Working with humans instead of agents is rough.
Asked claude to analyze the last month of PRs for code review comments to find patterns I could write skills on.
"Diagnosis: The cursor-bot caught correctness issues, human reviewers caught taste/hygiene issues"
The humans
Just had to do actual computer science. 🤮 I'm self taught, I'm not supposed to need those things.
Needed a sparse data structure to represent test status with retries across commit/time.
The way people retry CI actions is wrong. You don't want your test runner to retry. Let the pipeline fail and track the failure. Then automate the retry as a new run.
Track how often its failing and get the logs for it. Don't use things like pytest-retry or nick-fields/retry.
DX isn’t just pretty docs and fast builds. It’s trust. Flaky tests break that trust. You lose all the speed when you are debugging and retrying tests.
Why are all the Developer Experience teams building stuff that doesn't improve the daily life of the people they are serving?
