The last roadmap | Claire Vo

演讲者开篇指出，自上次Lenny's Summit以来，产品管理虽然没有消亡，但已发生了深刻的变化。他们认为，长期以来被视为行业标志性产物的传统产品路线图，现在实际上已经“死了”。这一大胆断言源于产品开发格局的重大转变，这主要是由人工智能和执行能力提升所驱动的。 根据演讲者所述，问题的核心在于执行速度已超越了发现有意义、有价值产品创意的能力。凭借智能编码代理、先进的开发者工具和众多人工智能模型，他们坦言“出货量比以往任何时候都多”——甚至拥有“40个Grok机器人”——但同时却“缺乏好的创意”。瓶颈已从“我能建造什么”转变为“我深信什么才是真正值得建造的”。 过去，工程能力是稀缺资源。产品经理的工作是优先处理想法、安排它们的顺序，并以“拒绝”大多数请求而闻名，充当宝贵开发时间的守门人。这项工作包括筛选掉糟糕的创意，因为根本没有足够的容量来构建所有东西。然而现在，演讲者感到被一种“真正信念”的差距所困扰：快速出货，却不确定代码是否真正重要。 他们通过一个个人轶事来阐明这一点，该轶事讲述了构建一个“聊天PRD的产品图谱”，这是一个利用人工智能为产品经理创建语义产品图谱的洞察引擎。尽管他们快速高效地构建了它，甚至在功能上与竞争对手旗鼓相当，但演讲者觉得它“应该被扔进垃圾桶”。他们的信念很低，因为它没有差异化；它感觉像是一个同质化产品，而不是能给客户带来惊喜或真正有价值的东西。这种经历凸显了，即使拥有强大的构建能力，如果基础理念缺乏信念，那么所有的努力都会感觉是浪费。 演讲者指出了将AI工厂与传统路线图结合所产生的三个主要“陷阱”： 1. **积压工作陷阱 (The Backlog Trap)**：AI会构建积压工作清单上的每一个项目，但清理请求并不能保证有意义的业务或客户进展。 2. **模仿陷阱 (The Parody Trap)**：竞争对手都使用类似的工具和洞察，最终构建出非常相似、缺乏差异化的产品，从而压平了市场创新。 3. **流失陷阱 (The Churn Trap)**：产品被推出，如果最初的接受度很低或反响嘈杂，它们就会在没有适当学习或积累的情况下被放弃，导致精力浪费。 这些陷阱，表面上看起来富有成效，但最终加速了走向“平庸”的道路。这导致了“路线图归零”的概念，即每一个可见的功能都变得合理且可构建。在这种情况下，基于工作量或可构建性的传统优先级方法失去了意义，而作为战略产物的路线图变得危险。一个执行过时路线图的AI工厂，可以将糟糕的判断加速到“机器速度”，使坏主意迅速演变为真正的问题。 演讲者提出了一个新方法，取代路线图，专注于“我有什么足够坚定的信念，可以去尝试并证明？”剩下的制约不再是代码，而是“真相”——真实、落地的客户洞察。AI无法将未经检验的假设变成事实；它需要现实。 建议的新框架包括： * **建立信念 (Building Convictions)**：定义一个长期未来（1-2年），以及真正需要存在的是什么。 * **确定证据 (Determining Evidence)**：清晰地阐明哪些数据或结果能够证明或反驳一个信念。 * **利用工厂 (Leveraging the Factory)**：利用AI驱动的构建能力，“非常非常快地与现实交汇”。 * **分配投资 (Allocating Investment)**：根据实际学习情况持续重新分配资源。 这个新世界要求“持久的信念，但可抛弃的功能”。产品领导者需要是“好的固执”——坚持解决问题并修改解决方案——而不是“坏的固执”——因为廉价的执行使其变得容易而随意改变目标。这也意味着将“出货”的观念从承诺转变为假设。并非每个出货的功能都是承诺；有些是探索或实验。诚实地对待承诺水平（探索、实验还是承诺）至关重要，并记住“代码是充足的，但客户信任不是”。 演讲者最后敦促产品领导者构建他们的“最后一张路线图”，这意味着不再有电子表格中详细的功能列表和日期。相反，他们应该专注于提升信念、拥抱雄心，并为未来几年的成功定义宏伟的创意。目标是培养一个“能让你惊喜的工厂”，它能够在必要时放弃软件，对AI生成好创意的能力保持高标准，并拥抱一个不可预测但可能更好的未来。重点从纯粹的功能速度转向“雄心游戏”——每月能进行多少巨大的、变革性的实验。

The speaker opens by declaring that product management, while not dead, has profoundly changed since the last Lenny's Summit. They contend that the traditional product roadmap, long considered the defining artifact of the industry, is now effectively dead. This bold statement stems from a significant shift in the landscape of product development, primarily driven by advancements in AI and execution capacity. The core of the problem, according to the speaker, is that execution has outpaced the ability to discover meaningful, valuable product ideas. With access to intelligent coding agents, advanced developer tools, and numerous AI models, they confess to "shipping more than ever" – even possessing "40 Grok bots" – but are simultaneously "out of good ideas." The bottleneck has moved from "what can I build" to "what do I believe is actually worth building." Historically, engineering capacity was the scarce resource. Product managers existed to prioritize ideas, sequence them, and famously "say no" to most requests, acting as gatekeepers to precious development time. The job involved filtering out bad ideas because there simply wasn't enough capacity to build everything. Now, however, the speaker feels burdened by a "true conviction" gap: shipping rapidly but unsure if the code truly matters. They illustrate this with a personal anecdote about building a "product graph for chat PRD," an insights engine that leveraged AI to create a semantic product graph for product managers. Despite building it quickly and efficiently, even matching