Nobody does it correctly—ai or not.

It’s about breaking the problem down into epics, tasks, and acceptance criteria that is reviewed. Review the written code and adjust as needed.

Tests… a lot of tests.