TL;DR
The short version
Claire Vo argues that when AI makes execution nearly free, classic feature roadmaps become dangerous: they clear every backlog item, converge on competitive parity, and churn without learning — mediocrity at machine speed.
Replace date-stamped lists with durable 1–2 year convictions, a fast factory aimed at reality, disposable features, honest probe/experiment/promise labels, and an ambition scoreboard (huge swings per month) instead of raw PR velocity.
Based on Claire Vo's talk on Lenny's Podcast (published 24 Sep 2026; recorded at Lenny & Friends Summit).
The bottleneck moved from build to believe
Vo confesses she's shipping more than ever — agents, MCP, skills, dozens of bots — and she's out of good ideas. Execution outran discovery of meaningful, commercializable products. Engineering scarcity used to force cut-lines that quietly killed weak ideas. Cheap tokens remove that filter.
Now we can ship all our bad ideas. Like congratulations to us.
Claire Vo
Three traps at machine speed
When a classic roadmap meets an AI factory, three traps appear: the backlog trap (clearing requests isn't progress), the parity trap (same inputs and taste skills produce the same obvious product), and the churn trap (ship, see noise, abandon, never compound). From the inside it feels productive. From the outside it accelerates toward mid.
Roadmap zero — and what replaces it
"Roadmap zero" is not an empty backlog. It means everything on it is buildable, so effort-based prioritization stops being strategy. The remaining constraint is truth: customers, data, repeated tests, a unique point of view.
Build convictions for one to two years out — including what evidence would prove you right or make you stop. Keep the factory for fast contact with reality. Allocate tokens toward proofs. Practice durable convictions with disposable features: good stubbornness revises the solution; bad stubbornness moves the goalposts because tokens are cheap.
Label ships as probes, durable experiments, or promises. Vo feature-flagged her ChatPRD Product Graph off until she believed it deserved attention — competitive and near-oneshot, but not differentiated enough in her judgment. Code is abundant; customer trust is not.
Next year is the ambition game
The last 12–18 months trained velocity. Useful muscle. Wrong scoreboard for what's next. Ask how many huge experiments you run a month, expecting most to fail — not how many PRs you merged.
How many big big swings are you taking a month with the presumption that most of them won't work out?
Claire Vo
Write your last roadmap: not your last plan — your last feature-and-date spreadsheet. Same-summit companion: Dan Shipper's labs/product split answers who explores; Vo answers what the plan artifact is for once exploration is cheap.
