Founders usually arrive with a feature list. The useful question is different: what's the riskiest assumption in this business, and what's the cheapest software that tests it?
Cut to one workflow
An MVP that does one job end-to-end beats one that half-does five. Depth converts; breadth demos.
Buy everything that isn't differentiating
Auth, billing, email, analytics — integrate, don't build. Your custom code budget belongs on the thing only you do.
Instrument from day one
Activation and retention numbers are the real deliverable. A launch without analytics is a party, not a test.
Plan the second version before the first ships
Knowing what you'll build if the test succeeds keeps the architecture honest without gold-plating it.
Eight to twelve weeks is enough — if you're disciplined about what "minimum" means.