How We Actually Use AI to Build Software

Clients ask whether AI writes our code. Partly, yes — and being straight about how is more useful than pretending either that it does everything or that we don't use it.

Where it genuinely helps

• Boilerplate — forms, CRUD screens, repetitive test cases
• Reviewing a change for edge cases a tired reviewer misses
• Explaining unfamiliar code we've inherited from another team
• Writing documentation nobody would otherwise write
• Drafting the second language of bilingual content

Where it makes things worse

Architecture decisions, anything touching money or permissions, and any code where being subtly wrong is worse than being obviously broken. Generated code that looks right and behaves wrong costs more to find than code that simply fails.

The rule we hold to

Nothing reaches a client's system without a person who understands it reading it line by line. AI changes how fast a developer works. It doesn't change who is responsible when something breaks at 2am.