Post

> In part, it is near impossible to get AI/agents to do actual TDD, particularly writing the tests first. This. I have not seen LLMs writing tests first and then start the implementation, even after hard prompting. LLMs will happily write decent tests after implementing though. Related ->

We're all still working it out. However Adam Thornhill (author of code as a crime scene, founder of CodeScene which is an excellent tool, long time XP/TDD practitioner) is the kind of person I will pay attention to.

Lots of slower ones (as we climb the test pyramid). The more cumulative functionality the more of those and their execution time scales the worst.

...most of your tests are very fast low level ones, and with e2e tests spending a lot of effort making sure they are as decoupled as poss, fast, reliable etc

And Adam's approach is still very much "test-first" - driving each change from a failing test at the boundary and letting the implementation follow is much the same shape as ATDD/BDD/outside-in. What's gone is the inner loop, where the unit tests were the design tool as well as the verification.