A16z Partner Elman Says Retention Moats Require Narrow Wedge Over Hype

a16z合伙人Elman称留存护城河需依赖精准切入点而非炒作 消费级AI应用若试图包揽一切往往会失败，唯有通过单一任务取代固化习惯，才有资格进一步拓展业务。 Josh Elman回顾了自己在早期社交网络、苹果及风险投资领域的职业历程，阐述了为何在当今注意力极易获取的时代，留住屏幕饱和的现代用户却比以往任何时候都更加困难。他认为，成功的消费级AI产品必须在开箱即用时解决一项具体任务以取代固有习惯，从而逐步赢得信任并获得扩展至更广泛服务的许可。对话探讨了对话式商务和AI微短剧的未来、2015年以来缺乏新型现象级社交网络的现象，以及多人AI智能体在后台协同以促进更丰富现实人际联系的潜力。Elman还剖析了他的消费级项目评估标准，涵盖可变推理成本、抵御快速复制的防守能力，以及在Twitter超高速增长时期从Evan Williams身上学到的根本性产品经验。 在现代消费级软件格局中，通过病毒式社交视频获取初始注意力已变得十分容易，但留住用户却前所未有地艰难，因为新工具必须直接取代根深蒂固的数字习惯。 (0:00) 主流消费级AI的普及仍处于起步阶段，正从基础搜索查询转向主动型个人智能体以及创作者驱动的互动微短剧。 (5:00) 与早期互联网和云时代的固定资本支出不同，生成式AI的运作依赖于单次查询的可变推理成本，这要求初创企业持续补贴算力，直至不断下降的推理价格与用户日益增长的付费意愿相交汇。 (42:07) 消费级AI的竞争护城河将较少依赖可能在数天内被复制的表层功能，而更多依赖于赢得的信任、长期记忆以及真实独特的产品标识。 (50:00)

Venture capitalist Josh Elman argues that because modern digital consumer attention is saturated, winning consumer AI applications must establish a narrow wedge that directly displaces an entrenched habit, earning user trust and the right to expand before pursuing broad platform ambitions. The Wedge Strategy and Habit Displacement (0:00) Unlike the early smartphone era when software developers filled open canvases, contemporary consumer tech operates in an environment where capturing initial attention is easy through algorithmic feeds, but retention has never been harder. Elman evaluates consumer investments by asking what specific habit a user stopped doing when adopting a new product. Winning consumer applications must deliver immediate, out-of-the-box utility on a single, focused job rather than launching as broad, all-in-one generalist tools. Model Velocity Versus Mainstream Consumer Trust (7:35) While technologists celebrate rapid weekly model drops and build autonomous frameworks like OpenClaw or Hermes, scaling consumer adoption requires a far more deliberate cadence. Elman distinguishes early adopters and workplace users from mainstream consumers, whose interactions remain predominantly basic, such as conversational search and homework help. Because autonomous personal agents require access to financial credentials and private schedules, long-term consumer retention hinges on transparent data alignment and zero-kickback advocacy rather than raw model benchmark speed. Collaborative Agents and Real-World Catalysts (30:00) Consumer social behavior has shifted away from broadcast feeds toward private group messaging, niche hubs, and offline experiences. Elman envisions moving consumer AI beyond isolated, single-player copilots into multiplayer networks where personal agents coordinate with one another behind the scenes. By cross-referencing user schedules and low-friction social cues, these systems can resolve coordination anxiety, automate event bookings, and prompt real-world gatherings—exemplified by products like Partiful—rather than maximizing passive screen engagement. Defensibility Beyond Fast Surface Clones (50:00) With modern generative tooling allowing competitors to clone visible interface features in days, technical output alone offers little defensibility. Sustainable moats instead stem from distinct creator taste, accumulated user context, and product identity. Drawing from his operating experience at Twitter under Evan Williams, Elman highlights how listening closely to user behaviors must be paired with firm design principles, pointing to the introduction of the native retweet as an intentional choice to preserve creator attribution rather than simply copying manual community hacks. A critical challenge remains in navigating the economics of variable inference costs while scaling mainstream consumer adoption, alongside unresolved questions about how effectively AI can transition from single-player text interfaces into trust-heavy, multiplayer autonomous tasks.

投资人 Josh Elman 指出，消费级科技的竞争核心已从争夺流量转向建立长期留存；消费级 AI 的成功不在于全能式功能堆叠，而在于通过极具针对性的切入点（wedge）有效替代既有习惯，并以纯粹的用户信任与后台协同赋能真实的线下生活。 切口策略与习惯替代：消费级留存法则 (0:00) Josh Elman 认为，在数字生活高度饱和的当下，通过社交短视频获取关注极为容易，但让用户持续留存的难度空前上升。一款成功的消费级产品无法一开始就做通用型平台，而必须找到清晰狭窄的切口，在开箱即用的状态下将单点任务做到极致。他以 Discord 为例指出，该平台起初仅解决游戏玩家的实时语音沟通痛点，在彻底赢得用户信任后，才逐步扩展为多元社群。消费级 AI 产品必须直面一个核心评估标准：“当用户开始使用该产品时，他们停止了做什么？”若不能实质性替代既有行为，产品便无法沉淀为持久习惯。 面对科技巨头和竞品在数天内快速抄袭的常态，Elman 强调功能复制无法动摇真正的底层护城河。模仿者只能复制显性的表面功能，却无法复制产品迭代背后的愿景与用户信任。消费级产品的生命力来自于持续累积的历史数据、独特的审美定位以及用户与品牌之间建立的默契，这些非功能性特质构成了抵御快速克隆的壁垒。 信任门槛与搜索演进：零回扣智能导购 (10:00) 虽然大语言模型已深度渗透至硅谷极客与知识工作者的日常工作流程，但 Elman 认为大众消费级 AI 仍处于极早期的查询与润色阶段。当前开发者热衷于每周迭代模型基准，但大众用户建立信任的节奏却相对平缓。他类比 Apple 坚持的年度发布周期，主张向普通消费者推广智能助理必须遵循循序渐进的节奏，因为涉及信用卡凭证、日程安排及个人隐私托管的操作，无法脱离长期的安全感积累。 在搜索与消费领域，传统基于关键词的信息检索正被对话式交互取代。Elman 以自己在 Amazon 和 Google 耗费数小时寻找特定钥匙扣的经历为例，说明传统搜索在面对模糊或多维约束时效率低下，而 ChatGPT 能直接拆解机械结构与耐用性并锁定小众独立店铺。他进一步描绘了理想的 AI 购物顾问：其核心价值在于完全不依赖佣金或分成（zero kickbacks），纯粹站在买家利益与审美偏好一侧提供诚恳反馈。他展示了自身穿着的一件由 AI 依据审美提示词直接生成规格、并在中国定制生产的黑色翻领毛衣，指出消费场景正逐步走向“个性化对话加按需即时制造”。 多人智能体协同：以数字协作赋能现实社交 (30:00) 经历疫情时期的极端屏幕饱和后，消费者行为出现了向现实生活回归的趋势。用户不再渴望无休止的信息流刷屏，而是需要能消除琐碎摩擦、腾出线下时间的自动化工具。Elman 观察到，传统社交网络中的公开广播行为正在瓦解，熟人私密群聊已成为主要的社交纽带。然而，目前的消费级 AI 几乎完全局限于单人操作，缺乏应对多人群体互动的感知与协同机制。 Elman 构想了“多人协同 AI”（Multiplayer AI）的场景：用户个人的智能体在后台相互沟通，代替人类试探周末聚会意向或共同兴趣，从而彻底化解在群聊中发起邀约却遭遇沉默尴尬的社交心理阻力。例如，智能体在识别到多位好友周五夜晚均处在独处空闲状态时，可主动撮合并在附近的桌游咖啡馆预订座位。类似 Partiful 这类旨在将线上筹划转化为线下聚会的工具，印证了下一代社交软件的核心使命在于充当催化剂，引导人们走出数字孤岛并重返真实的实体人际交往。 叙事微短剧与推理成本下的商业变现 (20:00) 在互动娱乐层面，Elman 结合早期投资 Musical.ly 的经验指出，消费者的娱乐需求正从零散的短视频转向具有连贯戏剧张力的短剧（Microdramas）。以往影视创作受制于演员、场地和拍摄预算等物理瓶颈，而生成式 AI 抹平了制作门槛。Ollie Forsyth 以自身利用 Plot Party 制作耐克创始人创业史微短剧为例，说明创作者只需掌控剧本与叙事品味即可输出高水准剧集。Elman 坚信，内容领域不会被冰冷的合成媒体淹没，创作者的个人审美、叙事身份以及与观众的情感连接，依然是抵御同质化内容泛滥的核心护城河。 针对商业变现路径，Elman 梳理了基础设施的演进周期：从 LinkedIn 早期的自建机房机架，到云计算时代的基础设施轻量化，再到如今按调用计费的“可变推理成本”（Variable Inference Costs）。虽然企业软件可以通过公司报销顺利变现，但消费级 AI 必须直面每次推理带来的真金白银消耗。所幸移动端订阅模式已培养了消费者的付费习惯；随着小模型演进和开源技术压缩单位推理成本，Elman 预测未来几年内算力成本下降与消费者付费意愿将达成平衡，支撑订阅、微交易与广告补贴的多元混合盈利模式。 Josh Elman 总结认为，当前技术圈日新月异的模型发布节奏与大众用户缓慢建立信任的心智周期之间存在显著断层。消费级 AI 的胜负不在于能否打造出面面俱到的全能型代理，而在于能否以绝对的用户利益对齐为前提，在单一真实场景中替用户省下时间，并最终将注意力归还给现实世界。

## Background 本期内容来自 New Economies 播客，由 Ollie Forsyth 主持并于 2026 年 9 月 19 日在 RSS 及各大音频平台发布。节目邀请了 a16z 消费级合伙人 Josh Elman，共同探讨下一波消费级 AI 浪潮，以及何种产品能真正从“尝鲜试用”沉淀为大众日常生活不可或缺的一部分。 ## TL;DR Josh Elman 指出，如今通过推荐算法获取注意力前所未有地容易，但让用户建立长期留存却极为艰难，消费级 AI 必须通过极窄且锋利的切入点极致做好单一功能，进而赢得拓展边界的权利。对话深入剖析了 AI 时代从传统搜索到专属导购、微短剧等互动娱乐的范式转移，并探讨了如何通过幕后多 Agent 协作将人们拉回线下真实社交，而非沉溺于屏幕孤立。同时，Josh 结合其在 Twitter、LinkedIn 和 Apple 的操盘经验，强调信任是产品破圈的终极护城河，随着推理成本可变性上升，唯有兼具鲜明个性与深度替代价值的原生产品才能在抄袭浪潮中立足。 ## Atmosphere & style 整场对话沉稳睿智且充满行业一线的掌故温度，兼具资深操盘手的实战手感与顶级投资人的宏观前瞻。主持人 Ollie 提问敏锐直接，Josh Elman 则以平和真诚的口吻娓娓道来，从当年全员人工搬运 LinkedIn 机架的艰辛，到 Twitter 顶住声浪推行转推机制的博弈，叙事层层递进。两人在观点交锋中既有对极客圈激进迭代的审慎克制，又饱含对科技最终能促成更多真实人性连接的温情与乐观。 ## Bulletin - `00:00` 消费级 AI 的核心挑战已从“获取注意力”转变为“促成留存”，新产品必须通过极窄且锋利的切入点极致解决单一任务，进而替代用户的既有数字习惯。 - `05:00` 当前社会对 AI 的采用呈现出三级分化：硅谷极客依靠 Agent 集群全面接管生活与代码，知识工作者深度依赖生产力插件，而大众消费者仍仅将其视作增强版搜索引擎。 - `08:00` 底层基础模型在长程任务执行与自我改进闭环上的突破推动了 Agentic 范式跃迁，但要跨越到数亿大众用户，严苛的数据隐私与信任机制是不可逾越的护城河。 - `42:15` AI 时代的推理成本重新引入了随用量波动的可变成本，但得益于过去移动应用订阅与内购的普及，消费者对“创造真实价值后付费”的接受度已远超以往。 ## Key takeaways - `35:10` 消费级产品必须遵循“赢得做更多事情的权利”哲学：切忌一开始包揽所有功能，而应先将单一核心任务做到极致，逐步建立用户信任后再拓展边界。 - `48:04` 当下获取用户眼球因算法分发变得前所未有地容易，但由于用户日常生活已被现有数字工具填满，改写旧习惯并建立长期留存变得极其艰难。 - `12:35` 未来电商将从关键词搜索转向对话式专属导购，并进一步演化为由用户意图直接驱动供应链制造的端到端定制生成模式。 - `18:22` Whatnot 等直播电商在欧美破局的核心在于“深度参与感”与“叙事化沟通”，消费者购买的不仅是功能性商品，更是背后的故事与情感连接。 - `20:00` AI 微短剧正在重塑娱乐产业，生成式工具将制作门槛降至接近于零，行业的竞争壁垒随之完全转移至创作者的品味、叙事共鸣以及自有 IP 资产沉淀。 - `25:00` 过去数年缺乏新型超级社交网络，是因为现有平台已完全饱和了纯线上围观动态的需求；未来的社交机会聚焦于促成线下真实聚会（IRL）与多人协同共创体验。 - `30:00` 社交 Agent 的突破口在于从单机模式走向多玩家群聊协作：让不同用户的个人 Agent 在后台互相协商沟通，消除发起聚会的社交脆弱感，促成线下面对面连接。 - `37:34` 语音交互是绝佳的“大声思考”与即兴对练工具，但因其表达天然带有冗余，未来产品必须能够根据使用场景在语音、文字和多模态之间丝滑自如切换。 - `50:00` 初创企业面临的快速像素级抄袭无法阻断具备愿景的原创者；追赶者只能复刻表层形态，却无法复刻产品演进方向、独特 AI 个性与深层信任羁绊。 - `52:00` 旧金山湾区已不再是消费创意的唯一全球诞生地，但它依然是汇聚顶尖操盘手、将早期火苗快速规模化并构建庞大网络效应的最关键枢纽。 - `54:59` Twitter 早期原生转推（Retweet）功能的迭代印证了产品设计的核心真谛：在充分倾听社区诉求的同时，仍需坚守自身理念并向用户透彻传达决策逻辑。 ## Detailed walkthrough ### `00:00` 消费级产品的立足法则与重返投资界 在注意力经济泛滥的今天，消费级 AI 产品面临的本质难题并非吸引眼球，而是留住用户。如今获取消费者的注意力比以往任何时候都简单，但真正的挑战在于让用户不仅愿意尝试，还能真正留存下来。一款产品必须在开箱即用的状态下，把单一核心功能做到极致，迫使消费者要么放弃原有的旧习惯，要么愿意将更多日常行为托付给它。由于每个人的数字生活早被各类应用塞满，新产品必须在用户现有的数字疆界中找到切入点（wedge），通过演进为不可或缺的日常习惯来建立黏性，进而裂变吸纳更多人群。 ```mermaid flowchart LR accTitle: 消费级产品的增长飞轮 accDescr: 展示产品从切入点到形成日常习惯并最终拓展能力的递进过程。 subgraph S1["切入与验证"] A["精准切入口 Wedge"] --> B["极致完成单一任务"] end subgraph S2["留存与习惯"] B --> C["替代旧习惯 / 托管行为"] C --> D["建立高频日常习惯"] end subgraph S3["拓展与规模化"] D --> E["赢得拓展更多业务的权利"] end ``` 播客主持人 Ollie Forsyth 介绍了本期嘉宾——a16z 合伙人 Josh Elman。此前 Anisha Acharya、Ali Yahya 和 David Haber 等合伙人均已做客节目，这次围绕消费级技术的新浪潮展开深度探讨。Josh 拥有跨越投资人、创业者以及顶级科技公司高管（Operator）的复合履历。回顾自己的职业轨迹，Josh 提到他从 1990 年代末互联网破晓期入行，始终为“打造能够改变世界、触达数亿人并改善生活的产品”而着迷。在经历过早期 LinkedIn、Facebook 和 Twitter 的关键增长期后，他在 2011 年转型成为风险投资人，押注了移动互联网与社交网络的黄金浪潮。约八年前，他选择重回一线参与建设，先后加盟 Robinhood 与 Apple，在 Apple 期间主导推进了新一代 Siri（Siri AI）的研发。 促使 Josh 离开 Apple 重回 a16z 担任投资人的转折点，源于他对 AI 技术演进的判断：行业正站在消费级 AI 体验寒武纪大爆发（Cambrian explosion）的起点。他深感自己的职业生涯还有最后一段宏大旅程，那就是重回创业者身旁，帮助他们创立划时代的伟大企业，而消费级技术领域的最佳平台正是 a16z。 > “It has to be such a good product doing that one thing really well out of the box that I either stop doing something else or I turn more behaviors over to it.” —— Josh Elman - **Josh Elman 职业生涯关键节点**： - 1990 年代末：投身早期互联网产品研发； - 2000 年代中后期：先后在 LinkedIn、Facebook、Twitter 等社交网络巨头负责核心产品与增长； - 2011 年：转入 VC 领域，投资移动互联网与社交浪潮； - 近八年：重回一线操盘，先后加入 Robinhood 与 Apple（主导新一代 Siri AI 体验研发）； - 2024 年至今：重返投资界，加盟 Andreessen Horowitz（a16z）担任消费级方向合伙人。 --- ### `05:00` 疫情后消费行为重构与 AI 采用的三级分化 对比五六年前疫情初期与当下的消费级体验，用户的线上习惯发生了深刻变化。在疫情封锁最严峻的阶段，人类探寻了数字在线生活的承载极限：视频通话从一种略显突兀的提议变成了主流沟通方式；社交游戏与手机游戏爆发式增长，Roblox 借此完成了跨越式发展。然而，纯线上体验最终触碰到了心理与生理的天花板，人们意识到现实物理世界无法完全被屏幕替代。当疫情缓和后，大众心态走向再平衡：既保留了高效率的在线工具，又迫切渴望重返现实世界，更具临场感地去体验生活。而这正是新一代消费级 AI 的核心使命所在——充当个人助理与生产力外脑，帮人类搞定日常繁杂事务，从而把时间真正归还给用户去享受现实生活。 对于“AI 究竟是已经走得很深还是仅仅浮于表面”的争论，Josh 认为当下社会可以分为三个截然不同的采用梯队（Buckets）： | 梯队人群 | 核心代表工具与技术 | 当前应用场景与渗透程度 | |:--- |:--- |:--- | | **硅谷极客与早期极客**<br>（Living at the edge） | OpenClaw、Hermes agents、自主 Agent 集群 | 全面托管个人生活与代码工作；开发者不再逐行手动写代码，而是调度多 Agent 集群并行执行复杂任务。 | | **职场知识工作者**<br>（The Workforce） | AI 编程插件、文档分析引擎、智能表格工具 | 深度渗透生产力环境；辅助构建电子表格、分析冗长材料、进行前沿调研，全面改变日常办公模式。 | | **大众普通消费者**<br>（General Consumers） | ChatGPT、Google Gemini 等聊天/搜索工具 | 处于极早期阶段；多数人仅将其当成增强版搜索引擎（省去逐个点开链接的麻烦）、作业辅导辅助、起草困难邮件或应对基础法律/理财咨询。 | 在大众消费者的认知中，AI 远未像当年的智能手机那样彻底重构他们的生活形态或沉浸体验。而在娱乐领域，虽然用户追求好故事的核心本质未变，但生成式 AI 工具如何介入剧本创作、影视制作以及传统叙事机制，目前仍充满摩擦与博弈。 > “For early adopters, what people are exploring and seeing in Silicon Valley, we've got people who are truly living at the edge. They have AI managing their lives, managing their work.” —— Josh Elman --- ### `08:00` Agentic 范式跃迁与大众接纳的信任壁垒 Ollie 提到 ChatGPT 周活跃用户已突破 9 亿大关，大众日常用它查食谱、规划周末出行或处理会议转录，AI 似乎在极短时间内完成了跨越。然而，在科技圈极客与大众之间，存在着巨大的感知割裂：科技前沿以“周”为单位甚至以“天”为单位发布新模型和 Agentic 工具，而大众产品（如 Apple 硬件与 iOS）往往遵循一年一次的稳定更新节奏。 Josh 分析指出，真正的底层拐点发生在去年年底至今年初：底层基础模型取得了质的飞跃，首次具备了处理“长程任务”（long-running tasks）与“自我改进闭环”（loops of improvement）的能力，Agentic（代理式）工作流由此真正跑通。例如，个人开发者独自在公寓里就能基于全新模型能力构建出 OpenClaw 这类工具，通过“技能文件（skills files）”让模型拥有长期记忆并调度工具链，彻底刷新了极客圈对 AI 交互形态的想象。 然而，将前沿突破转化为真正的大众消费级产品，无法靠激进的技术崇拜一蹴而就。极客们习惯于每周换一套技术栈、追逐最新刷榜的开源模型与 Agent 应用；但大众消费者需要的是审慎、确定性强的平滑演进。让 AI 真正深入大众生活，必须直面“信任”（Trust）与“隐私”（Privacy）两大生死关口：当要求普通人交出个人信用卡号、完全开放个人生活隐私数据，并授权 AI 代替自己进行真实消费和决策时，用户不可能每隔两个月推翻一次认知。技术界需要以更沉稳的步伐，带领大众理解让代理人（Agent）代为行事的深层边界。 > “To really bring this to consumers, it's going to take a more measured pace. We have to bring people step by step into understanding what does it mean to have AI doing things for you? What does trust mean here?” —— Josh Elman - **模型能力质变的技术拐点指标**： - 支持复杂、多步骤的长周期任务（Long-running tasks）； - 具备执行中自我纠错与反馈迭代闭环（Loops of improvement）； - Agentic 任务与自主编码能力跨过实用门槛； - 记忆机制工程化落地（如结合 skills files 组织外部工具与长期记忆）。 ### `10:00` 信任是突破大众市场的护城河与搜索交互的范式转移 Josh Elman 指出，对于那些日常生活和工作尚未被 AI 深刻重塑的普通大众而言，他们接受新技术的节奏并不如从业者那么快，因此他正在寻找那些能够带领普通人走上这一旅程的创始人和公司。谈及 Ollie Forsyth 提到的“信任是否是消费级 AI 应用的新护城河”，Josh 坦言在早期采用阶段，实用性（utility）往往会压倒信任，人们愿意为了尝鲜而去测试各种工具。然而，一旦产品想要破圈进入主流视野，触达数亿乃至数十亿用户，信任便会成为决定成败的核心分水岭。这取决于一家公司展现出的严肃态度：用户的数据绝不该被拿去改善模型并与第三方交易，而是“仅用于为你提供更好的体验”。 随后，两人讨论了日常信息检索习惯的剧变。Ollie 直言自己已经很久没有“Google 一下”了，转而使用 Claude、ChatGPT 或 Gemini。Josh 赞同这一观点，并分享了他最近的一个真实购物经历：他想找一款便于在代客泊车（valet parking）时快速拆分、事后又能轻松合上的钥匙扣。他在 Google 和 Amazon 上搜索时十分痛苦，输入关键词后很难精准表达需求，不得不挨个点击链接自行筛选。于是他打开 ChatGPT 展开对话，AI 不仅为他详细梳理了各种连接配件的优缺点、哪些容易损坏、哪些常年耐用，还在他提出“想要美国本土制造”后，精准定位到了密歇根州一家店铺生产的产品。不过 Josh 也强调，即便体验比传统搜索好得多，产品化层面仍有很大提升空间：他在对话末尾其实仍然渴望一个“浏览”（browse）的环节，但 ChatGPT 最终只收敛给出了一个单一选项，而他希望在明确自身所有复杂限定条件后，能在一组高度匹配的完美候选集里进行沉浸式挑选。 “在采用初期，实用性往往胜过信任。人们会尝试新事物，会去折腾试验。但如果你真正想打入主流市场，触达数亿乃至数十亿人，信任最终会变得至高无上。”——Josh Elman --- ### `12:35` AI 赋能的未来电商：从专属导购到端到端定制生成 在谈及尚未出现的下一代线上购物体验时，Josh 纠正了主持人的口误，说明自己在 Apple 工作了将近六年半（而非十年）。他认为实体商业中最顶级的体验莫过于店铺里的专属导购（concierge）服务：一位真正的专家在身边提供建议，帮你挑选试穿，而最关键的是“他们的唯一动力是你的成功”，既不是为了让你掏空钱包，也不是为了拿特定商品的回扣。Josh 期待未来的 AI 消费级产品能扮演这一站在用户这边的角色，在浩瀚如烟的商品海洋中做减法，让用户完全建立起信任。用户可以随手拍下自己现有的衣物、提出升级需求，或者试穿后拍照让 AI 提供穿搭点评。 为了证明这一趋势的现实可行性，Josh 透露他当天身上穿的黑色翻领毛衣（polo sweater）就是完全借助 AI 设计的。他当时只是向工具描述了自己想要一件什么样的黑色 Polo 毛衣，AI 协助敲定了具体版式设计，并直接对接了中国工厂完成生产制造交付。从单纯的关键词搜索，到自然语言导购，再到按需反向定制，AI 正在将消费体验推向全新维度。 ```mermaid flowchart LR accTitle: AI购物体验演进路径 accDescr: 展示从传统搜索到对话式导购，再到生成式反向定制的购物演进流程 subgraph Search["传统搜索模式"] A["输入关键词"] --> B["海量链接与比价"] B --> C["用户自行人工筛选"] end subgraph Agent["AI 专属导购"] D["自然语言意图表达"] --> E["多轮交互/限制条件过滤"] E --> F["高度匹配的精选候选集"] end subgraph C2M["生成式定制制造"] G["自然语言构想设计"] --> H["AI 辅助打板出图"] H --> I["柔性供应链直接制造交付"] end Search -.->|"范式升级"| Agent Agent -.->|"深度进化"| C2M ``` “我身上穿的这件衣服，实际上完全是用 AI 设计的；我说我想要一件新的黑色 Polo 毛衣，它帮我搞定了设计，并且真正在中国帮我代工生产了出来。”——Josh Elman --- ### `14:45` 消费级 AI 的三大核心赛道：个人 Agent、交互式娱乐与垂类服务 展望未来 12 个月乃至更长周期的消费级初创生态，Josh 梳理了他最为看好的几个关键方向： 1. **全天候私人生活 Agent（Personal Agents）**：区别于市面上铺天盖地的企业办公提效 Agent，这类产品专注服务于个人生活。它可以在周末根据天气或本地节日帮用户规划出行、管理健康与财务状况、提醒日程，甚至只是在傍晚温和地问候一句“今天过得怎么样”。如果做得很烂，这种产品会变成恼人的弹窗催促（naggy）；但如果做对了，它就像身边时刻为你着想的伙伴。 2. **深度交互式娱乐与世界构建（Future of Entertainment）**：现在的影视娱乐多为被动接收。Josh 以 Apple TV+ 的科幻剧集《为全人类》（*For All Mankind*）为例，观众常常渴望深入探索某个支线宇宙，或者推演“如果当时发生了另一件事会怎样”。AI 赋予了每个人生成交互式体验与同人衍生世界的能力。他提到 Roblox 构建的多人互动生态极为成功，但受众相对集中；未来基于生成式 AI 的新型体验空间将比 Roblox 广阔得多，任何人都能成为世界构建者。 3. **垂直生活服务与人际连接（Vertical Services & Connection）**：在旅行等低频高客单领域，用户花费了辛辛苦苦赚来的时间和金钱，却往往因为信息不透明而无法实现个性化体验，垂直 AI 专家能填补这一鸿沟。此外，在社交领域，AI 可以作为人际沟通的“练兵场”——用户可以先和 AI 进行角色扮演试探对话，甚至让 Agent 代为收集前置信息，最终促成更高质量、更具深度的现实人际互动，而不是用机器去取代真实的人类联系。 “我不想去取代人与人之间的连接，但只要它能让我更进一步、去拥有更优质且更有影响力的人性时刻，那就会无比美妙。”——Josh Elman --- ### `18:22` 直播电商与社区参与感：Whatnot 的崛起逻辑 Ollie 谈及欧美直播购物（live shopping）赛道时表示，这一模式在亚洲和中国已被验证极其成功，但欧美市场此前一直反响平平，直到 a16z 投资的 Whatnot 迅速崛起并逆势完成巨额融资，令人刮目相看。 Josh 深度剖析了 Whatnot 破局的核心机制：传统电商本质上是冰冷、静态的货架展示，用户只能点击图片和购买按钮；而 Whatnot 代表的直播电商核心在于“深度参与感”（participation）。在平台上，用户不仅仅是买一件功能性的工具，更是在聆听商品背后的故事。无论是听独立设计师拆解一件服装的灵感，还是听专业买手分享其精挑细选的审美逻辑，买家都可以在公屏实时提问、与主播互动交流。这种富有深度的故事化沟通，让用户在点击下单时拥有了更强的信任感、归属感以及情感共鸣，这也正是 Whatnot 能够在欧美消费市场持续扩张并拉开身位的关键所在。 “当我们真正在 Whatnot 上时，我们听到的是产品背后的故事……我们买的不再是一个在生活中起某种作用的扁平物品，我们买下的是一段完整的故事，以及一件新物件走进我们生活的方式。”——Josh Elman ### `20:00` 直播带货、微短剧与 AI 驱动的内容创作浪潮 对话从直播带货切入。主持人 Ollie Forsyth 与 Josh Elman 探讨直播电商能否真正像在中国那样在西方（如美国）生根发芽。在相关模式成熟的市场，用户展现出了完全不同的购买行为，交易不仅更真实，而且建立在高度的人性化连接之上。 紧接着，话题自然延伸至互动式直播娱乐与当前正在爆发的新趋势——AI 微短剧（Microdramas）。Josh Elman 认为短剧是下一波娱乐演进的核心载体： - 这一演化路径最早可以追溯到 Musical.ly（他在 Greylock 任职期间早期投资的项目，后被字节跳动收购演变成 TikTok）。 - 当时 Musical.ly 的创新在于利用短视频配合背景音乐打造强烈的情感共鸣，算法在此基础上学习用户的共鸣偏好，从而沉淀出极其黏性的内容流。 - 如今用户已经习惯了碎片化的短视频，但往往渴望更深度的叙事与更丰满的情节。微短剧恰好填补了这一空白：在 TikTok 或 YouTube 上，用户先通过几秒的切片被迅速钩住（hooked），接着便迫切想要把整个故事追完。 针对这一趋势，AI 正在彻底打破影视制作的工业门槛。Josh Elman 指出，过去拍摄一部影视作品需要庞大的制作棚、专业演员和昂贵的布景，能够生产的故事数量非常有限。现在有了生成式 AI，那些拥有绝佳构思却缺乏资源的写作者和创作者也能将想法转化为视觉作品。未来不仅会在现有社交平台进行分发，更可能催生出全新的故事交易市场（Marketplaces），成为用户寻求沉浸式叙事内容的首选平台。 > “微短剧是下一波娱乐浪潮中极其重要的一部分。” —— Josh Elman 随后，Ollie Forsyth 分享了一个切身体验：他们刚刚制作完成了自己的第一部 AI 微短剧。 - **创作构想**：Ollie 原本希望能面对面专访全球顶尖的商业创始人（如马云、Phil Knight 等），但这在现实中操作难度极大。于是他们决定将这些伟大公司的起源故事改编成 4 至 5 分钟的微短剧，并融入一些戏剧化与浪漫爱情元素。 - **合作分工**：他们与一家名为 Plot Party 的早期初创公司合作，Ollie 的团队负责叙事和剧本设定，Plot Party 负责利用 AI 工具进行视觉与视频渲染，成片发布后迅速收获了极高播放量。 ```mermaid flowchart LR accTitle: AI 微短剧生产与消费链路 accDescr: 展示从创作者构思、AI 工具制作到社交分发与内容交易平台的完整链路。 subgraph 创作端["创作端 (Creation)"] A["创作者剧本与叙事灵感"] --> B["AI 制作工具 (如 Plot Party)"] end subgraph 分发端["分发与消费端 (Distribution)"] B --> C["社交短视频引流 (TikTok / YouTube)"] B --> D["沉浸式故事交易市场 (新媒体渠道)"] end ``` ### `22:36` 创作的平民化与创作者资产沉淀 针对 Ollie 团队制作的耐克创业短剧，Josh 提到了好莱坞拍摄的耐克传记电影《AIR》（气垫传奇），并指出将企业发展史转化为充满戏剧张力的影视作品本身极其困难。如今借由 AI 工具，任何普通人都可以制作出类似质感的短剧。 但 Josh 强调，当技术把制作门槛降至接近于零时，行业的瓶颈发生了转移： 1. **核心壁垒从制作转向“品味”与“共鸣”**：当人人都能创作时，能够脱颖而出的一定是懂得如何打动观众、积累受众并维系复访关系的创作者。 2. **IP 与产品的自有化**：Ollie 指出，在这个新时代，创作者不仅要讲故事，更应该构建属于自己的、能够拥有所有权的产品与企业。 3. **真实的人性连接不可替代**：即便未来大量内容由 AI 辅助生成，受众买单的底层逻辑仍然是对创作者个人身份（Identity）的认同。Josh 直言，内容消费绝不是由冷冰冰的“博格人 AI”（Borg AI）自说自话，观众追随的是背后有血有肉的讲故事者。 > “在当下，最艰难的环节已经不再是制作本身……能够引发共鸣、找到观众、并与观众建立起让他们愿意不断回访的关系，这依然是关键。” —— Josh Elman ### `25:00` 为何过去几年没有诞生新的超级社交网络？ Ollie 抛出了一个核心疑问：过去几年技术飞速演进，AI 浪潮汹涌，但为何市场上迟迟没有出现新的主流社交平台？Josh Elman 凭借在 LinkedIn、Facebook、Twitter/X 的工作经验，以及作为 Discord 早期董事会成员和 Musical.ly 投资人的履历，给出了深度剖析。 Josh 认为最根本的原因在于：“在某种程度上，我们已经不再需要新的社交网络了。” - **历史周期的对照**：在 2015 年他投资 Discord 和 Musical.ly 时，市场上同样充斥着“已经有了 Facebook、Instagram、WhatsApp、Twitter 和 LinkedIn，为什么还需要新产品”的质疑。但 Discord 解决的是玩家实时语聊和社群协作的真实痛点；Musical.ly 解决的则是大众创作表达的问题。 - **前置摄像头 vs 后置摄像头**：Josh 生动地比喻道，早期的 Instagram 很大程度上代表的是“后置摄像头”视角（看我所在的高级餐厅、看我喝的拉花拿铁，强调生活展示）；而 Musical.ly 则是“前置摄像头”视角（即使背景是卧室里没洗的脏衣服堆，只要配上节拍跳一段舞，就能骄傲地发布创作）。 - **微创新与衍生品的陷阱**：过去几年涌现的多数所谓“新社交产品”本质上只是现有巨头的衍生品（Derivative）。例如各种形式的 Twitter/X 仿品，仅仅换了一批细分人群，在核心交互和网络效应上缺乏质变，因此很难真正走向主流大众。 ### `27:38` 消费习惯的变迁与重返真实世界（IRL） Ollie 提到自己本周刚刚采访了社交网络鼻祖之一 Bebo 的创始人 Michael Birch（这也是 Michael 近十年来首次接受播客采访），两人同样探讨了用户消费习惯的巨变： - 现在的用户几乎不再像过去那样频繁在 Facebook 个人时间线上发帖。 - X 变成了快速获取各类资讯的核心阵地；Substack（同样为 a16z 投资组合公司）承载了深度内容与长文消费。 - 大众对纯线上“围观熟人动态”的欲望正在边际递减。 Josh 对此表示赞同。大众的时间已经被现有成熟产品填满：刷完 Instagram 上朋友的照片、在 Facebook 上看熟人动态、再去 TikTok 刷几段视频，用户的日常社交与娱乐需求基本已被完全满足，不再有“我还想看看朋友说了什么”的饥渴感。 真正的增量机会反而蕴藏在两个方向： 1. **连接线上与线下的现实社交（IRL - In Real Life）**：随着疫情结束，能够促成现实相聚、打破数字屏幕隔阂的产品极具价值。例如 a16z 投资的活动平台 Partiful，它专注于帮助用户更轻松地组织派对、探索线下活动，在年轻群体中建立了极高黏性。 2. **共创与协同体验（Co-experience）**：社交正从单向的“发布-点赞”转变为“共同做一件事”，这也是 Discord 能够持续繁荣的核心密码。 > “做完这一切之后，我不会再去想：‘我的朋友们还有什么话要说？’ 但我确实认为，我们将看到围绕‘人们正在创造什么’以及‘我可以在哪里与他人一起共同体验’而诞生的全新事物。” —— Josh Elman ### `30:00` 从“单机版”到“多玩家”：群聊与社交 AI Agent Josh Elman 回顾了 Discord 的崛起历程：早期的 Discord 创造了一种“一边轻松打游戏一边实时语音通话”的新行为，这种全新的行为模式把人们聚拢在一起，随后用户便留在了这个平台上闲聊和互动。反观当下的 AI 领域，大多数产品依然停留在极度的“单机（Single Player）”模式——用户单独与 AI 交互，让它处理个人生活中的琐事，但 AI 尚未融入群体，也不懂群体交往中的微妙情绪和潜台词。 近年来群聊（Group Chats）已经演变成最重要、最普及的社交形态，甚至构成了人们未曾察觉的新型社交网络。Josh 指出，自己现在的本地群聊数量远超十年前。如果将 AI 引入群聊，或者让多个代表不同个人的 AI Agent 在幕后相互沟通，将开启全新的社交互动范式。Josh 兴奋地描绘这种协作前景：“嘿，我正在和它说话，你也在和它说话，它们之间彼此交流，这样当我们真正坐下来聊天时，大家都掌握了更充分的信息。” ```mermaid flowchart LR accTitle: 多智能体社交撮合流程 accDescr: 展现从个人用户向智能体表达意向，到智能体互通匹配，最终引导至真实世界线下面对面的完整链路。 subgraph UserLayer["用户交互层"] U1["用户 A<br/>（想看电影/想聚会）"] U2["用户 B<br/>（寻找周末活动）"] end subgraph AgentLayer["Agent 协调层"] A1["Agent A<br/>（探查意向与空闲档期）"] A2["Agent B<br/>（同步需求与偏好）"] end subgraph OfflineLayer["真实世界线下撮合"] Match["智能匹配与预约<br/>（例如：Noe Valley 桌游咖啡馆）"] end U1 -->|"私密沟通"| A1 U2 -->|"私密沟通"| A2 A1 <-->|"后台多智能体协商"| A2 A1 --> Match A2 --> Match ``` Ollie Forsyth 随即追问，这种多智能体与人类之间的“多玩家沟通”落到现实中究竟长什么样？难道仅仅是 Josh、Ollie 和一帮共同好友想要在周末碰头、纠结去哪里玩这么简单？Josh 确认这正是最直观且最切中痛点的切入点： - **消除社交脆弱感**：在现实中，人们在群里问一句“周末有人想出来吗”却遭遇冷场（Crickets）是极其尴尬且令人感到脆弱（Vulnerable）的。 - **Agent 幕后探查**：用户只需对自己的专属 Agent 随口提一句“去帮我到处打听打听，看这周末有没有朋友想出来”，Agent 便去和其他朋友的 Agent 交流撮合。 - **促成现实连接的“社交黏合剂”**：有人可能会觉得这听起来有些反乌托邦，但 Josh 认为这恰恰是让 AI 成为人际关系的黏合剂。比如有人随口问“有人看《奥德赛》（Odyssey）了吗？”，Agent 就能在后台找到同样想看该电影的朋友，顺水推舟促成一次聚会。 - **从线上孤立走向线下真实体验**：Josh 描绘了一个理想愿景：周五晚上，旧金山诺伊谷（Noe Valley）可能有 10 个人各自坐在家里和 AI 聊天；AI 识别到这一聚集需求后主动提议：“你们 10 个人都在诺伊谷，为什么不一起去桌游咖啡馆玩呢？他们已经准备好两张桌子了，我现在就可以帮你们预订。”这种将人们从屏幕前拉回真实世界的能力，才是技术最大的价值。 --- ### `33:04` 极致助手设想：生活全能管家与“严父慈母” 顺着这个思路，Ollie 提出如果跳出基础中间层、把设想推向极致（Crazy Big），未来的生活会变成什么样？Josh 勾勒了一个彻底解放个人精力的终极个人助理图景： - **主动式社交与日程编排**：周六早晨醒来，你不再需要焦虑地规划时间，Agent 已经为你排好了丰富多彩的一天——几点和哪位朋友见面、下午留出休息时间、晚上与哪些同样在找活动的朋友聚餐、甚至周日和下周二晚上的安排都井然有序。 - **无缝的生活执行力**：当你随口感叹一句常穿的衬衫磨破了，Agent 已经自动下单了合适的新款并寄送到家。 - **兼顾陪伴、自律与留白**：真正的助理不仅替你包揽不想做的杂事、省出时间享受热爱的事物，还会主动守护你的独处时间（Personal Time）。它既能督促你锻炼、健康饮食，又允许你在适当时机放纵一下。 谈到理想中的交互关系，Josh 幽默地拿自己的母亲做类比。母亲是世界上最懂他的人，有时是最大的拉拉队长（Cheerleader），有时又是最严厉的批评者（Critic），给予他最需要的“严厉的爱（Tough Love）”。Josh 坦言：“我深爱我的母亲，她能给我严厉的爱与鼓励，但我并不会每天把所有事情都跟她聊。”拥有一个具备全天候上下文、真正理解你的长期目标并在耳边适时推你一把或敲打你的 AI 伴侣，将是一种前所未有的体验；甚至如果用户整整一周没跟妈妈通电话，Agent 还会贴心地跳出来提醒该打个电话了。 --- ### `35:10` 平台壁垒与破局：“赢取拓展边界的权利” 面对如此庞大的个人助理愿景，Ollie 提出了关键的商业与分发疑问：这样的超级平台究竟会由谁创造？是 ChatGPT、Claude 这样拥有现成庞大分发网络（Distribution System）的底层模型巨头，还是会有初创公司杀出重围？初创企业在缺乏巨额分发优势的情况下该如何生存？ Josh 提出了核心产品哲学——**“赢得做更多事情的权利（Earn the right to do more）”**： - **切忌包山包海**：要想成为一家巨头，绝不能一开始就试图无所不包，而是必须先挑少数几件事做到极致，从而赢取扩展能力的信任。 - **Discord 模式的启示**：Discord 最早仅仅专注做好一件事——成为游戏玩家在联机时进行低延迟语音通话的工具。正是因为把这一点打透并赢得了信任，玩家才开始在上面聊更多游戏话题，随后聊日常琐事；接着非游戏玩家也因为朋友都在这里而被吸引过来，最终成长为一个综合性的社区平台。 - **通用型助手的信任陷阱**：那些一上来就宣称“我是全能通用助手，能帮你搞定一切”的产品，往往因为缺乏深度，根本无法在任何具体任务上建立信任。 - **Siri 的历史教训**：Josh 结合自己在 Siri 上的长期思考指出，Siri 最初其实成功赢得了帮用户做好几件简单任务的信任，例如打电话、设倒计时、设提醒事项和发短信；但当用户试图让它处理更多复杂任务时，它却频频出错崩溃，直接伸手拒绝并回复一句“抱歉，我帮不了你”。产品能让用户尝试“再多试一件事、再多试五件事”的机会窗口是非常有限的。下一代 AI 助理不仅需要按部就班（Step by step）地拓展边界，还要在无法完成任务时向用户解释清楚原因，展开真正的对话，而不是冷冰冰地筑起高墙。 Josh 对此总结道：“我始终深信，如果你想成为一家极其庞大的公司，你绝不能一开始就试图包揽所有事情，而是必须先把少数几件事做到极致，从而赢得做更多事情的权利。” --- ### `37:34` 语音交互的真正潜力：思维草稿箱与对练伙伴 谈及消费级科技的新规则，Ollie 与 Josh 一致认为语音（Voice）将迎来巨大的爆发浪潮，但目前行业仍处于“仅触及皮毛（Scratching the surface）”的初级阶段。 结合此前在 Siri 的经历与自身日常习惯，Josh 深度剖析了语音作为交流媒介的优势与局限： - **速度与流利度优势**：人们说话的速度通常远快于打字。对于像 Josh 这样习惯“大声思考（Think out loud）”的人来说，不断说话是理清逻辑、完善观点的有效方式。 - **语音的“有损性（Lossy）”**：语音表达往往伴随着漫无边际的闲扯（Ramble）和游离；而文本输入虽然起步慢，但用户在编辑和润色发送给对方的过程中，思维已经被大幅提纯。 - **个人场景下的“思维提炼器”**：工作场景下已经有了出色的会议记录与总结工具；但在个人生活中，当用户脑海中闪过零碎灵感或漫无目的地做白日梦时，语音是门槛最低的记录形态，助手应当能够倾听长篇大论的唠叨，并精准提炼出核心要点。 不过，Josh 明确表达了自己对极端趋势的怀疑：“我目前并不相信我们所有的交互都会彻底转向语音。我认为这始终是一种平衡，纯语音 Agent 的设想在现实中其实站不住脚。”在安静的公共场所、涉及隐私的沟通环境，或者需要提前把观点组织精炼的场景下，文字依然不可或缺。 ```mermaid flowchart TD accTitle: 语音与文字交互场景权衡 accDescr: 对比语音输入与文本输入在不同场景下的优劣势及协同关系。 subgraph VoiceInput["语音交互模式（Voice）"] V1["高输入速度与流利度"] V2["大声思考与发散性漫谈"] V3["车载驾驶等双手受限场景"] V4["播客/演讲等实时角色扮演对练"] end subgraph TextInput["文本交互模式（Text）"] T1["高精度与结构化润色"] T2["安静公共环境与隐私保护"] T3["复杂逻辑精炼表达"] end VoiceInput -->|"AI 提炼与总结"| Hybrid["混合交互体验<br/>（按需切换，互为补充）"] TextInput --> Hybrid ``` 对话最后，两人探讨了全双工模式（Duplex modes）为语音带来的革命性体验： - **即兴对练伙伴（Sparring Partner）**：在参加播客录制或公开演讲之前，具备实时打断与自然流转能力的新技术让 AI 成为极佳的模拟演练对象，能够与用户进行深度角色扮演、推演对话走势，帮助其打磨论点。 - **双手被占用的专注场景**：Josh 分享道，自己在开车时非常喜欢通过语音与 AI 来回切磋讨论，这不仅完全不干扰驾驶专注度，还能高效利用通勤时间理清复杂思路。 ### `40:00` 多模态切换与 Reid AI 的“数字分身”实验 在日常交互中，用户面对不同场景会自发切换输入模式：出门在路上时，语音是最自然的沟通手段；而窝在沙发上时，往往并不想开口对手机或电脑说话，打字、阅读和划动屏幕依然是最舒适的交互方式。因此，优秀的产品必须能够极其丝滑地在多种模态（语音、文字、视觉）之间来回切换。 Ollie Forsyth 提到，语音工具在演讲备稿或模拟思想碰撞（sparring partner）方面展现出惊人的潜力，例如用 ChatGPT 梳理思路效果就非常出色。随后他引申到 Josh Elman 曾在 Greylock 与 Reid Hoffman 紧密共事的经历，提到如果能有一个 Reid 的虚拟模型做预演互动，将会非常酷炫。Josh Elman 对此深表认同，并盛赞道：“Reid's one of the most incredible founders and thought leaders of our time.” 他透露 Reid 正在构建名为“Reid AI”的专属项目——这不仅仅是将通用 ChatGPT 当作陪练，而是把 Reid 的个人思想、文字与知识全部喂给模型，创造出一个数字化身。尽管数字分身依然无法替代与真人的即时交流，但它却是一个极佳的代理工具，能让人快速摸索他的思考逻辑和应对特定问题的切入点。在真正面对 Reid 之前，通过与他的数字分身切磋，访客往往能做好准备，从而促成更深入、更有成效的现实对话。 ### `42:15` 从固定机架到可变推理：C 端 AI 产品的商业化与成本十字路口 针对当前许多 AI 产品通过彻底免费或极低价格来疯狂获取流量与分发的现象，Ollie 质疑这种模式能否持续，以及它们何时会面临向投资人交差的变现压力。Josh Elman 回溯了互联网基础设施的成本演进史：在他职业生涯初期，建网站是一笔沉重的固定成本投入。他回忆起加入 LinkedIn 的第一周，公司仅仅为了把物理服务器从一组旧机架搬到数据中心更大的新机架上，就需要全公司大约 15 个人全员出动去人工搬运机器。随后，AWS 等云计算基础设施的成熟消除了这些重资产前期成本，使得规模化服务海量 C 端用户变得极度廉价，产品本身的创意与设计成了唯一的核心。 ```mermaid flowchart TD accTitle: 互联网产品成本与变现演进图 accDescr: 展示从早期自建机架、云原生时代到 AI 推理时代的成本结构与商业变现模式的转变 subgraph Early["早期互联网时代 (LinkedIn初期)"] A["物理机房与服务器机架"] -->|"前期固定成本极高<br/>需要重度融资建机房"| B["高门槛服务交付"] end subgraph Cloud["云计算时代 (AWS成熟期)"] C["按需租用云算力"] -->|"基础设施极度廉价<br/>边际服务成本趋近于0"| D["免费模式爆发 (Freemium)"] end subgraph AI["现代 AI 时代 (当前)"] E["模型 API 与 Token 消耗"] -->|"重现可变成本 (推理成本)<br/>每次互动均产生成本"| F["新型混合变现模式"] F --> G["应用商店订阅制"] F --> H["开源模型降本"] F --> I["广告变现交叉补贴"] end ``` 然而，AI 时代的来临重新引入了成本难题，而且是随着使用量线性波动的可变成本——推理成本（inference costs）。B 端软件可以依托产品驱动增长（PLG）让员工通过企业报销甚至签署大额合约来买单，但 C 端产品往往无法要求用户在前置环节就掏腰包，团队必须探索如何去补贴这笔推理账单。不过，得益于过去移动游戏与应用订阅制（App Store、Google Play）的普及，如今消费者对数字商品付费的接受度远高于十到二十年前。Josh Elman 强调：“When people hit a certain level of engagement with a product, asking them to pay isn't the same hand to the face that it used to be. It's now very much an invitation: if this is finding you value, it's costing us money.” 随着开源模型普及以及推理单价的急剧下降，未来几年技术降本的曲线终将与用户愿意付费的心理阈值相交，形成广告、内购与订阅协同支撑推理开销的健康商业模型。 - **早年重资产案例**：Josh 加入 LinkedIn 的第一周，全公司仅约 15 名员工，当时大家需要一起去数据中心人工搬运物理服务器机架。 - **成本性质转变**：从云原生时代的极低边际固定成本，转变为 AI 时代随每次 Prompt/Token 生成即时产生的可变推理成本（Variable Inference Costs）。 - **用户付费心智**：经过应用商店订阅生态与移动游戏的十余年教育，向深度高频用户提出付费请求已不再被视作冒犯，而是一次合理的价值置换邀请。 ### `45:26` 消费级产品的黄金法则：好奇、替代、传播与留存闭环 当被问及作为投资者在审视成千上万家初创企业时，内心是否有评判优秀 C 端产品的“黄金清单”时，Josh Elman 拆解了他最为看重的完整生命周期闭环： 1. **引发惊叹（Raise their eyebrows）**：产品必须具备极其清晰明确的价值主张，一句话就能向别人解释清楚它实现了以往无法想象的功能。在 AI 时代，做到让用户挑眉惊叹本身并不罕见。 