A great way to use this would be to create very good tests (manually) and then let the AI write the code. Maybe even with a feedback loop: when a test fails, the AI automatically tries a different approach.