3 AI as a partner and problem

A senior engineer is reviewing a pull request. The code looks promising. It has a clean structure, comprehensive error handling, and well-named functions. Tests pass. The implementation matches the API contract. She approves it and moves to the next PR in her queue. Three weeks later, the on-call engineer gets paged. The endpoint is returning 409, a conflict for cancelled orders. The product team is confused. They never specified that cancelled orders should be treated differently. The engineer who wrote it can't explain where the rule came from. She prompted the AI to "implement the endpoint," examined and accepted the output, and shipped it for review. The 409 response looked reasonable, so the reviewer approved it. Nobody decided it should work that way. It just appeared. The problem isn't that the code is broken. It works exactly as written. The problem is that nobody chose this behavior. The team is now maintaining the business logic that was invented by autocomplete.

3.1 Keeping the human above the loop

3.1.1 The problem with being in the loop

3.1.2 Positioning the human above the loop

3.1.3 What this means in practice

3.1.4 From collaborator to power tool

3.2 How AI helps you stay lean

3.2.1 The API service example

3.2.2 The Lean principle at work

3.3 Where AI creates waste

3.3.1 The unconstrained prompt

3.3.2 Breaking down the waste

3.4 Establishing your AI standard work

3.4.1 What the API service team established

3.4.2 Making AI boundaries explicit

3.4.3 Standard work as flow protection

3.5 Exercise: Compare handwritten vs AI-generated code

3.6 Summary