Launching an AI feature is not the end of the work. It is the point where you start learning from real use.

Before launch, you test against the examples you already know. After launch, users show you the examples you missed. They ask questions in unexpected formats, use different product vocabulary, paste messy data, switch languages, hit unclear tool behavior, and reveal gaps in your evaluation set.

Iterating in production means improving the system from real usage without experimenting carelessly on users. The practical loop is to collect signals responsibly, turn them into hypotheses, test changes against offline evals, release through controlled experiments, and measure whether the product actually got better.

This chapter shows how to run that loop.