Glossary

AI agent

An AI agent is software that uses a model, usually a language model, to decide its next step and then acts on it. It calls tools, systems or other agents, with whatever access it holds.

In a run

Each agent signs in as itself and gets a credential for this job only. Every call it makes is checked against its access policy, and every call is recorded, allowed or refused.

Our word for it

An agent that does the work.

What an agent can reach

An agent’s reach is set by what it can call. Give it a broad login and it can do anything that login allows. On Ares it can do only what its permission allows for that job, and the permission expires when the job ends.

When one agent hands work to another, the permission narrows. It never grows. A downstream agent never holds more than the one that called it.

An agent can also be talked into a step by the content it reads. A supplier document with a hidden instruction is read at the door, before the agent sees it.

Agents you already run

If an agent already works for you, keep it. Ares enrols existing agents without a rewrite: see A2A for how they connect. The agent’s code does not change. Its limits apply from the next run.

Where it stops

Ares does not find agents you have not enrolled, and its guarantees do not cover them. It also does not test whether the agent’s answers are correct.

Evaluate Ares with one of your agents, or watch agents work in the sandbox.

Related terms

  • Agentic AIAgentic AI is AI that takes actions toward a goal, not only answers questions.
  • Agent evaluationAI agent evaluation is testing whether an agent does its job correctly and stays inside its limits before it acts on real systems.

Tell us about one workflow.

We will say whether an agent should do it, and what the first step costs.

A senior member of the team replies within two business days.