Agent governance: how to set limits and approvals for AI agents

Useful agents need boundaries. Governance is what lets a team trust them with real work.

Start with a role

In Constella, you write agents in plain English, each with a role, expertise, and guardrails. A good definition says what the agent is for, what it knows, and what it must not do.

Set limits

• Scope: which knowledge the agent can draw on.
• Actions: what it may do, and what it may only propose.
• Permissions: access that mirrors your existing rules. See trust and controls.

Add approvals

For anything that matters, such as sending a message to a client, a person should approve first. Approval steps for these actions are on the way. Read why sending waits for a manager.

Keep humans in charge

People set goals and limits; agents work within them. Constella doesn't promise fully autonomous AI, and that is deliberate.

For larger organizations

Enterprises can pair governed agents with options like self-hosted models. See Constella for enterprise.

Watch the walkthrough: Agent governance: limits and approvals.