Running agents
Governed agents: the people, training and tools that keep them in line
An organization governs its agents when it has people whose job is to keep them under control and pointed at the business, trains those people for the work, and gives them tools that show whether every control is holding.
Managed is about the agents. Governed is about the organization.
A managed agent can be checked by someone who did not build it. That tells you the controls exist. It does not tell you who decided what the controls should be, who is watching them, or whether the agent still matters to the business. Answering those questions is governance.
People who own it
Governance starts with named people. In a small company one person may hold several of these roles. In a large one there is a governance team, a review board, and owners in each business unit.
- Agent owners answer for what each agent is for and whether it is worth what it costs.
- A governance lead writes the policy and sets each agent's risk tier.
- Reviewers approve new agents, changes and exceptions.
- Auditors check the evidence, from inside the company or outside it.
Trained for agents
Agents are not ordinary software. They act on their own, spend money, work with other agents, and behave differently when their model changes. The people who govern them need to know how agents fail: two agents stuck in a loop, authority that grows as work is passed along, a prompt edited by hand on a server, or answers that slowly get worse. Their training covers those failures, the organization's own policy, and how to read the evidence.
Tools that show what is happening
People cannot govern what they cannot see. They need one place that lists every agent and every relationship between agents, with its owner, its risk tier and the state of each control objective: met, not met, or not checked recently. When a control slips, the right person hears about it. When an auditor asks for proof, the evidence is there, and nobody can have changed it.
What they decide
- Purpose. Every agent serves a business objective, with an owner and a date to review it.
- Obligations. Each agent follows the contracts, laws and internal policies that apply to what it does.
- Risk tiers. The tier sets how strict each control must be. An experiment needs little. An agent that moves money or faces customers needs a lot. A relationship between two agents gets a tier too.
- Relationships. Policy says which agents may work together, and authority only ever narrows as work passes from one agent to the next.
- Change. Every update and every rollback is approved, with a reason.
How it fits together
Governance sets the rules. Management enforces them across each agent's life and across the network. Assurance checks the evidence. What the checks find goes back to the governance team, and the rules get better.
Built around the definition
AgentMesh gives a governance team that view. It shows every agent and relationship with its owner and limits, enforces policy at the gates of the Agentic Development Lifecycle, keeps records nobody can alter unnoticed, and puts a kill switch in the hands of the people responsible. What each agent has to show is set out in the Managed Agent Framework.