2. **行为替代（Behavior substitution）**：当用户试用产品并发出“哇”的感叹后，核心在于它如何融入日常生活，并具体取代了什么旧行为。Josh Elman 指出他最爱问的问题：“I always like to ask people, what did you stop doing when you started using this new thing?” 如果能得到切中要害的答案，说明该产品真正形成了有效替代。 3. **自发传播（Organic evangelism）**：用户是否兴奋到会主动向他人安利？正如 ChatGPT 早期那样，大家因为惊艳于“再也不用去 Google 翻网页了，这里直接给你答案”而引发野火般的口口相传。 4. **终极留存与网络效应（Long-term retention）**：留存是检验一切的核心。最坚固的留存壁垒来自双边或多边网络效应（让人们聚在一起使用）；次之则是基于个人数据沉淀的转移成本——用户在产品中投入的信息和使用历史越厚，就越不愿迁移到别处；在内容与消费生态中，优秀的创作者为了受众而聚集，受众又因创作者而留下，从而形成正向飞轮。 ### `48:04` 争夺数字生活“微小切口”：注意力容易，建立留存极难 针对 Ollie 提出“如今争夺用户注意力是否比以往任何时候都更难”的观点，Josh Elman 给出了截然相反的反直觉判断。他认为，得益于 TikTok、Instagram Reels 和 YouTube 等庞大算法推荐渠道的存在，新鲜好玩的东西极易实现病毒式传播，“在当下，获取消费者的眼球反而是前所未有地容易”。 真正难如登天的是将尝鲜者转变为长期留存者。Josh 直言，消费者的现实生活和数字生活早已被各种现有工具塞得满满当当，COVID 疫情更是将这种数字饱和度推向了极限。15 年前智能手机刚普及的时代，创业者是在一片空白的原野上做“增量拓展”，大量生活行为首次被搬上移动端；而今天的数字产品，必须在拥挤的现有习惯中找到一个极其锐利的“微小切口”（wedge），并且做得足够出类拔萃，才能实现“残酷替代”。对于新一代消费级 AI 产品而言，唯有展现出数量级维度的体验优势，才有机会改写用户根深蒂固的行为范式并真正扎下根来。 ### `50:00` 抄袭者的困境与 AI 产品的核心“灵魂” 面对当今市场上大公司与初创企业对新兴创意的极速跟风与像素级抄袭，Josh Elman 指出，克隆一个产品的时间周期已经从过去的数月、数周急剧缩短到了现在的区区数天。然而，这种抄袭并不能真正杀死具备核心洞察力的原创者。他强调，追赶者所能复制的仅仅是当前浮于表面的形态，却永远无法复刻创始人的前瞻视野与演进路径：“即便其他人能抄袭你已经做出来的东西，他们也无法抄袭你未来的走向。”（“Even if other people can copy what you did, they can't copy where you're going.”）对于当下的创始人而言，在激烈的红海竞争中坚守自己的路线并践行最初的愿景，正是构筑护城河最有力的方式。 随着时间推移，原创产品与用户之间会沉淀出一种深刻的信任与情感连接。用户即便尝试了竞品的仿冒版，也会敏锐地察觉到那些替代品“缺少了某种真正特别的精髓”，从而迅速回流到具备独特性格的原生产品中。Josh 认为，在当前的 AI 浪潮下，打造产品绝不仅仅是实现特定功能那么简单，它必须拥有鲜明的个性、能够与用户建立长久羁绊的身份认同，以及在交互过程中持续学习并演进出懂你的独特视角。正是这些不可替代的软性维度，让那些能够经受住考验的持久平台在喧嚣之中脱颖而出。 ```mermaid flowchart LR accTitle: 复制品与持久产品的竞争差异 accDescr: 对比浅层抄袭与具备愿景、个性与持久网络效应的原创产品路径 subgraph Copycat["跟风与克隆者"] A["逆向已有功能"] --> B["表面形态复刻"] B --> C["缺乏产品灵魂与演进路线"] end subgraph Durable["持久原生平台"] D["独特洞察与愿景"] --> E["塑造 AI 个性与身份认同"] E --> F["伴随用户持续学习演化"] F --> G["建立深度情感羁绊与网络效应"] end C -.->|"用户流失回流"| G ``` ### `52:00` 旧金山不再是创业起点，而是规模扩张的关键枢纽 针对 Ollie Forsyth 提出的疑问——为何过去半到一年内涌现的优质消费级与社交 App 几乎全被美国团队占据、欧洲是否在消费社交领域掉队，Josh 提出了不同看法。他列举了瑞典的 Lovable、起源并不在旧金山的 ElevenLabs，以及发源于中国团队的 Musical.ly（后发展为 TikTok 的基石）和中国蓬勃发展的直播电商生态，指出创新早已在全球各地开花结果，绝非硅谷所独占。然而，旧金山湾区依然保留着它独一无二的稀缺价值：这里聚集了一代又一代懂得如何把早期微小火苗规模化为独立巨头、并构建起庞大网络效应的顶尖人才与操盘手。 在回应此前 Reid Hoffman 经常被问到的经典问题“旧金山依然是全球最好的创业起步之地吗”时，Josh 给出了十分精辟的论断：“我不认为它是世界上最适合创办公司的地点，但我认为它是世界上最适合进行公司规模化（Scale）的关键之地。”（“I don't think it's the best place in the world to start a company, but I think it a very important place in the world to scale a company.”）如果团队正在打造 B2B 软件或面向科技公司销售工具，湾区的浓厚客群是非常优渥的试验场；但如果做的是消费级产品，其本质在于连接大众文化。创始人若一味戴着“旧金山眼镜”（SF Glasses）闭门造车，便很容易脱离全球各地主流消费群体的真实生活习惯与在地文化。 - **全球创新的多极化**：Lovable（瑞典）、ElevenLabs、Musical.ly（中国团队发源）以及中国直播电商，证明了颠覆性的消费创意完全可以在硅谷之外孕育。 - **湾区的真正红利**：不是最初那一瞬间的点子，而是拥有大量经历过从 0 到 1 再到 100 完整周期的操盘手，能为处于关键拐点的公司注入规模化扩张的经验。 - **警惕“硅谷滤镜”**：旧金山独特的极客与科技文化并不等同于全球主流大众文化，消费类产品必须深入理解多元城市的地域生态。 ### `54:59` 黄金年代的 Twitter 与 Evan Williams 的产品哲学 回顾自身在 Apple、Robinhood、Facebook、LinkedIn、Musical.ly 以及 Twitter 等多家科技巨头的工作履历，Josh 坦言虽然很难挑出唯一的最爱，但 2009 年底至 2010 年底在 Twitter 的那一年是他职业生涯中最难以忘怀的黄金时期。当时整个团队怀揣着一种近乎传教士般的纯粹信念，坚信这个简陋的平台能为世界带来不可估量的改变。彼时大家齐心协力攻坚最艰难的技术与产品难题，公司尚处在从未被完全定义为成熟大企业的上升期。Josh 将自己定义为一名“早期加入者”（Early Joiner），比起从零起步创立公司，他最大的热情在于找到已经初具势能的伟大火苗，并帮助其长成参天大树。在他加入 Twitter 时，平台活跃用户仅有约 1000 万，而在他离开时，这一数字已经突破了 1 亿大关。 谈及管理与产品决策，Josh 特别澄清他在 Twitter 任职期间正值联合创始人 Evan Williams（同时也是 Blogger 的创始人）执掌公司的时期。他从 Evan 身上学到的最核心一课是：“我们楼里有很多聪明人，每天使用我们产品的也有很多聪明人；别自以为是地觉得自己无所不知，多去倾听他们、向他们学习，而在做出决策时，务必将自己的思考逻辑完整地传达给他们。” 为了阐述这一理念，Josh 拆解了 Twitter 早期最具争议的产品改动之一——原生“转推”（Retweet）功能的上线过程： - **早期民间转推的痛点**：在官方功能推出前，用户转发推文必须手动输入“RT @用户名”，这导致转发出来的推文以转发者本人的头像和账号发出，在形式上侵占了他人的表达。 - **坚守原始声音的真实性**：Twitter 团队敏锐地倾听到了社区对于扩散放大他人声音的强烈诉求，但在具体机制上采取了非常坚定的立场——团队坚持保留原作者的头像与账号署名，通过展示“Josh 转推了 Ollie 的推文”来保障原作者的版权与主体性。 - **倾听与主见的平衡**：这项改动在刚上线时因改变了既有用户习惯而招致了巨大的反对声浪，但团队坚定地向外界阐释自身理念，并最终沉淀为塑造当今社交媒体基础架构的核心范式：“在充分倾听的同时，仍然保持极其坚定的立场，然后把这种思考逻辑对外部做深度沟通并展开探讨，即便许多人一开始因为习惯被打破而并不高兴。”（“And there was this incredible lesson on listening, but still having a strong point of view and then communicating that back out and engaging in discussion, even when a lot of people weren't happy with it at first because it changed the behavior a little bit.”） ### `58:00` 尾声：对新一代科技浪潮的展望与致谢 Ollie 随后笑着感慨，自己在 2009 年时才仅有十一二岁，但依稀记得当年注册 Twitter 账户时的情景，并盛赞 Josh 伴随这些社交巨头一路走来的传奇历程。他呼吁播客听众在社交媒体上持续关注 Josh 针对前沿科技与消费产品生态的深度洞察。 在节目的最后，两位嘉宾提及了即将由 a16z 在旧金山主办的盛会 SF Tech Week。Ollie 分享了自己去年参会的震撼体验，指出该活动期间全城汇聚了多达两千场周边聚会与峰会，是近距离感知湾区创业活力与技术脉搏的绝佳窗口。作为 Josh Elman 加盟 a16z 后录制的首档官方播客，双方在轻松愉快的致谢中结束了本次对话。节目最后也附带了标准的合规声明，强调播客内容仅用于教育与信息交流，不构成任何投资或买卖金融产品的专业建议。 ## Notable quotes > `35:10` 我始终深信，如果你想成为一家极其庞大的公司，你绝不能一开始就试图包揽所有事情，而是必须先把少数几件事做到极致，从而赢得做更多事情的权利。 — Josh Elman > `10:00` 在采用初期，实用性往往胜过信任。人们会尝试新事物，会去折腾试验。但如果你真正想打入主流市场，触达数亿乃至数十亿人，信任最终会变得至高无上。 — Josh Elman > `37:34` 我目前并不相信我们所有的交互都会彻底转向语音。我认为这始终是一种平衡，纯语音 Agent 的设想在现实中其实站不住脚。 — Josh Elman ## Mentioned Josh Elman, Ollie Forsyth, Anisha Acharya, Ali Yahya, David Haber, Reid Hoffman, Evan Williams, Michael Birch, Andreessen Horowitz (a16z), Greylock, LinkedIn, Facebook, Twitter, Robinhood, Apple, Siri, OpenClaw, Hermes agents, Roblox, ChatGPT, Claude, Google Gemini, Amazon, Whatnot, Musical.ly, TikTok, YouTube, Plot Party, Discord, Instagram, WhatsApp, Bebo, Substack, Partiful, Reid AI, Lovable, ElevenLabs, Blogger, 《为全人类》（For All Mankind）, 《AIR》（气垫传奇）, SF Tech Week ## Video description a16z Partner Josh Elman joins Ollie Forsyth on New Economies to discuss the next wave of consumer AI and what separates a product people try once from one that becomes part of their everyday lives. Josh argues that getting attention has actually become easier, but getting consumers to stick is harder than ever. He explains what he looks for in consumer products, why the best ones start with a narrow wedge and earn the right to do more, and why trust becomes increasingly important as AI agents gain access to more of our personal lives. They also explore personal AI agents, the future of shopping and entertainment, why we haven’t seen another major social network emerge, and how AI could make technology more social rather than less, including agents that help people spend more time together in the real world. This conversation originally appeared on the New Economies podcast. [Source](https://mgln.ai/e/1344/afp-848985-injected.calisto.simplecastaudio.com/3f86df7b-51c6-4101-88a2-550dba782de8/episodes/60c2b678-3eec-440d-a9fa-890e85aa5a23/audio/128/default.mp3?aid=rss_feed&awCollectionId=3f86df7b-51c6-4101-88a2-550dba782de8&awEpisodeId=60c2b678-3eec-440d-a9fa-890e85aa5a23&feed=JGE3yC0V)

[35:10] Josh Elman: 我始终深信，如果你想成为一家极其庞大的公司，你绝不能一开始就试图包揽所有事情，而是必须先把少数几件事做到极致，从而赢得做更多事情的权利。 [10:00] Josh Elman: 在采用初期，实用性往往胜过信任。人们会尝试新事物，会去折腾试验。但如果你真正想打入主流市场，触达数亿乃至数十亿人，信任最终会变得至高无上。 [37:34] Josh Elman: 我目前并不相信我们所有的交互都会彻底转向语音。我认为这始终是一种平衡，纯语音 Agent 的设想在现实中其实站不住脚。

[35:00] Josh Elman: I'm a big believer that if you want to become a very big company, you can't start by trying to do everything, but you have to start by doing a few things really, really well and earn the right to do more. [45:26] Josh Elman: I always like to ask people, what did you stop doing when you started using this new thing? [22:38] Josh Elman: It's not just like some Borg AI is going to be making things. It's still going to be people making things to tell stories that they have. [37:30] Josh Elman: I think we're going to have a lot of times where voice assistants should be able to let you ramble and let you have that long thought pattern and then help you distill the real meaning out of it and capture those core thoughts.

[0:00] It's never been easier to get consumers' attention. What's harder, I think, is getting me not just to try something, but to actually stick. It has to be such a good product doing that one thing really well out of the box that I either stop doing something else or I turn more behaviors over to it. And we have a lot that we already have in our digital life. So now as a product, you have to figure out some new wedge into somebody's digital life. And then becoming so good that it becomes a habit and that you can pull other people in. clear value that you can describe to somebody. If you want to become a very big company, you have to start by... It's never been easier to get consumers' attention. The hard part is getting them to stick. In this episode, A16Z partner Josh Ellman joins Ollie Forsyth on New Economies to discuss what it takes to build a lasting consumer product in the AI era. Josh explains why the best products start by doing one thing exceptionally well, become a habit, and then earn the right to do more. They also discuss why trust could increasingly important as AI moves deeper into our lives, from shopping and entertainment, to personal agents that know our preferences and act on our behalf. They also look at what might come after today's social networks, how AI could bring people together rather than replace human connection, and the signals Josh looks for when deciding whether a new consumer product can actually last. Thank you. Anisha Acharya, Ali Yaha, David Haber, and now you. So I'm super excited to have you here. So thank you so much for joining us. You know, when I was thinking about the title for today's episode, there is so much happening in consumer tech and can't wait to dive in more with you. But let's start off with the basics. You've actually been on the investor side, you've been the founder side, the operator side, now it's kind of like come full circle. Why are you back in investing mode? When I think about a lot of my career when I got started in the late 90s, it was the dawn of the internet. And this idea that we could make products that could change the world, that could get into hundreds of millions of people's hands and help them have a better life was so exciting to me. And I've kind of had this amazing career of doing it, and I've always tried to find the right place. Sometimes it's been diving in and working on something that I was so passionate about that I wanted to help bring to the world. Sometimes it was, you know, in 2011, I had already been worked at LinkedIn, Facebook, and Twitter, and I thought, wow, I'm going Maybe there's a great opportunity for me to help the next wave of companies. And so I became a VC and got to invest in a lot of the next wave of the mobile Internet. and the social internet, About eight years ago, I was like, maybe my best times will be spent helping build companies again. And I went back and I joined Robinhood. And then I spent the last many years at Apple. And then starting this year, I was so excited for what we were doing at Apple. And we were kind of bringing out this new version of Siri, Siri AI. But at the same time, I was like, we're about at the beginning of a Cambrian explosion of new consumer experiences around AI. Now's the time I really believed it could happen. And I felt like in my career, I have one more path. I want to get back on the side of founders. I want to help founders create incredible companies. And I thought, you know, at this point, what better firm to do this in the consumer world than Andreessen Horowitz? And was lucky enough to have known people like Anish for a very, very long time and brought me in to help kind of get excited about this next wave. You know, the rules of consumer tech have changed so much now. And you mentioned a little bit around the different types of experiences that consumers are now experiencing versus maybe even five years ago. Can we maybe do a comparison? What are the new consumer experiences five years ago versus what they are today? Five years ago is a really interesting time because that was the heart of COVID five or six years ago. And... I like to say in the period of COVID, we realized just how much time we can spend online and where our limits are because we had nothing else to do. We were on video calls, which normalized video calls. That used to be kind of a weird thing for someone to propose getting on a video call. It's become crazy. Something that makes sense when it makes sense when we can't be physically in person. We had so much sharing going on. We saw massive rises in gaming, both mobile gaming, social gaming, where we all play together. Roblox has been an incredible phenomenon that's continued growing through that. And we saw so many companies, but I almost say they kind of hit a cap and we realized that we still need time for ourselves in the world. And as COVID sort of relaxed and people