The Prototype-to-Production Field Guide
How founders take an AI prototype to a production product — architecture, costs, timelines, and the traps.
Why prototypes stall
Every week, thousands of founders get an AI prototype working in an afternoon — and then spend six months discovering why it isn't a product. The gap isn't intelligence or effort. It's that a prototype answers "can this work?" while a product answers "can this work for ten thousand strangers, unattended, at 3am?"
This guide maps the terrain between those two questions: what actually changes (auth, data, billing, observability), what it costs, and where founders predictably lose months.
The production gap, itemized
Real authentication and account lifecycle. Real data modeling with authorization enforced at the database. Payments and the tax/receipt/refund machinery around them. Observability — logging, error tracking, alerting — so problems surface before users report them. And the unglamorous 40%: edge cases, empty states, retries, and the mobile keyboard covering your submit button.
None of this appears in a prototype, and all of it is why "it already works" and "it's ready to launch" are different sentences.