Assertion: it is more difficult to review and fix LLM-generated code than it is to write the code correctly yourself. Using AI to generate code is transferring the intelligence of coding from writing to reviewing.
@shac I think this also explains why there’s a wide range in reported benefits: an experienced developer who’s used to handing a lot of implementation over has already internalized many of the skills needed to write a better spec & good approaches for testing.
Add comment