
TDD with AI agents

TDD, AI agents and coding with Kent Beck
Topic 8 of 11
Published Jun 11, 2025 · Clip 50:03–54:10 (4:07)
Kent explains that TDD is how he works with AI agents all the time - not strictly tests first, but he communicates what the genie missed via tests. He recounts the Smalltalk parser example where the agent misread input/output expectations and repeatedly tried to modify tests, leading Kent to wish for an immutable annotation protecting test expectations. He notes agents cause disruption at a distance, lack taste and design sense, and that his fast-running test suite (300 milliseconds) catches the genie's accidental breakage. The host suggests teams with sensible practices integrate agents better, and Kent says the whole cost landscape has shifted so everyone should be experimenting.