felt more comfortable getting out in the world, we've kind of seen this rebalance where a lot of things that happened we're still spending time on. But now... We want to be more in the world. And we actually, with this incredible rise of AI, we want things to help us actually be more present in the world. What's incredible about AI right now is AI, There's all these new tools that can help us get things done so we can do more of what we want to do. Whether those are agents, whether those are personal agents. [5:00] Obviously, a lot of work and productivity tools. And we're just at the cusp of what I think we're going to be able to use AI for. And I think all of it's going to be like this period of coming out of COVID, giving us more time to experience things and do the things we really want to be doing. I think AI has obviously changed completely how we're going to work and live in the future forever. But honestly, sometimes I think we're actually pretty far ahead with AI already. Some people I speak to, we're just scratching the surface. Where are we on this incredible journey for AI, especially for consumers? I mean, I think there's three different buckets of people. I think for early adopters, what people are exploring and seeing in Silicon Valley, we've got people who are truly living at the edge. They have AI managing their lives, managing their work. Developers who are used to writing lines of code manually don't have to do that anymore. They get direct swarms of agents to write things for them. And with OpenClaw and Hermes agents and other things like that that have come out over the past year, six months or eight months, really. people are now getting things done in ways and their tasks done that they could never imagine. If you then go to the workforce, we're seeing a whole bunch of tools truly enter the workforce. Obviously, AI coding is a massive phenomenon. We seeing people use AI to help them build spreadsheets analyze documents do research and it truly making work more and more productive When I look at consumers we are very very early Most consumers are a lot of people are using Chat2BT and Gemini and other things, and they've used them to kind of run searches more effectively. Instead of running a search and clicking on every single link to read the pages, you can get a better synthesized answer. People use this for homework. Some of them use them to consult and brainstorm on things for their financial problems or financial opportunities or legal challenges or simple things they're working through. Just if you want to write a hard email, AI is this great new consultant to do that. But in terms of really transforming their lives in this new wave of either productivity or experiencing more things, we're barely scratching the surface. And then on the same time, on the entertainment side, there's a lot of questions about where does AI fit in to the future of entertainment? People want great stories and they want to just consume them wherever they come from. But we're seeing a lot of interesting things happen as AI becomes either a tool for that or create some challenges with the way that we're traditionally telling stories. I love to hear your take. I actually, I love all of those points. I kind of think this year has been the main breakthrough year for AI, especially for consumers You know GBT now 900 million weekly active users This thing only launched four and a half years ago And if we just scratching the surface here you know it can now do code for us All the things you just mentioned, like writing emails, giving it our transcripts, figuring out like what to cook, like got a bunch of things in the fridge, like what can I do useful, right? Where to go on call spots on the weekends. You know, this is just scratching the surface, right? What do you think comes next? If we've already experienced this in the last four and a half years, I kind of think of these iterations as product updates, right? You're at Apple, so I'd love to obviously hear how you spend your time now. Have a new iPhone once a year. We can't do that with AI. Updates happen every few weeks. I think we're in this really interesting moment in time where the people who've been living and breathing this AI world wonder why everybody else isn't on that cycle and cadence of updating. But I really think it started at the beginning of this year, late last year. the models got so much better towards the end of last year that they could do long-running tasks. They could actually have loops of improvement. Agentic tasks finally started to work, and the coding and envisioning of something got so much better. And that class of models resulted in something like OpenClaw that one person wrote by himself almost in his apartment and came up with a whole new way of interacting with models and ways that models could remember things in skills files and put all of that together in this modern experience that showed the people who are very deep in AI how different it could be. And since then, we've obviously been on this race to try to bring this to so many people. But you talked about this cycle of adoption. And we sometimes, when we're living in this AI world, it's breathless and you can keep trying new things and every week a new model comes out and you want to see how that compares to the other ones. And a new incredible app comes out that everybody wants to try and see what it can agentically do for you. There's been a few hot ones even in the past couple of weeks that people are having incredible experiences in the tech world that weren't possible before. But to really bring this to consumers, it's going to take a more measured pace. We have to bring people step by step into understanding what does it mean to have AI doing things for you? What does trust mean here? You're asking it to like have my credit card, take care of things, understand my life. how do we build trust and privacy in? And you talk about how Apple kind of ships a new version of the software and the hardware yearly. And it's an incredible motion to bring people along on this journey. And sometimes I always wonder how much faster are people really ready to keep changing their lives up? And they bet on one thing and then they're wrong two months later and they have to change everything. A lot of regular consumers who are just trying to [10:00] through their lives, their jobs, maybe things that aren't as immediately impacted by AI, aren't necessarily at that same pace So I'm looking for those companies and those founders that can bring people really on that journey. I love you mentioned trust, Sage. You think trust is a new big mode for consumer AI? apps and companies. I mean, I think in these early days of adoption, utility often trumps trust. People will try new things and they'll experiment with it. But if you truly want to break into the mainstream and really get to the hundreds of millions or billions, you're going to be a big deal. trust ends up becoming paramount and the seriousness of a company and how it doesn't want your data to make itself better and trade that with other people, but wants your data to be. only in order to help give you a better experience becomes important dividing lines. I think The things that will break into the mainstream will have a fundamental tenet of trust. And just the way we search for items or things for online now for research, like I didn't know when the last time you Googled something, But I haven't Googled something in ages. I go to Claude, or I go to GPT, Gemini sometimes have different things. The fundamental way we search for online now has radically changed forever. It's a great point. I think Google for a search when I kind of know what I want, but I don't quite know and I want to describe it. It's hard to get exactly what I want out of that. And I personally find myself going to Tachibiti or Gemini as a starting point for a lot of these conversations. I was recently looking for a keychain that makes it really easy to split for when I have to park at a valet garage and then put it back together and... I was having so much trouble on Google trying to find something, on Amazon trying to find something. And I went to ChatGPT and started having this great conversation. And it found me a bunch of, it kind of described to me a bunch of the different attachments and the way that they could connect and easily break. A bunch of the ways that those either had maintenance problems or the ones that were most likely to be durable over the years. And then I said, and I really want one that's made in America and helped me find just the right one that was made by a store in Michigan. not... Googling, clicking on links and having to do all the work myself. And I think that experience is just one of so many, but there's still so many ways that I could have productized that to be a lot better experience. And still, I kind of still wanted a chance to browse at the end of my journey. And ChatGPT had kind of helped me narrow on like one option. And I really still wanted now that shopping browse experience, but with all of my conditions laid out. So then I could really kind of explore from like the perfect set of options. When we think about those new experiences for shopping online, especially through these AI models, what would you like to see that still hasn't emerged yet? You spent 10 years at Apple, so you've probably got a whole bunch of ideas on how you can make this product so much better. Because right now, it is still pretty basic. It's only six and a half, not quite six years at Apple, but it was an incredible time there. But when I think about shopping, What's really interesting is the best experience is when you can have a concierge experience at a store and you have somebody who's a true expert. They're guiding you, commenting on maybe what looks great, helping you try different things on. And even better is when you know that their only incentive is your success. Their incentive is not to get you to spend the most money or buy something where they get a kickback, but it's truly that guidance in helping you. And I'm really excited to find the companies that can sort of Bring us that experience, but using AI to actually simplify from the vast array of options. But it's truly on our side and it's something that we learn to trust because it has our best interests and all of our questions. And I'm very happy to, you know, try something on, take a picture, take a picture of stuff I have that I want to upgrade and have AI comment on that and make it better. This thing I'm wearing, I actually designed completely with AI, where I said I wanted a new black polo sweater and it actually helped me figure out the design and went and actually had it manufactured for me in China. And so I'm so excited by sort of all of these ways that we can improve the experience, but it truly has to be something that's on my side. There you go, Josh. If EC doesn't work out for you, I'm sure it will, That was pretty good. I treat it all to the tool. I just described what I wanted and it did an amazing job for me. But I think it goes to show, you know, this is how far we have come in four and a half years. The next even 12 months, I think it's going to be incredibly exciting. Well, there's some of the trends you're really excited about coming from a consumer lens that you're really excited to see in startups and hopefully back them. There are so many ways that AI is going to reinvent the experiences that we do every day. I mean, it almost goes from anything. You know, I start with being really excited about the personal agent that is sort of the thing in my corner that I can rely on when I'm trying to figure out what to do this weekend and, maybe scheduling appointments or helping me monitor my health, whether it's helping me think about my finances and whether there's opportunities I'm missing out, whether it's just simply [15:00] checking in and making sure I've had a good day and seeing if there's anything I need to just do that's on my mind. That moment of something, even just checking in with you, can either be naggy if done poorly, but amazing when done right. And it feels like it's really on your side and it's helping you think through things. So this area of personal