly to [15:57] intersect reality very very fast. So I'm [16:00] not saying get rid of it. I'm saying put [16:02] it at the right place in the process. [16:05] And then you need to allocate. You need [16:07] to go through that cycle. And when you [16:09] get into reality, you need to allocate [16:12] your investment, your conviction, your [16:14] effort, your tokens in the right [16:16] direction. And what's interesting I [16:18] think most about this new world we're [16:20] about to move into is I think you need [16:22] to have durable convictions but [16:24] disposable features. And I was talking [16:26] to Mara who is going to come on the [16:27] stage later and she's like but Claire [16:30] enterprises love feature road maps and [16:34] I'm going to talk to some folks at the [16:36] end of the day who have changed a lot of [16:38] features around on us a lot of products [16:40] around us and we as consumers at least [16:43] some of us have accepted that level of [16:45] churn. And so I think we're going to get [16:47] into this place where customers, where [16:49] teams are able to shift to, do I bet on [16:52] these convictions? Do I bet on this [16:54] team? Do I bet on this space? Do I bet [16:55] on this vision? And I understand [16:57] features need to come and go. They might [17:00] be unique to me. They might be unique to [17:01] a market. And so what you want to see is [17:04] clear progress against your convictions, [17:06] but zero ego about your solutions. And [17:09] what this would look like, I think [17:11] there's two ways this could go. You [17:12] could be you have to be stubborn here [17:13] with your durable convictions. But [17:15] there's good stubborn and bad stubborn. [17:18] Good stubborn is you like stay with the [17:20] problem. You revise the solution. You [17:23] know, you hold true what's going to [17:25] happen even if the flavor of the week is [17:27] trending. Bad stubborn is because you [17:30] have tokens. You just move the goalpost. [17:31] You're like, "Ah, we're close. We'll [17:32] just ship again and ship again and ship [17:34] again." And so I want you to think about [17:37] what are my durable convictions? Am I [17:39] being good stubborn? Am I being bad [17:40] stubborn? And then how do you actually [17:42] build into your system the ability to [17:44] absorb disposable features and keeping [17:47] rigor around quality and learning? And [17:51] this is really scary, especially when it [17:53] hits customers. But I think we're moving [17:55] to a place where not everything we ship [17:58] is a promise. And I I don't has anybody [18:01] feeling come on. Like has anybody [18:03] shipped some thank you right in the [18:05] middle? Like you ship things and you're [18:07] like this is a hypo this is like truly [18:09] an experiment and I might get it wrong [18:12] or the market might move underneath me [18:13] or the technology might change, you [18:15] know, tremendously. And so I do think we [18:17] have to think about this idea of like [18:18] not every ship is a process, a promise. [18:21] And we used to do these road maps and I [18:23] would like go hand on heart to go to [18:25] market and I'd be like I swear on my [18:27] children this feature will ship with [18:30] these specific things on this specific [18:32] date. You can definitely sell it in the [18:35] contract like you know and make these [18:37] promises. And now I think we have to [18:39] really think how we communicate [18:40] internally, how we communicate with [18:42] customers. And so I think there's going [18:44] to be everything from a probe. This is [18:46] like kind of a bet. We're going to [18:48] explore a little bit. Um to a more [18:52] durable kind of experiment against your [18:54] conviction. I think this is something we [18:56] really believe in. We're going to stick [18:58] with it till we get it right. And then [19:00] there are promises. Promises are I have [19:02] shipped something. Customers can rely on [19:04] it. They can build their customers on [19:05] it. We really believe and we'll keep [19:07] compounding here. And I think you need [19:09] to be honest about what your commitment [19:11] is on any feature. I think that's almost [19:13] the most um helpful lens on the new road [19:15] map is like how strong is our conviction [19:17] here and how durable is this promise [19:20] versus like how hard is it to build and [19:22] what is our estimated impact and while [19:24] we do this I you know again you need to [19:26] remember code is abundant customer trust [19:29] is not. So going back to that product [19:31] graph, that feature that I built over [19:32] and over and over again until I felt [19:34] conviction, it was feature flagged off [19:37] because I just knew that as soon as I [19:39] put this in front