competitors feature-for-feature, the speaker felt it belonged "in the trash." Their conviction was low because it wasn't differentiated; it felt like a parity product, not something surprising or truly valuable for customers. This experience highlighted that even with immense building power, if the underlying idea lacks conviction, the effort feels wasted. The speaker identifies three major "traps" created by combining an AI factory with traditional roadmaps: 1. **The Backlog Trap:** AI will build every item on a backlog, but clearing requests doesn't guarantee meaningful business or customer progress. 2. **The Parody Trap:** Competitors, all using similar tools and insights, end up building very similar, undifferentiated products, flattening market innovation. 3. **The Churn Trap:** Products are shipped, and if initial pick-up is low or noisy, they are abandoned without proper learning or compounding, leading to wasted effort. These traps, while feeling productive on the surface, ultimately accelerate the path to "mid." This leads to the concept of "Roadmap Zero," where every visible feature becomes plausible and buildable. In this scenario, traditional prioritization methods based on effort or buildability lose their meaning, and the roadmap as a strategy artifact becomes dangerous. An AI factory executing an outdated roadmap can accelerate weak judgment to "machine speed," making bad ideas quickly manifest as real problems. Instead of roadmaps, the speaker proposes a new approach focused on "what do I believe strongly enough to go out and try and prove?" The remaining constraint is not code, but "truth" – real, on-the-ground customer insight. AI can't turn untested assumptions into facts; it needs reality. The suggested new framework involves: * **Building Convictions:** Defining a long-term future (1-2 years out) and what truly needs to exist. * **Determining Evidence:** Clearly outlining what data or outcomes would prove or disprove a conviction. * **Leveraging the Factory:** Using the AI-driven building capacity to "intersect reality very, very fast." * **Allocating Investment:** Continuously re-allocating resources based on real-world learning. This new world calls for "durable convictions but disposable features." Product leaders need to be "good stubborn" – staying with a problem and revising solutions – rather than "bad stubborn" – moving goalposts because cheap execution makes it easy. It also means changing the perception of "shipping" from a promise to a hypothesis. Not every feature shipped is a promise; some are probes or experiments. Being honest about the commitment level (probe, experiment, or promise) is crucial, remembering that "code is abundant, customer trust is not." The speaker concludes by urging product leaders to build their "last roadmap," meaning no more detailed lists of features and dates in spreadsheets. Instead, they should focus on raising conviction, embracing ambition, and defining big ideas for success years into the future. The goal is to cultivate a "factory that can surprise you," capable of discarding software when necessary, holding AI to a high bar for generating good ideas, and embracing an unpredictable, but potentially much better, future. The focus shifts from raw feature velocity to an "ambition game" – how many huge, transformative experiments can be run monthly.