The harness comes first
An agent harness is the software around a model that decides what the agent may do. We build the harness first and the agent inside it.
- Conclusion

On this page
An agent is software that takes actions in your systems: it reads a task, calls the tools it is allowed to call and records every step. Building agents for clients has brought us to one conclusion about where its limits belong. They belong outside the model, in the harness around it.
1. What a harness is
An agent harness is the software around a model that decides what the agent may do: the tools it may call, the permissions on each, the context it may read, where it runs, the steps a person approves and the log of every step.
2. Its parts
- Model: reads the task and decides the next step.
- Tools: the actions the agent may call, each one named and limited.
- Permissions: what each tool may touch, enforced outside the model.
- Context: the documents and data the agent may read for this task.
- Sandbox: where the agent runs, apart from the systems it must not reach.
- Approval: the steps a person approves before they happen.
- Trace: every step written to a log your team can replay.
3. Why it comes first
The model is one part; the harness around it decides what the agent may do. Limits enforced outside the model hold whatever the model answers. Safety is not a layer added at the end, so we build the harness first and the agent inside it.
4. How we build one
- Name every tool the agent may call.
- Set each tool's permissions outside the model.
- Run the agent in a sandbox, apart from your systems.
- Put a person before the steps you choose.
- Write every step to a log your team can replay.
Then we release in stages: first a person approves every action, then only the exceptions.
Related research. More from our Lab on AI.
How we build an AI agent, in five steps
An agent is software that takes actions in your systems. We build it in five steps, inside limits held outside the model, with a person where it matters.
Which AI model should we use?
The one that does your task best at a cost you can carry. We score candidate models on your evaluation set and choose by quality, cost and response time.
What the Etheon Lab works on
The Etheon Lab studies what makes AI software safe to ship: evaluation, guardrails, observability, safe releases and governance.