This goes over how we built LLM support into Dagger, and more importantly what it took to implementing testing for it.