agents, and I went through a few of the verticals too, things like health and finances and setting your goals and helping you kind of get through your day better. Maybe it's helping you shop and make sure you have the right amount of food home that you need. Maybe it's helping you plan something fun because it can notice, hey, I'm This weekend, you have a lighter weekend and maybe it's time to go out and there's amazing weather and or this incredible festival is happening. That's one area of just these personal agents that become part of our lives that are very, very different than all the work agents that are helping us work more productively. The second area I get really excited about is the future of entertainment. Right now, There's so many things to watch and things to experience, but you end up having to just consume and absorb. And what's really fun is when media is interactive and exciting and it's your world. And, you know, when you're watching a story, one of my favorite shows on is on Apple TV. It's called For All Mankind. And there's so many times that I'm like, I would love to go deeper on the backstory here. Or I would love to imagine if this happened instead. Now with AI, you could imagine that any world that you're in, you can go much deeper. You can have these experiences. And even so, I wanted to be able to share them with somebody. And that's sort of in the fan fiction world, but you can do this even for new worlds and worlds that somebody else creates and lets me start to experience and explore and add into it. I think of what Roblox has become as just an incredible phenomenon of all of these games, all these shared experiences people have. And I think that there's, It's still very much in the game world and has a certain group that really enjoy Roblox. And I think that opportunity is even bigger for so many more types of generated experiences. And we're going to have a world of creators that can imagine these worlds that we then get to participate in and interact in. I think in terms of, you know, verticals, I mentioned travel or I mentioned health and finance, but I also believe in travel. You know, right now, when you want to go somewhere, you are spending some of your hardest earned time and money to have a great experience. And yet you don't always have full visibility in what's there and you can't always express your own personal interests. Having an expert that can actually help bridge that gap is an incredible opportunity. home life in ways that you don't think about. So this whole world of kind of vertical services that are much more powered by AI and technology. very different interaction model and very, very personalized to you. Those are just some of the things I get really excited about. I mean, there's so many more. You know, we talked about fashion and being able to actually have something made for you that you've designed in your mind, being able to connect with people in ways that AI facilitates. So it's like, hey, I want to ask this question, but you can role play it with AI and then it might actually do some of the work to go reach out on your behalf to some things and bring you back answers before you actually then go have the human connection. I don't want to add a replace human connection, but anything it can do to get me closer to having even better, more impactful human moments is going to be incredible. I love all of these trends. One, you mentioned that shopping. One of your companies, whatnot. I'll be very honest. When I saw these live shopping companies in Europe and in the US, I was like, what? They work super well in China, by its most recent time. They work super well in Asia, but to adopt that in the US or Europe is obviously very different. One of the just raised a massive round, I think, a few weeks ago. What have they done so well that maybe consumers just didn't quite see you? and the bestest as well. I mean, I think live shopping is just an incredible category. And I think Whatnot as a company has done a fabulous job really expanding their marketplace and expanding all the ways in which people can buy things. But what's really fun with something like live shopping is that it's much more about participation. I can go to the store, I can go to a static store and just click on a few things. But when I'm actually on WhatNot, I'm hearing the story behind the products. I can actually ask questions. I can interact with the people who are presenting that. I can explore different things and different products and have just so much more depth and story. And we're not just buying a flat thing that does something in our lives. We're actually buying a whole story in a way that some new product comes into our lives. And whether that's fashion where you're talking to the designer or fashion where you're talking to a curator who's done so much work where they're telling their story of why they think these are the best, so many new ways for that interaction to happen. And it makes the shopping, when you make your purchase, you feel much more both confident, engaged, and have a connection to what you're buying. So I've just been blown away by their execution over the number of years. They were, again, one of those companies that [20:00] Like you said at the beginning, Is live shopping ever really going to work? in the U.S. You know, there were clearly some evidence of it working in China and, you know, when it starts to really flourish, you see massive new behavior and you see people buying and selling things that, in a way that is, again, much more real and much more about a human connection. I think live shopping and also entertainment, right? You mentioned kind of live entertainment with interactions. There's this whole new trend happening now, AI microdramas. I think that's going to be a really interesting category to watch. I mean, I think that, micro dramas are a huge part of the next wave of entertainment. You know, it kind of started with TikTok and actually even Musical.ly, which I had backed it when I was at Greylock. We found the team that was building Musical.ly and they had an idea that, short form video, again, initially with even like a music soundtrack, can create a whole different experience that you watch and you have a meaningful connection before you go on to the next one. And that as they learn which ones you have meaningful connections, it becomes this really wonderful feed of content that speaks to you, lets you both connect with the creators and connect with the content in a new way. And as that was sold to ByteDance and became TikTok, it has flourished in a way that it truly has some of the richest, best content anywhere. But now what happens is those are all these nice short videos. And often you want to go deeper. And often you want a richer story. And so the micro dramas have come in and people have been able to create these perfect vignettes. And they fit really well into places like TikTok and YouTube helped first discover them. But then you're like, you're immediately hooked and you want to finish the story. Now, we can only make so many of those. And each of these companies has done a great job making a lot of different stories in a lot of categories. But with AI, we can open up this creativity to so many more people, people who might not have a studio or a cast or sets to go film something. and have that made. And what I'm excited about is, How do we find those writers, those creators, those people with brilliant ideas, give them the tools to express them, help that get in front of more people? I think there might be incredible new marketplaces where people come together to see the creations from others, read those, engage in those stories and be entirely new media channels. Right now, a lot of the distribution might still be driven through existing social media channels. But I think there's room for massive new marketplaces that become the first place you go when, hey, I want a story. I just want to engage. Super interesting. So guess what? We actually just created our first AI microdroma. You did. What is it? So it's on the Nike story. So an idea I had was how cool it'd be to go and interview all of these incredible founders, the Jack Mars, the Phil Knights of the world in person. Slightly difficult though. But hang on, why don't we tell these stories in four to five minute microdramas on how the most incredible companies got started, their origins? Obviously, there's always a bit of a romanticized love element part of each microdrama. So we collaborated with this company called Plot Party, which is like an early stage startup. And yeah, we published it, our first one a couple of days ago. I'll send it to you afterwards, but super interesting. We do all the storytelling. They do a lot of the creating. It just goes to show the different types of entertainment you can create. I'm not a movie director, just hopefully a good storyteller. And we can now create dramas and get millions of views from it. Oh, that's amazing. I can't wait to watch it. And I mean, you're a great storyteller. I've seen you through these podcasts, be able to put together just amazing stories. Now I can't wait to see how you do that. I remember the Nike movie from a couple of years ago. I think it was Air. if I remember. And I mean, that, And that was like an incredible movie. And it's always hard to take a corporate story and actually bring drama into it and make it feel cinematic. But, Now that anybody can have access to these tools, I'm sure you did the same in just a short little micro drama. And I think so many people have that talent. And the hardest part now is going to be less about the creation. We can have a lot of people with talent, but it's still going to be the things that resonate. Finding an audience, figuring out how to build a relationship with the audience where they come back to you. The Nike one's great. Now I might want to hear the Amazon story or the Microsoft story. So think about great companies from the Pacific Northwest. Or you might go around the whole world and keep telling these stories that people don't know, interest. And it's all about taste, right? But I think, you know, as we enter this new era for media, creators, it's also just what can you do so differently that you can own? You know, I kind of feel like creators should be building products that they can own. And obviously, like just build hopefully into incredible companies. Now's the time to be doing it. Yeah, and I think for creators, there really is an identity that they build. And they build a relationship through the content that they produce, through the stories they tell directly, through the stories that they bring out from other people, through the story [25:00] that they might... you know, create through AI, but it still becomes an identity and a relationship over time. It's not that you can slop lots of things in lots of different categories, but it's really that relationship with the creator that is what's powered the most recent wave of these incredible, you know, user-generated content companies. And I think even as we move into an AI world, It's not just like, Some Borg AI is going to be making things. It's still going to be people making things to tell stories that they have. And those relationships you make with those storytellers, those creators is still going to be very important. And for each person, it's building that set of relationships. You know, one thing I've been super surprised about, to be honest, is we've had all of this amazing technology emerge in the last couple of years, but we haven't seen any new social media networks ever. sides. emerge. You've worked at LinkedIn, you've worked at Facebook, you've been at Discord on their board, you've also been at X, and maybe a bunch of others are missing. Why haven't we had a new social network emerge in the last few years? It's a great question. And I think, you know, in some ways, We haven't needed to. Like the social networks we have are at billions and billions of users, you know, or hundreds and hundreds of millions. You know, you mentioned Discord and Musical.ly, which, you know, Musical.ly became TikTok. Both really started growing in 2015. That was the year that I got to invest in both of them in those early days. And