Arpit Mishra
Fullstack AI Developer
Socials
Harnesses, environments and terminal-based coding agents
Why a shell — not a bespoke sandbox — is the environment that makes coding agents genuinely useful, and how the propose-execute-observe loop underneath them actually works.
An agent is only as good as the world it's dropped into — what it can observe, what it can do, and what happens when it gets something wrong.
A model only predicts the next token. Everything that turns that into a working agent — context, tools, loop control, and permissions — lives in the harness.
