🔗 100 percent code coverage is not as good as you think

When I see people talk about code coverage, there are usually caveats that you don't need to worry about 100% coverage. Some argue that it ends up being a victim of Goodhart's Law1, others say that you get diminishing returns. Some will say that full code coverage is just a by-product of the proper development process, which should be your focus. Others will point out that some things just don't need test coverage. "Be pragmatic!", they'll say.

But lately, since AI agents have become more common, I do see people starting to shoot for that perfect 100%. Some feel that since they can just throw AI at it, they might as well ask for it. Others think of it as a hedge against the unexpected things their agent might do, that it's just the kind of guardrails you need in this kind of workflow.

There's a fundamental point that's missed here.