Testing in an agentic coding world
I’ve been writing a lot of code with agents recently. I’ve even written about a lot of it here. However, I began to notice a pattern about testing. My LLM is great about writing tests to go along with their code and it always ends up with a 100% pass rate. Most of the time, the LLM will modify their tests when they do fail the first time. But what am I really testing here? Agents write code to get to completion as quickly as possible. It’s built into their DNA - please the user as fa...