What Does an AI Operations Manager Do, and Should I Hire One?
An AI operations manager owns the AI agents that already run parts of your company: they clear the exception queues every day, update each agent's rules when a price, supplier or policy changes, measure every agent against its baseline each month, and choose the next process to convert. You should hire one once you have 3 or more agents in production across at least 2 functions. Before that you need someone who builds and ships the first agents, which is a different job.
Why companies hire this role too early
The title is new, so most companies copy a job description from a job board. It asks for prompt engineering, vendor selection, an AI strategy, change management and staff training, all in one manager level hire. Then the person starts and there are no agents in production. There's nothing to operate.
They do what's left. They evaluate tools, write a policy, run lunch and learns and start two pilots. I'd put the cost at a full year's salary and 6 to 9 months before anyone asks what's live. The answer is usually nothing, because building agents and running agents need different people.
The opposite mistake is also common. A company gets 4 or 5 agents live with an outside builder, the builder leaves, and nobody owns them. Within a quarter the supplier list has changed, the pricing has changed, and the agents are coding invoices against last year's rules.
Who you need at each stage
I use one test before anyone writes a job ad: count the agents that are live, meaning they act on real work every day without a person redoing it. Here's how that count maps to who you need.
| Agents live | Who owns them | Hire an AI operations manager? |
|---|---|---|
| 0 | An AI operator who builds and ships the first ones | No. Nothing to manage yet |
| 1 to 2, one function | The head of that function, about 2 hours a week each | No. The function head can carry it |
| 3 to 6, two or more functions | A named AI operations manager | Yes. Start the handover now |
| 7 or more, 150+ people | The manager plus one engineer | Yes, and the builder moves to quarterly reviews |
How I set the role up inside a client
1. Write the job as four duties with numbers
Clear every exception queue within one working day. Update an agent's rules within a week of any change to prices, suppliers or policy. Report each agent against its four week baseline every month: hours removed, error rate, cycle time. Convert one new process every 6 to 8 weeks. If a duty can't be measured, it doesn't go in the job description.
The week has a shape too. Monday is the agent report to the COO or CFO. Tuesday to Thursday is queues, rule changes and the next process in shadow. Friday is a one hour review of anything an agent got wrong that week and the rule that would have stopped it. If the manager spends more than half their week on queues after month 3, an agent's rules are too loose and need tightening before you add another one.
2. Hire from operations, not data science
The best people I've put in this seat ran a finance operations or revenue operations team before. They know what a wrong invoice costs and who to call about it. They need to read an agent's log and a simple SQL query. They don't need to train a model. My interview test is a week of real agent logs with 3 errors planted in it. Strong candidates find all 3 in under an hour and tell me which one would have cost money.
3. Overlap with the builder for 4 to 8 weeks
The manager shadows the operator for the first 2 to 4 weeks, then runs the queues while the operator stays on call. Every agent gets a one page runbook before the handover closes: what it does, what it's allowed to touch, what goes to the exception queue and who signs off a change.
4. Report to the COO or CFO
The agents run operations, so the person who runs them reports to the person who owns operations. Put the role under the CTO and it gets treated as an IT ticket queue. Put it under the COO or CFO and the monthly report sits next to the numbers it moves.
5. Give them a hard limit on what agents can do
The manager can change an agent's rules. They can't widen what it's allowed to touch without a second sign off. An agent prepares a payment, a named person releases it. That rule doesn't move when the team gets comfortable.
From my operating seat
Inside one client I run, an agent researches the sales pipeline overnight and the team has a ranked list of accounts by the morning. The person who keeps it right came from their revenue operations team. Most mornings that's 20 to 30 minutes: read the exceptions, fix the two accounts it got wrong, tell me if a rule needs changing. That's the job.
I've spent twenty years in the C-suite, with three exits and three FT100 listings. In every one of those companies the person who chose a system and the person who kept it accurate were different people, and the second one mattered more at exit. Buyers ask who runs this when you're gone. An AI operations manager with a runbook per agent is a good answer to that.
What should an AI operations manager job description include?
Four duties with numbers on them. Clear every agent's exception queue within one working day. Update an agent's rules within a week of any change to prices, suppliers or policy. Report each agent against its four week baseline every month. Convert one new process every 6 to 8 weeks. Leave out vendor research, AI strategy and company wide training. Those belong to whoever owns the plan.
What's the difference between an AI operations manager and an AI operator?
An AI operator designs, builds and ships the agents, and decides which processes go first. An AI operations manager runs them once they're live. The operator's work is heaviest in the first 6 to 12 months. The manager's work starts when there's enough running to manage. Most companies need the operator first, often fractional, and the manager second, often hired from their own operations team.
Can AI agents replace my operations team?
Agents can take most of the repeat work: matching, coding, chasing, routing, first drafts. They can't own exceptions, judgement calls or the relationships with suppliers and customers. In a 100 to 200 person company I'd expect the operations team to stop growing while the company grows, and the people in it to move from doing the tasks to checking and improving the agents that do them.
Where I come in
This is what I build and run inside companies. I come in as the AI operator, ship the first agents one process at a time, then hire and train your AI operations manager and hand the system over with a runbook per agent. If you have agents live today, who owns their exception queue tomorrow morning? See how I work as your AI agent operator, or start at meethayat.com.