The Canary Test: Run AI-Generated Code in a Sandbox Before It Touches Your Repo

You wouldn't deploy to production without staging. So why do most developers let AI-generated code land directly in their working branch? I started running every AI code suggestion through a canary test — a quick, isolated validation step — and it's saved me from shipping broken logic more times than I can count. The Problem AI coding assistants are fast. Too fast, sometimes. They'll generate 200 lines that look correct, pass a quick eyeball review, and then blow up at runtime because of an edge case the model didn't consider. The temptation is to paste the output, run the tests, and fix what breaks. But by then, you've already polluted your git history and maybe introduced subtle bugs that don't trigger test failures. The Canary Test Workflow Here's the workflow I follow for any AI-generated change larger than 10 lines: Step 1: Isolate # Create a throwaway branch git checkout -b canary/ai-$(date +%s) Step 2: Apply + Validate Paste the AI output. Then run this checklist: - [ ] Does it