Every Small Business Is About to Have Agent Bosses
John Abbitt
·
·
8 min read
Every small business that uses AI agents needs someone to supervise them. That person is probably your best frontline employee, and here's what the job looks like.
Picture the office manager at a plumbing company. She knows which customers pay late, which jobs need a confirmation call the night before, and which supplier always ships the wrong fittings. Nobody wrote any of that down. It lives in her head, and the business runs on it.
A few months ago, her boss set up an AI agent to answer customer emails, book appointments and chase unpaid invoices. It works. She's getting home on time for the first time in years.
Now ask the owner a simple question: if that agent quoted the wrong price this morning, emailed a payment reminder to a customer who already paid, or booked three jobs on top of each other, how would anyone know?
When I ask owners some version of that question, most give me an honest answer. They're not sure. And the follow-up question I hear next is almost always the same: "Do I need to hire someone to manage the AI?"
My answer is that the person who should manage your AI agents probably already works for you. In most small businesses, it's the person who knows the work best. That might be your office manager, your dispatcher, your bookkeeper or your lead estimator. Supervising an agent is a real job, and it's one of the best promotions you can give your strongest people, as long as you give them the time, the training and the authority to do it well.
Microsoft already has a name for this job
In its 2025 Work Trend Index, Microsoft surveyed 31,000 workers across 31 countries and introduced a term I think every owner should know: the "agent boss." Microsoft defines an agent boss as "someone who builds, delegates to, and manages agents to amplify their impact."
The research found that 36% of leaders expect managing agents to be part of their job within five years, and 41% expect to be training agents. Microsoft also suggested a new metric that I like a lot: the human-agent ratio. How many agents does each role need, and how many people does it take to guide them? Too few agents and you're leaving value on the table. Too many and you overwhelm the people responsible for checking the work, which is where mistakes and burnout start.
One idea from a Microsoft researcher in the report stuck with me: working with agents is a lot like onboarding a new team member. You don't micromanage, but you need "informed trust."
Informed trust is the whole game. You get it the same way you get it with a new hire: someone who knows the job watches the work, corrects it, and decides when it's ready for more responsibility.
Companies are giving agents job titles and skipping the manager
Here's where most businesses are today. In late October, Vida, a company that builds AI agents for phone and customer service work, surveyed 215 U.S. decision-makers about the AI agents running inside their companies. Sixty-three percent give their agents names. Sixty percent give them formal job titles. Only 41% assign a human manager to oversee them.
The quality numbers tell the same story. Seventy-two percent said output quality and reliability was their top concern, and only 12% actually track accuracy or error rates.
It's a vendor's survey with a small sample, so I'd hold the exact percentages loosely. The pattern matches what we see. Companies are treating agents like employees in every way except the one that matters: nobody is responsible for their work.
That gap has a cost. A Zapier survey of 518 executives this summer found that 38% say unsupervised AI has already cost their company time, money or reputation. Those are companies with formal AI governance policies and at least 100 employees. A 20-person business with no policy at all is in a riskier spot.
Why your best frontline person is the right agent boss
When owners hear "manage the AI," they tend to picture a technical hire. Someone who understands models and prompts and integrations. That person is useful for building the agent. They're usually the wrong person to supervise it day to day.
The agent boss needs to know what right looks like. Your office manager can glance at a booked schedule and see that a certain customer's job can't go on a Friday because they're never home. Your bookkeeper knows that one customer always pays 45 days late and gets annoyed by reminders at 30. An outside AI specialist will never know those things, and neither will the agent unless someone teaches it.
We're seeing this play out with a paving client right now. They lost the specialist who did all of their takeoffs, the measurements and quantities behind every bid, and their estimating nearly stalled because the expertise left with him. We've been building a system that runs takeoffs on their exact calculations. The interesting part is what happens to the role. The person in that seat stops doing every takeoff by hand and starts supervising the system: checking its numbers, catching the jobs that don't fit the pattern, and deciding when it can be trusted. That's an agent boss job, and it goes to someone who understands paving, with a little training on the tool.
This is also why I don't see agent supervision as a threat to your team. The people who are best at their jobs are exactly the people you need checking the agent's work. Their knowledge becomes more valuable once an agent is in the picture, because now it shapes how hundreds of tasks get done, all of them running under their judgment.
What the job actually looks like in a 20-person business
An agent boss in a small business is rarely a full-time role. It's a set of responsibilities added to someone's existing job, and it needs to be written down like any other responsibility. Here's what I'd put on the list.
