Skip to content

mike-diff/agent-harness-overview

Repository files navigation

harness / agent

A small, visual explainer that answers one question: what is an agent harness, and how is a coding agent built from one?

What this is

A short, illustrated essay site. It takes the idea behind every AI agent and pulls it apart into its pieces, in plain language, one page at a time: the loop at the core, the tools wired into it, the context it works inside, the memory it keeps, the prompt that steers it, the gates that keep it in check, and the way one agent can run a team of others.

It is built as a warm, magazine-style long read rather than a wall of API docs. The aim is for someone with no background to come away understanding how these systems are actually put together, and for someone who already builds them to find the mental model stated cleanly.

Why we built it

The language around AI agents has blurred together. "Harness," "agent," "coding agent," and "framework" get thrown around as if they were the same thing, and a genuinely simple idea ends up sounding like jargon.

The simple idea is this: a harness is a reusable engine, a loop plus a handful of primitives, and a coding agent is just that same engine with coding tools wired in and a codebase to work in. Swap the tools and the very same core becomes a research agent or a trading agent. Change how you run it, all at once or as a live conversation you watch unfold, and nothing about the core changes at all.

That is the thesis the site exists to make obvious: composition, not inheritance. One core, wrapped two ways, many products. We built it because the clearest way to show that an idea is simple is to draw it.


Built as a static site with Astro and Tailwind CSS.

About

A visual, plain-language explainer of what an agent harness is, and how a coding agent is just one thing you build with it.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors