How to Make AI Write Tests That Don’t Suck!

Models write terrible tests? Many developers believe that. But in my experience the opposite can be true. In this video I show what made AI-generated tests work surprisingly well in several real projects. The key is not the model, but how you structure your testing approach. We look at acceptance-level tests, clear guardrails for AI, good examples, and proper test infrastructure. If you are using GitHub Copilot or other AI tools for coding, these ideas can help you build reliable regression test suites instead of fragile unit tests. Learn how to make AI-assisted testing actually useful in modern software development. 0:00 Intro 0:20 Vibe Rewrite from JavaScript to Rust 1:21 Cornerstone - Acceptance Tests 4:07 Cornerstone - Test specific Guardrails 4:41 Cornerstone - Examples 5:44 Cornerstone - Test Infrastructure 6:50 Conclusion