Decide whether to keep an agent that fixes CI

You wired an autonomous agent into CI. When a test fails, it reads the repo, edits code, reruns the full 6-minute suite and loops until the build is green, up to 10 times. It does fix some failures. But CI cost has tripled, and median time to PR feedback went from 8 to 27 minutes.

Do you keep it? Decide what you measure first and how you would reshape it.