Testing is the case I've found AI actually useful. Write one good test, maybe happy path, and then tell your AI to test for scenario XYZ and how it should fail, etc.

The generated code is in general 90% there.

This allows me to write many more tests than before to try to catch all scenarios.