My experience with harnesses is entirely about UX, personally. You could just use an LLM directly and pipe its output directly into your source files, but that

My experience with harnesses is entirely about UX, personally. You could just use an LLM directly and pipe its output directly into your source files, but that would produce terrible results in practice. Harnesses / agents are just better versions of “curl https://llm.com https://llm.com > source.{py/js/cpp/etc}”, imo Long term I’m bullish on an open source harness “winning” the foot race, in a similar way that Linux “won” over Windows and MacOS (that is, debatably)