Every agent needs an owner. What that owner actually does

One name per agent, not a team. That person checks it, changes it, answers for it and can switch it off. Someone central can look after the plumbing, but the agent itself needs an owner who cares whether the job gets done well.

By Matthew Steele3 min read28 October 2026

A sales manager at their desk reading a short printed log of an automated tool's weekly actions, pen in hand, coffee going cold
Illustrative scene, not a customer's premises.

Why one name

Think about a Thursday afternoon. The agent sends a wrong reply to a customer. Who gets the phone call? If the answer is "the AI team", three people each assume one of the others is on it. If the answer is a name, that person's probably already looking at the log.

The slower failure works the same way. Your quoting process changes in March. The agent you built in January keeps quoting the old way – correctly, by its own rules – for weeks. An owner who runs the quoting team notices on the first morning. A central team notices when a customer complains.

The owner's card

The owner's cardA card listing the six duties of an agent's owner: checks a sample weekly then monthly, owns the rules and the prompt, approves changes after running the saved test cases, knows the off switch and has used it, reports a line a month, and hands over before leaving. At the bottom: one name, not a team.Owner of: the quoting assistantChecks a sampleweekly at first, monthly once settledOwns the rulesnobody changes them without the owner knowingApproves changesafter the saved test cases passKnows the off switchand has used it once to prove it worksReportsone line a month: what it did, what went wrongHands overa new name before the last day, or it is switched offOne name, not a team. The person who most wants the job done well, usually in the department that uses it.
The owner's duties, on a card. If someone can't do these six things for an agent, they're not really its owner.
  • Checks a sample: weekly at first, monthly once it's settled. Reads what the agent did and whether it was right.
  • Owns the rules: the instructions, the prompt, the thresholds. Nobody else changes them without the owner knowing.
  • Approves changes: tests on the saved examples before anything goes live. One source of truth has the routine we'd use.
  • Knows the off switch: can stop it in 5 minutes without raising a ticket, and has done it once to prove it works.
  • Reports: a line a month on what it did, what went wrong and what changed.
  • Hands over: when the owner leaves or changes role, the agent gets a new name before their last day, or it gets switched off.

Where the central role fits

A central person or small team does have a job – it's just a different one.

The owner (in the department) The central role (platform)
Whether the agent does the job well Whether the agent can run at all
The instructions and the thresholds The connections to your systems, and who gets access to what
The weekly sample The log, the backups, the monitoring
The decision to change or stop it Building the change the owner asked for
Reports to the department head Reports to the AI lead and the systems owner

In a company of 30 people the central role is about half of one person's job. In a company of 200 it might be two people. What it shouldn't ever be is the owner of the agents, because it doesn't do the work the agents do.

How many one person can own

Roughly an hour a week each once an agent has settled, more in the first month. A department head with a real job to do can own 3, maybe 5 if they're small. Past that the sample checks quietly stop happening, and that's usually the beginning of the end. If a department needs more agents than its people can own, it probably needs fewer agents – or a person whose job includes owning them.