1. Own the agent's rules. Every action the agent can take gets sorted: it does this on its own, it does this and tells someone, it asks first, or it never does this. The agent boss decides where each action lands and moves things between buckets as trust builds. Anything that moves money, makes a promise to a customer or deletes something should start in "asks first."
2. Check a sample of the work every day. In the first few weeks, that means looking at a real slice of what the agent did yesterday: the emails it sent, the appointments it booked, the invoices it chased. Fifteen or twenty minutes a day catches most problems before a customer does.
3. Track a few simple numbers. How many mistakes did the agent make this week? How many did it escalate? How much time did it save? Remember that only 12% of the companies in Vida's survey track error rates. Being in that 12% puts you ahead of most of the market.
4. Update the agent when the business changes. New pricing, a new service area, a supplier you stopped using. The agent only knows what someone tells it. The agent boss is the person who tells it.
5. Have the authority to pause it. If something looks wrong, the agent boss can turn it off without waiting for permission. A supervisor who can't stop the work isn't really supervising it.
That's a meaningful set of responsibilities, and it's also very learnable. Most of it is judgment your people already have. What they need from you is the structure around it.
How to make room for it
The most common way this goes wrong is that the owner adds "keep an eye on the AI" to someone's plate without taking anything off. The agent boss gets busy, the daily checks slip, and three weeks later nobody has looked at what the agent is doing. That's how you end up in the 38%.
If you're going to put an agent to work, here's how I'd set up the person supervising it.
Give them time. Plan for a few hours a week in the first month, written into their schedule. That time comes back as the agent takes over more of the routine work, but it has to be spent up front.
Give them training. They should understand, in plain terms, what the agent can and can't do, where it gets its information, and what kinds of mistakes it tends to make. Microsoft's research found that 47% of leaders list upskilling their existing people as a top workforce strategy for the next year or so. For a small business, this is where that upskilling should start.
Give them authority. Make it clear to the rest of the team that the agent boss owns the agent's rules. If someone wants the agent to do something new, it goes through them.
Recognize the role. A new title, a raise, or simply saying out loud that this is a promotion goes a long way. You're asking someone to take responsibility for work they didn't do personally. That deserves recognition, and it tells everyone else on the team that learning to work with AI is how you move up in your company.
The questions to ask before you turn on your next agent
If you have an agent running today, or you're about to, sit down with your team and answer these:
Who is the agent boss for this agent, by name?
Does that person know what the agent is allowed to do on its own, and what it has to ask about?
When did someone last look at a sample of its work?
How many mistakes did it make last month, and how do you know?
Can the agent boss pause it today without asking anyone?
What did you take off their plate to make room for this?
If any of those answers is "I'm not sure," you've found your first piece of work, and it has nothing to do with the technology.
Where we stand
At Avolis, we build agents, but we spend at least as much time on the people around them. When we put an agent into a client's business, we work out who supervises it, write down the rules with that person, train them on how to check the work, and stay on the hook until they trust the system and the numbers prove it out. The tool is the easy part, and the person who can tell when it's wrong is what makes it pay off.
Every small business that uses AI agents is going to have agent bosses, whether the owner plans for it or not. The ones that plan for it will pick their best people, give them real time and real authority, and end up with agents they can trust. The ones that don't will find out what their agent has been doing the same way the rest of the 38% did.
If you want to talk through who your agent boss should be, or what your first agent should and shouldn't be allowed to do, reach out to us. It's a conversation we have every week.
Sources
Microsoft WorkLab, "2025: The Year the Frontier Firm Is Born," Work Trend Index Annual Report, April 23, 2025. https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born
Jon Swartz, "U.S. Businesses Fast-Track AI Employees but Lag on Oversight: Survey," Techstrong.ai, November 12, 2026 (Vida Digital Workforce Readiness Survey, 215 U.S. decision-makers, late October 2026). https://techstrong.ai/agentic-ai/u-s-businesses-fast-track-ai-employees-but-lag-on-oversight-survey/
Lane Gillespie, "AI and human approval," Zapier, September 8, 2026 (Centiment survey of 518 U.S. executives, July 1 to 7, 2026). https://zapier.com/blog/ai-and-human-approval/
Ready to make AI work for you?
Book an AI readiness evaluation. If there’s nothing worth automating, we’ll tell you.