we had to navigate this question even in 2015. Well, we already had Instagram and Facebook and WhatsApp and Twitter and LinkedIn in 2015. Why would we even need things like Discord and Musical.ly? problem in our lives. Discord was great for people who were trying to come together to play games and engage. Musically was great for people who wanted to actually tell stories and view stories from other people. So much of Instagram, I used to say, was the back-facing camera. Look at the incredible place I'm in or the incredible latte I'm having. Musically was the I'm just going to create something. I'm going to put some music behind it, do a dance. I might have my piles of laundry behind me in my room. But because of my dance and because of this song, it's going to feel like a great creation. I'm proud to share. And it's obviously grown up a ton from there. And we need something that feels new, that feels different. I've got a story I want to tell. I'm going to bring everyone together who's telling these kinds of stories. And this becomes the place you go to read those stories. Maybe that looks more like a marketplace than a social network. But I think there's a huge opportunity for that. lot of the attempts at social media between then and now were things that felt derivative, that felt like, well, I'm already doing it elsewhere, come and do it here slightly differently. There's a lot of variance of X and Twitter are, that actually feel very similar to X and Twitter, which is a slightly different group... group of people, it's hard for those to catch and truly become mainstream. But I think we're on the cusp of seeing some things get invented that have us thinking in new ways, asking questions of very large groups and seeing information that isn't actually available and that people wouldn't otherwise be sharing in social media that can become a whole new network. I think we might see these creator marketplaces where people are actually creating new stories and publishing them for other people. I still think, as we talked about coming out of COVID, that really help us get back out in the world and do more in the world and kind of bridge that online to IRL life. Partiful is one example. And Andres and Horowitz portfolio that I think is a really exciting platform. thing that if you're going to different events, Partiful actually is the places you really want to be and it becomes a great way to go out and explore the world much more with the community that you have. In real life, though, it's super hard. Do you remember Bebo? I do. Yeah, so I spoke to the founder for the first time. It's the first podcast he's done in about 10 years. It was Michael Virch. He's incredible. Yeah, Michael Virch. We spoke on Monday and I asked him the same question. Why hasn't there been a new social network emerge? But I also just think the appetite and the consumer websites on how we engage with these social networks has massively changed. We don't post on our Facebook timelines as much anymore. X is our go-to maybe consumer app for all these content. which was an awesome app, Subsec, another A60Z portfolio company. There's just different appetites, right? Again, I kind of just wish there was a new social network which emerges. We haven't had any new apps in a few years. I mean, I've heard a lot of people... kind of say that and want it to emerge. But I also say, you know what? We have great ones. And maybe we're pretty satisfied. And after I've checked out the people who are sharing on Instagram and talked to some of my friends who are still on Facebook and seen what they're up to and commented on their posts and went to TikTok, I'm no longer going, what else do my friends have to say? But I do think we're going to see things about what are people creating. Where can I go co-experience things with people? Now, one of the amazing things with Discord was it was one of the rare [30:00] cases you could really easily game and talk at the same time. And it was actually a new behavior that brought people together that then became a place people stayed and chatted. And so I think about what are those things that are those new behaviors that bring us together that then once we're together, Now, we'll also stay and do other things. We also talk a lot about AI, and right now a lot of AI is very single player. I interact with the AI to help it do things in my life. But I don't yet have it plugged into our group and understanding our group nuances. And when I have it join a group, it's not helping me do other things with the same group of people. Group chats have become such a popular form of social media. Maybe that is in some ways the new network that we aren't thinking about that we really have developed over the past few years. I have way more local group chats than I probably had 10 years ago. But when we bring AI into some of those group chats, does that create a whole new pattern where we're now interacting? And even more, what if the AI is interacting our behalf to bring us together? And it's like, hey, I'm talking to it. You're talking to it. They're talking to each other. Now, when we have a conversation, we're much more informed. I think we're just at the cusp of many of these ideas. Starting to get explored and have people trying them. And I'm really optimistic in this wave a few of these will catch. I'd like to double down that. Okay, so if we're going to go down this approach of multiplayer communication between agents and consumers, especially on maybe communication apps or all of this, helping bring people together. So realistically, what does that actually look like? Is that Josh and Ollie communicating together, maybe with a bunch of their mutual friends, they're trying to meet up for the weekend, trying to figure out where to go? I mean, is that just the basic? I mean, I think it's a great basic place to start. That was actually the example I was about to give you, which is I get into my weekend and I'm like, which of my friends are even available? Now, it's a little bit embarrassing sometimes. We're like, who's up for something and get crickets back? And you feel very vulnerable when you're asking that question. But what if I could go to my agent and I'm like, can you just root around and see if anybody's up for something this weekend? And it could talk to some other agents and to see if anybody's asked their agent the same question. And once it has, it starts to make a match for us. And in some ways, people say this sounds a little bit dystopian. But I think in some ways, it's actually AI being the glue that brings us together, thoughts. I kind of want to do something. I don't really know. Has anybody seen Odyssey? You know, maybe everybody else has seen Odyssey who hasn't seen it yet. They might want to go. And I, If you can actually plant that seed in somebody who cares about you and can go root around and find somebody else, maybe that becomes a way that now brings us together. So I'm always looking for these opportunities where the agent can make us have better human experiences and more human experiences and make those happen. I've always dreamed that imagine that we're all on a Friday night, a bunch of people are sitting at home talking to their AI about something, and it recognizes, hey, there's like 10 of you in... San Francisco in Noe Valley, why don't you all go to the board game cafe and just play there? You know, they got two tables ready. I can even reserve them for you right now. Wouldn't that be amazing if it actually moved us from, talking to something online to bring us into that real world. That's insane. Okay, so a couple of questions there. It feels this is maybe like the basic middle layer of ideas of where we could go. If we think crazy big, How... Crazy can this, Ashi guys. If you think crazy big, you just wake up every Saturday morning and you talk to your agent or look at your calendar and you just go, oh, no. I've got a rich day for you. You're going to start here with this person. You're going to have a break this afternoon. And you're going to go have dinner with these friends because everybody was looking for something tonight. Sunday, you'll do this. Tuesday night, it's this. Your agent all of a sudden is just becoming your social calendar. And you're really getting out in the world a lot more. And every time you talk to it and you're like, oh man, my favorite shirt got really frayed and I want a new one. It's like great, ordered for you on your way. that true assistant looking out for you, is. And the goal, again, is to save you time from the things you don't want to be doing to do more of the things you do and that you enjoy and that you love. And by the way, a lot of people need time for themselves. The agent should be able to protect that and help remind people of that too. Coaching you, getting more exercise, whatever goals you have, wanting to Eat better or indulge yourself sometimes. Having a partner who understands your goals and sometimes gives you encouragement and sometimes gives you the tough love that you need. I mean, that's what we look for. I mean, I love my mother and she's great at doing both of those. Sometimes being my biggest cheerleader, sometimes being my biggest critic. And, it would be amazing to have something else in my life that is helping me do that. I love my mother. She gives me tough love and encouragement. But I don't talk about everything, [35:00] talk about everything every day. And there are times where I would love somebody who kind of has the day-to-day understanding that gives me some of that too. And even... If I haven't called my mom in a week, remind me to do that. I'm usually pretty good. Super fascinating. It makes me think, though, what type of platform could actually create this? Is it one with existing distribution, the likes of, GPT, Claude, are they going to enter this space? Because it's going to be a very difficult, I think, for a new, very small startup to merge into this category if they don't have the distribution system. these air models have. And I think it's going to be piece by piece. I'm a big believer that if you want to become a very big company, You can't start by trying to do everything, but you have to start by doing a few things really, really well and earn the right to do more. Again, I keep coming back to Discord, which earned the right to be the place that we played games. and that we had a live chat during games. And as it earned that right so well, it became a place we chatted more. And we chatted more about just playing games. And then we did spend more time there. And then more people did that. And even people who didn't play games came over to use the service to chat with friends who might also be using it to play games. But it slowly earned the right to be something that you relied on for more. And I think for a lot of these assistants, if you just say, hey, I'm a general thing. I can take care of everything. You don't earn the right to be really trusted for getting me a few things done. And so what does that agent do? And that assistant and that product that, It goes step by step by step. And once we learn to turn to it for one thing, it now earns the right for something else and something else. When I think about Siri, I know it's something that I spent a lot of time over the past many years thinking about. Siri was actually amazing to earn the right to do a few things for you really well. Place phone calls. Set timers. Set reminders. Send a text message. But when you try to do more things with it, it often couldn't. There were things you would expect it to do and get it right, and it didn't. But over time now, you only get so many chances to go, now try one more thing. Now try five more things. Now try a lot of things you can imagine. And you'll have a much better conversation when it can't do something. It can explain to you that it can't do it. really understand that as opposed to just like put up a hand and say, sorry, I can't help you. And so I think there's so much opportunity for new products to do the same thing, but they're going to have to go step by step. When we think about the new rules of consumer tech, voice is going to be huge. It feels we're just scratching the surface. You spend a bit of time on Siri. What are you