of a customer, they [19:41] were going to build on it. They were [19:42] going to use it. And if I didn't have [19:43] conviction that it was right, I was [19:45] really going to burn that customer [19:46] trust. And that's something I thought [19:47] about really deeply as I was rolling out [19:50] this feature. So I do still think you [19:53] need a road map, but I think that [19:54] roadmap needs to be more about ambition. [19:58] I think it needs to be bigger. I don't [20:00] want to see onesie twoosies on your road [20:01] map. I want to see the future you [20:05] believe should exist. I want to see what [20:10] evidence shows that we're making [20:12] progress. What would prove us wrong? And [20:15] then what earns more time tokens and [20:17] customer attention. And I want to go [20:19] back to ambition because I think the [20:22] last 12 to 18 months as you know it was [20:26] like you know the year of our claude and [20:27] we were shipping everything and PMs were [20:29] writing PRs and it was prototypes [20:31] everywhere. I really think we're in a [20:34] velocity game. We're in a inflect PRs [20:37] straight up velocity game. If I can get [20:40] prototypes to customers faster they can [20:42] give me feedback faster. If I can hand [20:44] things to customers or to engineering [20:46] faster they can do PRs faster. my agents [20:48] will take care of tech debt. My agents [20:50] will take care of features. I think [20:51] we're really in the last 12 to 18 months [20:53] building up our muscle for true [20:55] inflected velocity. I don't think that's [20:58] the game next year. I don't think that's [21:00] the game we should be playing for next [21:02] year. I think next year is the ambition [21:05] game. I think you should think like what [21:08] huge swings can I make? What experiments [21:11] can I run in two weeks, three weeks that [21:13] would have taken us a year last time? [21:16] and how do I build sort of this roadmap [21:19] of very very very large investments in [21:23] very very ambitious builds because I [21:26] think that is more important than raw [21:29] feature velocity and I don't think the [21:31] road map as it is serves that goal you [21:35] know I was talking to somebody and again [21:37] going back to OKRs unfortunately not [21:39] dead yet and she was like what should [21:40] our OKRs be relative to our AI [21:43] transformation like PRs revenue per [21:46] headcount like what it should be like [21:48] how many [snorts] huge experiments are [21:50] you running a month like I don't care [21:52] what the experiments are I don't know [21:54] what they should be like how many big [21:56] big swings are you taking a month with [21:59] the presumption that most of them won't [22:00] work out and I think that is just a very [22:03] new way to think about how you build [22:05] your road map and a very new way to [22:08] think about how you do your product [22:10] again it's big on the ambition and very [22:13] fuzzy on the specifics [22:16] So here is my ask slashgift to you all. [22:21] Build your last road mapap. And I do not [22:23] mean build your last plan. Of course [22:26] not. I just mean like no more lists of [22:30] ideas and features in spreadsheets where [22:33] you guess impact and where you like put [22:36] a date on them and then lock arms and [22:39] say we'll never we'll never change [22:41] anything because this is how we work. I [22:43] just think that day is over. Instead, [22:45] what I want you to do is truly raise [22:48] your conviction, raise your ambition, [22:51] come up with big ideas, define what [22:53] success looks like in like a year or [22:56] two. You're going to have to make some [22:57] big guesses because I can't honestly [22:58] tell what's going to happen in a week or [23:00] two. I really think you want to get to a [23:02] place where you have a factory that can [23:04] surprise you. And so, you really want to [23:08] be able to discard software and then [23:11] hold your factory to a high bar. And [23:13] look, you're going to write a lot of [23:15] code and then you're going to trash it. [23:17] Good. Your customers do not need trash [23:19] products. Your AI is going to have [23:22] better ideas than your team. Good. We [23:24] need good ideas. And you need to hold [23:26] that AI to a very, very high bar. And [23:29] you're not going to be able to promise [23:31] your team, your customers, what next [23:33] year looks like. Good. Because it's [23:36] probably going to be better than you can [23:37] imagine right now. So, this is my ask to [23:41] you. write your last road map I promise [23:43] you is extremely fun on this other side. [23:47] Enjoy the rest of your summit. I will [23:49] see you later this afternoon and please [23:51] come say hi. [23:53] [applause]