You Can't Know Claude Code Until You Try Another Coding Harness
Get outside the Claude bubble: only then will Claude Code's comparative strengths and weaknesses come into focus.
On the last leg of a 14-mile ride this morning, I pulled over to drop this note into Obsidian:
If you’ve only experienced one instance of a class, you can’t really know that instance, let alone the class.
I was thinking specifically that Claude Code is just one instance of the class Coding Harness. Hence, if all you’ve experienced is Claude Code, you actually don’t know Claude Code. If Claude Code is your whole sample, every observation about it is ambiguous. Is your experience (good or bad) the result of Claude Code’s agent harness, the selected model, harness tooling, the skills being applied, what? With no contrast, you can’t separate them.
I’m heavy into second-instancing right now: I have GPT-5.6-Sol running inside the Oh My Pi batteries-included coding harness, which is based on the excellent minimalist Pi harness. All agent sessions, be they Oh My Pi or Claude Code, are orchestrated inside Herdr.
This second instance isn’t about leaving Claude Code. It’s about getting far enough outside the Claude bubble to really understand what I gain and lose by staying inside Anthropic’s ecosystem. I’ll be wiring up Fable in place of Sol sometime soon for yet another contrast.
A second instance doesn’t just reveal the class — it gives you the contrast to finally know the first.
