Founders often ask us to “add AI” before they have defined the core loop. That is backwards. Week one of an MVP should answer: who uses this, what action do they take, and what outcome proves the business works? AI is a multiplier on that loop — not a substitute for it.
When AI does belong in v1, we scope it narrowly. A support copilot that drafts replies from your FAQ. A listing assistant that suggests titles from photos. A matching score that ranks providers — with a fallback when the model is slow or wrong. Each of these wraps a clear user moment and can be turned off without breaking checkout.
What we defer: open-ended chat, multi-step agents with tool access, and anything that needs fine-tuning before you have production data. Those are v2 problems, once you know what users actually ask and where automation saves real hours.
The 8-week window still applies. AI features get the same milestone treatment as payments or onboarding — demoable, measurable, and owned by the same engineers who ship the rest of the product.



