Comment by rconti

> At some point, execution speed starts to matter more than the elegance of the code.

This works better in greenfield deployments than in existing ones. We all know AI "hallucinates". This is reasonably easy to work around on new code where you can take any number of paths to get the right answer.

In existing deployments, being flat-out wrong has a greater cost. I've worked with AI in a ton of examples (working in existing codebases, troubleshooting system problems, etc) where it simply wasted my time, and it would have been better to solve the problem myself.