The Agentic Loop: Treating testing as a continuous cycle with specialized agents

And all we need to do is overcome LLMs' inherent reliability problems and prevent the errors from compounding over time. Plus somehow provide the system with enough context on every code change that it can write useful tests which capture the expected range of behaviours. And we definitely know what those behaviours are ahead of time, because we all write code that clearly and unambiguously expresses our human intentions. Of course.

I'm so sick of people sprinkling the word "agentic" on things and pretending that it somehow makes all the hard bits of a problem go away.