exe.dev essay: an agent is a model, a loop and a tool set

What the research found

exe.dev essay framing an agent as model, agentic loop and tool set, with the loop and tools together called the harness; it covers where models, loops and tools run (local vs hosted, remote MCP) and the cost and security tradeoffs. Conceptual, no new release.

Claims and how well the evidence supports them

1. An agent consists of a model, an agentic loop, and a set of tools 99%
2. The harness is the combination of the loop and the tools 98%
3. Remote MCP servers can expose tools running elsewhere 99%

Evidence

• blog.exe.dev https://blog.exe.dev/topology

Actionability rated low.