The Lean Startup by Eric Ries

Build, measure, learn in uncomfortably small loops, and track learning, not vanity.

Three takeaways for AI PMs

• Write the leap of faith before the spec
• Make the MVP uncomfortably small
• Replace vanity metrics with learning metrics: Sign-ups, prompts sent and tokens generated go up with any launch and say nothing about cause.

Do this Monday

• Before the next PRD review, add a short section called leap of faith assumptions, with at most three entries and the evidence that would disprove each.
• In your next metrics review, strike any number that would rise even if the feature were useless, and replace it with a per-cohort outcome.
• Before the next prompt change ships, pull fifty traces and record the baseline error rate by input type, so the change can be judged against it.