A Rails testing agent needs to run what it writes

Maxime Langelier and Mathis Grosmaitre described a Mistral agent that reads Rails code, creates or improves RSpec tests, and checks them in CI. Their workflow handles different file types and can run work in parallel. The emphasis is on executing tests rather than judging their appearance.

Context

This is a useful example of automation with concrete feedback. Coverage alone, however, does not establish that a test is valuable. A generated check may mirror the implementation or avoid a difficult case. Review should ask which real regression the new test would detect and whether it fails for the right reason.