A small SaaS team bumps a dependency on Monday morning and 14 tests go red. Nothing is seriously broken. A helper now rounds a little differently, and someone gives up the morning to fixes that are dull, careful and easy to get slightly wrong.
That morning is a good job for an AI coding agent that fixes failing tests. It takes the first pass while the team does something else, but only if "done" means the tests really pass and a person still approves the merge. An agent that says it's finished, or quietly edits the tests until they agree, costs more time than it saves.
By the end of this guide you'll have a short Node.js script that fixes a failing test, refuses to stop early, can't rewrite your tests, and opens a pull request for someone to review. It runs on a cloud model or on your own machine.
