When a business buys AI agents the way it buys software, information technology (IT) picks the supplier, sets up access and signs off security. Then the agents start doing work that belongs to finance, sales or service, and nobody in those teams is responsible for how well it is done.
That gap is the problem. An AI agent behaves less like a system and more like a new member of staff. It takes on a defined job, follows rules, makes mistakes and gets better when someone corrects it.
A system needs maintenance. A member of staff needs a manager.
This piece is for the chief operating officer (COO) deciding who in the business should own the AI agents once they go live.
Why agents fall between IT and the business
IT can keep an AI agent running. It cannot judge whether a supplier query was answered well, or whether a draft quote used the right terms. Those calls belong to the people who own the work.
When the business treats the agent as IT's project, the finance team logs a fault instead of correcting a colleague when something goes wrong. The agent is fixed in its settings, or not at all, and the lesson never reaches the rules it follows.
The result is an agent that runs but does not improve, and a business that cannot say whether it is safe to give it more to do. This is one way a pilot stalls even when the technology works.
What managing an AI agent means
The job is familiar to any good line manager. It has four parts.
- Set the job. Write down what the agent does, what it may do on its own, what it must ask about and what it must never touch.
- Read the record. Each week, look at what the agent finished, what a person corrected and what it handed back.
- Fix the cause. When the agent gets something wrong, correct the result, then change the rule that led to it, so the same mistake does not come back.
- Decide how far to trust it. Widen what the agent may do alone only when its record shows it has earned that. Narrow it when the record says otherwise.
The fourth part is earned autonomy. It needs a person who knows the work well enough to read the evidence, and that person sits in the business, not in IT.
Who the manager should be
The right manager is the person who would manage the work if a person did it. Supplier invoices: the finance manager. Customer queries: the service lead. First quotes: the head of sales.
They do not need to write code. They need three things: they know what good work looks like, they have the authority to change the rules, and they have time each week to read the record. If nobody in the business has all three, the agent is not ready to go live.
IT still matters. It keeps the agents running, secure and connected to the right systems. That is the role facilities and HR play for staff: essential, and separate from reviewing the work.
Where this gets harder than it sounds
Managing an agent takes time. Reading a week's record and adjusting rules is real work. Added to a full job with no time set aside, it gets skipped, and the agent drifts.
One manager can own too many agents. A person who manages agents across three teams is close to none of their work. Keep each agent's manager near the job it does.
Some work crosses teams. An agent that chases approvals touches everyone. Give it one manager anyway, usually the owner of the process, and let others raise problems with that person.
A good month can mislead. A clean record over quiet weeks says little about busy ones. Widen an agent's scope in steps, and check the record after each step.
What stays with people
Managing agents is a human job, and so are the decisions it protects. A person approves what matters: payments, anything a customer sees first, anything that changes a commitment. The manager decides how much the agent may do on its own. Some decisions never move to an agent, however good its record.
Put a name next to every agent
Before an AI agent goes live, write one name beside it: the person who will set its job, read its record and decide how far to trust it. If you cannot fill in that name, wait. An agent with a manager gets better with every correction. An agent without one becomes a system nobody owns.
Futuvara is a creative technology agency. We build, implement and run AI platforms for our partners. If you are deciding who should manage your AI agents once they go live, start a conversation.
