AI can turn a prompt into a working MVP in an afternoon. The real bottleneck isn't building anymore — it's deciding which of those MVPs actually deserve your team's sprint time.
I kept dropping AI-generated prototypes into sprints and watching half of them stall. So now I run three quick checks before anything earns a spot on the board.
1. Does it survive a cold read?
Hand the MVP to someone who wasn't in the prompt loop. If they can't say what problem it solves in one sentence, the scope isn't real yet — it's just plausible-looking output.
2. Is there a falsifiable success metric?
"Users will like it" is not a metric. "40% of trial users finish onboarding without help" is. If you can't write the number down before the sprint, you can't tell whether the sprint worked.
3. What's the smallest thing that would kill it?
Name the single assumption that, if wrong, makes the whole thing pointless — then test that first, not the easy parts.
Only after all three pass does the MVP get sprint time. Everything else goes back to the prompt.
I eventually wired these checks into a repeatable prompt -> MVP -> validation-gate workflow so I stop relitigating them every week. If you're curious how it's structured, I've been building it out at https://nxcode.io.
What checks do you run before an AI prototype earns real engineering time?
Top comments (0)