TDD with AI Coding Agents: The Workflow I Actually Trust

Combining the speed of AI coding agents with the predictability of test-driven development. AI coding agents can write a lot of code, fast — but not necessarily code you can trust. In this video I share the workflow I use to get them to write production code I'm happy to own: test-first, one step at a time, with me staying in control of the process. In this video: Why letting an agent write all the tests at once goes wrong How test-driven development gives an agent the feedback it needs The three things that make agents actually follow the discipline A live walkthrough of the red-green-refactor cycle with an agent Using a test list, a CLAUDE.md file, and a few simple skills to drive it Building a small app in Go — a language I've never used — and trusting the result Whether you're already practising TDD or just curious how to use AI coding agents without losing control of your codebase, this should give you a workflow you can try straight away. do-nothinger on Github https://github.com/douglaswaugh/do-no... — I'm Douglas Waugh. I've spent nearly 20 years in software engineering, around 15 practising Extreme Programming and TDD, and a decade teaching it. I also run training and consulting for teams adopting these techniques: → https://omnicorp.works I'd love to hear how you're getting on with AI coding agents — the good and the bad. Drop a comment below. — CHAPTERS: 0:00 Intro 0:30 The problem with AI agents 2:38 What changed my mind 3:46 A quick TDD primer 5:10 Why agents won't just do TDD 6:37 Getting agents to follow TDD 8:05 The workflow in theory 10:47 The workflow in practice (demo) 12:53 The start of the TDD process 13:11 Test: display different step name 17:32 Test: step complete 20:42 Using skills to improve flow 21:14 Test: multiple steps (using skills) 23:46 Thoughts and conclusion