excited about right now in voice? And again, it feels like we are just literally scratching the surface. When I think, you know, we're talking here and people are listening, like voice is an incredible medium to express a lot of thoughts. You can often talk much faster than you type. I'm one of these people that tends to think out loud. And so I'll talk and talk and talk until I kind of fully refine my point of view. Voice is also lossy, though. Sometimes you wander and ramble a little bit until you get something. When I'm typing, I might type it and ramble a little bit. But by the time I edit and polish the thing that I'm typing to send or communicate to somebody else, I've gotten it really tight. And so I do think a lot about where voice sits as one part of the entire continuum of communication. I think we're going to have a lot of times where voice assistants should be able to let you ramble and let you have that long thought pattern and then help you distill the real meaning out of it and capture those core thoughts. We have a lot of these, you know, on the work side, we have all these great meeting note takers that kind of record everything and synthesize them incredibly well. On our personal lives, when we have ideas, we want to be able to ramble and think and dream and have it sort of capture what we need to do. When you have just a quick idea and you want to remember it later, you're going to be able to remember it Voice is often a better paradigm than having to type it in because it's the easiest way just to quickly express your thought. I'm not yet a believer, though, that We're all going to move to voice for everything. I think it's always going to be a balance. And I think the idea of a voice-only agent doesn't really make sense. It's a, hey, there's times where I want to type. I might be in a quiet place. I might not want other people to hear me. And I do want to express something. And I think voice isn't always the best use if I want to refine my thoughts a little bit before I send them over. to an agent or to another person. But it's – We're doing a lot less with it today than we should be, and I'm really excited for this massive wave. Voice is also such a great tool. If you're coming to a podcast show or if you're preparing for a speech, it's actually a pretty cool interaction tool as a sparring partner. Oh, interesting. Yes. And with a lot of these new duplex modes and what some of the new technology is doing, the AIs can now let you do a lot of role playing and sort of preview conversations that you're having before you go have them and help you truly refine your thoughts. I mean, I find voice conversations when I'm in the car. I love using voice to just go back and forth with something because I have to otherwise be focused. [40:00] on the road. But when I'm sitting on my couch, I don't often want to talk to my phone or my computer. I often there do just like typing and reading and scrolling. So it's going to be a matter of being able to go back and forth very smoothly with all these modalities. Yeah, I think it's going to be really interesting. And, you know, especially if people use these voice tools to prepare for speeches, if you can prepare with the person. But virtually, that'd be quite cool, right? We just had to re-top on the show, for example. GPT writes as a sparring partner, just bouncing ideas off. And actually, it was pretty good. And I know you and Reid worked very closely together when you're at Greylock. And it was a pretty cool experience. If you have Reid virtually and you're doing a virtual kind of like preview mock-up, that could be quite cool. No, that's great. And I mean, Reed's one of the... the most incredible founders and thought leaders of our time. And, One of the fun things I've seen him doing is building REIT AI, which is actually not only can you just interact with ChatGVT as a sparring partner, but actually building something that has... ingested and learned about him and creating sort of a double for him. Now, it's still very different than having the real live person and interacting in the moment, but it's incredible proxy to understand how he thinks, how he might go about certain situations to test things. And to get really, you know, instead of just reading his writings or watching previous podcasts, this incredible way to synthesize it to at least see how that AI may apply things to new situations. And then when you get to interview or interact with the real read, you feel like you're ready to go have a much deeper conversation because you've actually experienced some of it before. Like, I think these meeting, but not in lieu of. Exactly. It did actually help. We talked about it a ton. So people can check it out. We'll put it in the link in the comments. I also think just like the way these consumer apps are thinking about pricing and how they're thinking about monetizing is very different, right? So many of these AI tools today, they're literally giving all of their products away as cheaply as possible, basically for free, right? Which is very normal just to get as much distribution and attention as possible. Is this still the right way to build a consumer app today? And at one point, does that become a challenge? when they start having to try and make some money so they can make investors happy. This is a really interesting challenge of our time in technology. For many, many years, as I first was getting my start in my career, actually building a website was actually quite expensive. You had to go buy physical servers, get racks at data centers, actually put them up in the data centers, manage them, have people who are managing them. So if you want to just build a website, you actually needed to raise a lot of capital just to fund that. I remember in my first week when I joined LinkedIn, one of the things we did was move from one set of racks bigger set of racks in the data center, and they needed the entire company. I mean, it was maybe 15 or so people at the time to actually go and move computers from one rack to another, just to remind you of the costs. What got amazing in this era of cloud scaling, Amazon Web Services, all these new stacks, was All of a sudden, you didn't need that same set of heavy upfront infrastructure costs. And it became very, very cheap or so cheap that you could scale these consumer things quite far to serve many, many people. And the product you were building mattered more. Fast forward to today, we have costs again, you know, and actually they're variable costs. They're these inference costs. You know, they aren't just even this fixed cost of infrastructure. And so we really need to figure out, you know, what you can give consumers for free, what you need to be able to charge for. But we also know that for consumer distribution, the willingness to pay doesn't always come up front. In the enterprise world, you can sell to a company and sometimes convince them that this is going to be valuable for them. They'll actually pay you. You want a contract, obviously with a lot of product-led growth, you see this sort of get in the hands of employees before they have to do it. But even if they want to pay, if they find it valuable, they can find ways to expense that service. So they don't necessarily need to think about it as paying out of their own pocket. On the consumer side, It is all going to be ways that we need to figure out how to subsidize the inference, having consumers pay for that. On the plus side, With the rise of mobile gaming, the rise of subscriptions, the rise of the world. Consumers are much more used to paying for digital things than they were 10 or 15 years ago, let alone 20 or 30 years ago. They are willing to pay for things that are valued through Apple's App Store, through Google Play, through just online subscriptions with all the different tools out there now. When people hit a certain level of engagement with a product, asking them to pay, isn't the same, you know, hand to the face that it used to be. It's now very much an invitation. If this is finding you value, it's costing us money. On the other hand, we need the cost to come down. with what's going on with AI, with inference, being able to switch models, having open source models. We're seeing just [45:00] It's the beginning of how much cheaper you can actually build these things. And I think over the next couple of years, we're going to intersect... with how much cheaper these can get to build and serve, with how much customers are willing to pay for things that are of value or entertainment for them. And I think we're going to be at the right intersection point where you'll be able to bring in both ad models, purchase models, subscription models, and fund the cost of inference. But right now, it's still a little bit expensive. And as investors, you're looking for Are you going to have enough value that people would be willing to pay or would be, you know, advertisers would be willing to pay to get in front of those customers? And will that intersection actually happen? And that's sort of what the big bet is over the next few years in consumer. But it all starts with a product of value. What is a great consumer product actually look like today and as an investor? when you're looking at all these thousands of companies, are there some characters that just tickle your box? What's the golden list for Josh? Things that I look for in a product are where it has a really clear value that you can describe to somebody. If you start using this product, you get this thing that didn't exist before where people will raise their eyebrows. Now what's cool with AI is there's so many cool new things you can do. It's easy to get people to raise their eyebrows. The second is when that person does go and try it, do they have a good experience? Are they actually going, wow. I can figure out how this fits in my life. And often it is going to replace something they were doing elsewhere. And is it a replacement that is an easy substitution or is it a really hard one? And so – I always like to ask people, what did you stop doing when you started using this new thing? So if they've adopted a new product, what are you no longer doing? And if you can get a good answer to that question, now you know it can intrigue people. you know that it can get them actually trying it and using it. And then, Do they want to spread it? Are they so excited about it that they'll evangelize it to other people? Sometimes that happens naturally through the product with network effects. Sometimes it just happens because, you know, with like Chat2BT, we're all so excited about what you could do. You don't need to Google anymore. Chat2BT just gives you the answer right there. Like that spread very, very fast. And then I look for, why will people stay retained? So not only do you need to have Got me intrigued. gotten me to adopt it, and stopped doing something else potentially, gotten me to share it and spread the word, but now stick. And if I keep engaging with this product over time, now I know, hey, we really are onto something. And one of the best ways to keep somebody using something is with a little bit of a network effect that keeps them and other people using it together. But it's not the only way. Some of it is the more I put in here, the more it does for me, the more I don't want to move my information elsewhere because now I've built up a history. Sometimes it's, you know, if I'm going to one of these marketplaces, or entertainment, these great creators are here and I want to see them keep telling stories. And if I'm there to keep seeing them tell stories, then other people want to be there to get in front of the audience too. And then you'll have this great effect where all the great creators will want to be on the same platform to reach the audience and it will connect. continue to go as a loop. So I really look for that entire process from Is it? intriguing in the first moment when people actually adopt it? Do they really use it and stop doing other things? Do they spread the word about it? And then do they stick long-term? It's never been harder though, right, to grab the attention of the consumer. So when consumers are thinking about, you know, is this thing going to save me time? Is it going to save