AI makes the first version of a feature cheap. It does not make the hundredth user, the leaked API key or the Friday-night outage any cheaper. Code written with AI fails in recognisable ways — missing authorisation checks, tests that assert nothing, dependencies nobody chose on purpose, error paths that were never run — and most of those failures are invisible until real traffic finds them. This course is about finding them first.

Eight modules take an existing AI-built SaaS — a booking tool with accounts, payments and an admin area, supplied as a repo — and make it production-grade one layer at a time. You write the tests the AI skipped and learn to prompt for tests that fail when they should. You run a security review with a checklist built for AI-generated code, fix what it finds and wire the checks into CI on GitHub Actions so they run on every pull request. Then come observability, performance and cost: error tracking, logs you can search, slow queries, and the model bill that grows quietly with every user.

The last module is for teams. Once several people build with AI on the same codebase, the problems are about consistency: shared rules files, review standards, who owns which part of the code and how new people get up to speed. Priya Raman has set these up for product teams moving to AI-assisted development, and the module ends with templates you can drop into your own repo.