When something goes wrong with an AI agent, the first question is simple: who owns it? In many companies the answer is "the team." That sounds like shared responsibility, but in practice it means no single person is watching the agent, and no single person can say "stop."

Disclosure: I run a small AI agent risk-check service called AgentRisk, so this topic is close to my work.

Why "the team" fails

An agent is not a one-time project. Its model gets updated, its prompts get tweaked, its permissions grow, and the people around it change jobs. Each of those changes needs someone who notices and decides what to do.

When ownership is spread across a group, a few things tend to happen:

  • Nobody feels responsible for reviewing what the agent is allowed to do.
  • Changes to the prompt or model go in without anyone signing off.
  • When something looks wrong, people assume someone else already raised it.
  • When the person who built it leaves, the agent keeps running with no one who understands it.

One name per agent

The fix is small. Give every agent exactly one named human owner. That person does not have to do all the work, but they are the one who:

  • Knows what the agent can access and what actions it can take
  • Approves changes to its model, prompt, and permissions
  • Has the authority to pause or switch it off
  • Is the first call when something looks off

A shared spreadsheet is enough to start. Columns: agent, purpose, owner, backup, last reviewed.

Handling people changes

Ownership has to survive reorganizations. When someone changes roles or leaves, their agents should be reassigned as part of that process, the same way you would reassign their accounts and shared documents. Naming a backup keeps there from ever being a gap.

Do this today

Make a list of every agent you run, and put a name next to each one. If you can't name a person for one of them, that tells you something useful: it is either unowned and needs an owner, or unused and should be turned off.

Questions for you

  • Does each agent you run have a single named owner?
  • Who in your organization is best placed to own an agent: the builder, the business user, or someone in IT or risk?
  • How do you handle ownership when people move on?

If you want to look at how AgentRisk approaches agent risk checks, the page is here: https://rleffke.com/agentrisk