Write Your Last Roadmap: Claire Vo on Conviction, Not Features
Key points
Key takeaways from Lenny's Podcast recording of ChatPRD founder Claire Vo's talk "The last roadmap" at the Lenny and Friends Summit (September 10, 2026, San Francisco):
From engineering scarcity to conviction scarcity. Vo argues the scarce resource has shifted: engineering capacity used to gate what shipped, but with AI coding agents, 40 Grok bots and API/MCP/CLI access she can now build almost anything — and her bottleneck is now believing what is worth building, not building it.
Road map zero. When every visible feature is both plausible and buildable, buildability and effort stop being proxies for what matters; prioritisation stops being strategy, and feature-and-date road maps become dangerous — able to feed bad ideas to an AI factory at machine speed.
Three traps. The backlog trap (AI will build every item on the backlog; clearing requests does not equal meaningful progress), the parity trap (competitors copy each other, talk to the same customers and arrive at the same products), and the churn trap (shipping, abandoning and never compounding learning) all feel productive from inside but, she warns, accelerate the path to mediocrity.
Competitive is not differentiated. She built a product graph for chat PRDs — a semantic insights engine generating an auto wiki for agents — feature-for-feature matching competitors in almost one shot, but consigned it to the trash because it felt competitive rather than differentiated; shipping undifferentiated products, she argues, puts customers' attention and trust at risk.
Durable convictions, disposable features. The new road map should lead with conviction — what you believe the future looks like one to two years out; "good stubborn" means adapting the solution while holding the problem, versus "bad stubborn" moving goalposts with cheap tokens. Features become probes, experiments or promises, with honest commitment levels.
Code is abundant, customer trust is not. AI cannot turn an untested assumption into fact; real customers, real data and repeated tests are required to validate conviction, and a very high quality bar plus a unique point of view are what earn more tokens, time and customer attention.
The ambition game. After 12–18 months of building AI-native velocity, she argues next year is about big swings, not raw feature velocity: two-to-three-week experiments that once took a year, durable convictions with disposable features, and OKRs that ask how many large experiments the team runs a month, with the presumption that most won't work out.
Write your last roadmap. She asks teams to replace feature-and-date spreadsheets with a roadmap of ambition — the future you believe should exist, what evidence would prove you right or wrong, and what earns continued investment — arguing that writing your last roadmap is "extremely fun on the other side".
Read more: Lenny's Podcast (YouTube) · Lenny and Friends Summit