me money? It's all around these network effects and the value adds, right? How are the network effects different today versus even two, three years ago? I mean, the first thing is actually I think it's never been easier to get consumers' attention. There are so many channels that we're paying attention to, you know, things that, you know, rise up on TikTok or Instagram Reels or YouTube go viral very, very quickly because people get excited about something new. So the idea of having something that as long as it's so compelling, people want to check it out. I've seen people go really, really fast just by getting in front of people. What's harder, I think, is to your point, Getting me not just to try something, but to actually stick. It has to be such a good product doing that one thing really well out of the box that I – either stop doing something else or I turn more behaviors over to it. And we have a lot that we already have in our digital life. And so we already have a lot that we have in our real lives. But even, you know, I used to say there's so many things that can move from real life to digital life. And COVID again sort of showed me what that saturation point is. So now as a product, you have to figure out some new wedge into somebody's digital life if you're a digital product. And then becoming so good that it becomes a habit and that you can pull other people in. get More and more and more people. These are very, very hard challenges because you're substituting versus just expanding, which I think 15 years ago, we were all getting our first phones. There were so many things we'd never done before that we could do on our phones. Today, you have to be better than those things. But I think with AI, we have a real possibility to be those products that I'm like, well, that is so much better that that's great. the last thing you didn't talk about that I think a lot in this competition world right now is that's cool, but if someone [50:00] Does it, can't everybody else do it and copy it very, very quickly? And so we do live in that world where all great ideas people are chasing fast, whether it's other startups or big companies. And the hardest part there is, If you're that founder who has the idea and the insight, Even if other people can copy what you did, they can't copy where you're going. They can't figure out where you're going ahead of you. They can't necessarily replicate that. staying on your line and really living up to your vision is sort of the best thing you can do in today's sort of competitive market. This is the hardest challenge for founders, right? The idea to clone people. is no longer months or weeks as long. basically days. But I can only clone what I see. I can't clone where it's going and I can't clone the next thing. And over time, you build that relationship with the user, which is like, yeah, I tried that other thing, but it was missing the essence of something that was really special. And so you stay with the product that really has that essence. And so much of building an AI right now isn't just a product that does things, but that has personality, that has relationship, that has an identity that you actually stick to, that has a way of interacting with you that it can learn over time to be even more personal to you. as a point of view. And so I do think that That while everything feels like there's eminently clonable things, The people who build these durable platforms are finding their way to really connect with their customers and connect with their networks in a way that is really rich and much better. And the hardest part is all of us in the early days is teasing out, you know, which are which. This is just a small observation, but as I've been thinking about all these apps I've been using over the last six or 12 months, they're all based in the US. Why are American startups so good at consumer social versus maybe the likes of Europe? I've got to think more if that's true. I mean, a company like Lovable came out of Sweden, for example. That's just an incredible company. You know, all the things Eleven Lab does, you know, didn't start here in San Francisco. So I don't think it's true that the best ones started here. Musically, for example, started actually as a team in China. They had spent some time in the United States, but they were actually a team in China that was learning. And there's so much innovation that's been happening in the Chinese ecosystem, for example. some of the original growth of live shopping in China. So I do think we're at this wonderful place in the world where more and more innovation is happening everywhere. What's really unique, though, about the Bay Area is, How many people here, maybe not just the founders with that idea, but the next wave of people have truly come and been able to help scale and find those companies and find those paths to independence and find those paths to network effects. And we have so many people in the Bay Area that have been part of some or many of these journeys. I mean, I myself live here and I've gotten to be a part of so many different companies that have made this journey from something that was a really cool idea to a really durable platform. the right expertise even faster. when you're interacting with people from the Bay Area. So whether that's, visiting, whether that's deciding you need to live here or hire people here. I think it's a very independent question for every founder. But I do think, If you are going on this journey, you want that expertise around you, which is part of why, even for me, I'm on the VC side now because I'd love to be able to help as many companies as I can navigate that incredible moment from, We kind of got something working to a really durable company and platform. I also read Hoffman's question, so I'll ask you as well. Is SF still the best place in the world to start the company? I don't think it's the best place in the world to start a company, but I think it's a very important place in the world to scale a company. Because I think if you believe that you have a great idea and you've figured out how to get this into the first set of people, but you want that expertise now to go, okay, how do I truly make this durable? There's just so many people in and around the Bay Area that can help you do that. But I don't know that that always means you have to live here full time. I think it does mean that it helps to be a part of the ecosystem. I also think if you are a B2B company or selling to other small companies, there's just so many here that it's an incredible place to start with adoption. But honestly, if you're building for consumers, a lot of what you're also building for is culture. And I'm not sure that the SF tech culture is 100% transferable to everywhere else either. Everything here, you want to also make sure you really understand the culture of other cities around the world and other communities around the world to make sure that you're really being able to do something that will affect mainstream consumers where they are. So you also don't want to get it too much SF Glasses. You know, I'm a little jealous. You worked at Apple. Robin Hood for the time. You've obviously held the team at X. Facebook, LinkedIn, Musical.ly. Which is your favorite? Oh my gosh. I mean, you can't really pick a favorite, but in my career, I will tell you that I had this period of time that I joined Twitter late 2009. [55:00] And so we kind of got to go through the... to 2010. And it was this period of just, true missionary work, believing that we had this incredible platform at Twitter that could be so much more for the world. And I got to live and breathe it every day with an incredible group of people. And it was at that nascent period that we knew there was something and we were trying to solve really hard problems. But before it was so obvious that it had to become a real mature company. And I just I if I look back, like if I could just have done that year on repeat in my career, kind of late 2009 to late 2010, that was one of the most special times of my entire life. But everywhere else I've gotten to work was just been I mean, I've got to work on things that it was always my dream to help people make things that matter for the world. helped do that. I was never a founder myself. I never went and started a company from scratch, but I was always what I call an early joiner, where I'd go find these things and say, we're going to go make something great. You have an incredible idea that's getting started, has some momentum. How can we make it last? And there's so much joy in that. And the stuff we did at Twitter is probably some of the most meaningful work of my life and and still see it to this day play out. And I still feel like I could even been so much bigger and so much more important than what it is, even though it's a product I still use the most every day. How many users were at Twitter when you left? When I left, we had gotten well over 100 million users. And when I joined, we were at maybe barely 10 million actives. What's the most interesting lesson you learned from Jack Dorsey? You know, what's funny is I didn't actually work with Jack. When I was at Twitter, it was the Evan Williams era. And Evan had founded Blogger and he had really helped build Blogger and he had this vision for how Twitter was going to become this next wave of media with people communicating. The single best lesson. that I remember... from Evan was, We've got a lot of smart people in the building. We've got a lot of smart people who are using our product every day. Let's not just think we know what's best. Let's listen to them. Let's learn. And then when we make decisions, let's make sure we engage with them on our thinking as well. One of the most controversial things that Twitter had launched was the retweet. The idea that, you know, originally, and this is, again, ancient history and sort of the the tech world now. But when the retweet first launched, the way people used to do retweet is it would come from their account and it would say RT and then the name of the person you were retweeting. And instead of building that product, everybody realized everyone wants to retweet and share voices from others, but decided at Twitter that the authenticity of the original poster was more important than switching the words to look like they came from your name. So the retweet became a Josh retweeted, Ollie's post and that you would see Ollie's face in your timeline. And that was one of those both listen to the world on how they wanted to repeat others' voices and amplify others' voices, but also take a very hard line that it was not about moving the words to the retweeter's voice, but actually maintaining the original tweeter's voice. And the retweet and the quote tweet feature that we have now are all part of how it still retains the original person's voice. And there was this incredible lesson on listening, but still having a strong point of view and then communicating that back out and engaging in discussion, even when a lot of people weren't happy with it at first because it changed the behavior a little bit. That's insane. 2009, I was only... 11, 12 years old. So I don't remember all of this, but I do remember creating my X account, Twitter account. And I think what has been useful now, obviously, today is awesome and let's see where it goes next. For the audience, if you're not following Josh, go and follow Josh. Always awesome insights into what's happening in the tech world. We shortly mentioned Tech Week is coming up in SF. So if you want to get a flavor of what's happening in the Bay Area and startup plans, A16Z hosts this awesome Tech Week every year, SF Tech Week. I went last year, there was like 2,000 events or something. And just awesome experience. So hopefully see you there. But Josh, this has been awesome. Your first official podcast since joining A16Z. So thank you so much for coming on the show. Thanks so much for having me. Spotify. Follow us on X at A16Z and subscribe to our sub stack at A16Z.substack.com. Thanks again for listening and I'll see you in the next episode. This information is for 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 LLC. [60:00] 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.