AI agent glossary
Agent harness
The software wrapped around a language model that runs the agent loop, supplies tools, manages context, and enforces permissions.
A model on its own only turns text into more text. The harness is everything else that makes it an agent: the loop that sends requests and runs tool calls, the tool implementations themselves, the system prompt, the logic that trims or compacts context, the permission checks, and the interface you interact with.
Coding agents such as Claude Code, OpenAI Codex, and Gemini CLI are harnesses around their respective models. Agent SDKs give you a harness you can customize, and a hand-written agent loop is a minimal harness you built yourself.
The harness matters as much as the model. The same model can perform very differently depending on which tools it is given, how results are formatted, and how context is managed. When an agent underperforms, the fix is often in the harness rather than in the prompt or the choice of model.
Two products built on the same model feel different because one harness gives the model a file search tool and runs your tests automatically, while the other only lets it read files you paste in.
Learn more in the course
Related terms
Go from definitions to doing
Learn to use and build AI agents in 65 practical lessons.