Hear this article read aloud.
Every growing business reaches the same point. The work outgrows the people, and the answer has always been the same: post the job, interview for a month, wait out a notice period, and spend another few months teaching the new person how things are done here. Then do it again for the next role.
There is now a different answer, and it starts with someone you already employ. The new role is the agent manager: a person who directs a crew of AI agents, checks their work, and owns the result. It is not a layoff plan. It is a promotion for the people you have, and it means the next three hires you were planning may never need to happen.
What an agent manager actually does
An AI agent is software that does a job, not just answers a question. It drafts the post, chases the invoice, books the appointment, runs the tests. A crew is several of them working on one function, handing work to each other.
The agent manager runs that crew. Their day has four parts:
- Brief. They say what done looks like, the same way they would explain a task to a new hire.
- Review. They check what the crew produced, approve it, and send back what missed.
- Set limits. They decide what runs on its own and what waits for a yes: dollar amounts, customers, systems, hours.
- Improve. Every correction becomes training, so the crew gets better at the job every week.
None of that is technical. It is the skill of someone who knows the work well and can explain it clearly, which is exactly the person you already have doing it.
The single-job role grows into the whole function
Look at the roles built around one task or one channel, and each has a bigger version.
A social media manager becomes the person running all of marketing: a crew handling content, social, ads, search, email, and reporting, with them deciding the voice and where the budget goes. A bookkeeper runs finance operations, with agents taking invoices in, chasing collections, reconciling, and preparing the close. An HR coordinator runs people operations. A support rep runs customer experience. An office manager runs operations.
In every case the person keeps the part that needs a person: the judgment calls, the relationships, the approvals over a limit, and the final say on what goes out. What changes is everything underneath that.
Engineering is where this has moved fastest
Software teams are the clearest example. Writing code to a clear spec, writing and running the tests, building and deploying, and watching for errors after release are all work agents do well today, around the clock.
A senior engineer who knows your system becomes the person who briefs the crew, reviews every change, and decides what ships. One agent writes the change. Another reviews it before the engineer sees it. Another writes the tests and runs the whole suite on every commit. Another deploys to staging and checks it. When the engineer says go, it releases, with a rollback ready. After release, an agent watches errors and uptime and drafts the fix when something breaks.
The developers, QA testers, and DevOps engineers you would otherwise hire to keep up become a crew one engineer directs. The engineer keeps the architecture, the security model, the hard bugs, and the final yes on every release. That last part matters: agents make a strong engineer far more productive, and they do not make an inexperienced team safe. The person in charge should be a senior one.
It is also how our own team works. Our engineers ship websites, iPhone and Android apps, and the platform our agents run on with crews doing much of the writing, testing, and deploying, and a person approving every release.
Three new hires, or one person and a crew
Compare the two ways to grow.
Three new full-time seats cost about $292,600 a year in wages and benefits at the national private-industry average of $46.89 an hour, according to the Bureau of Labor Statistics. Then add the recruiting, the ramp-up, the payroll taxes, the equipment, and the time it takes to manage three more people. Each of them covers about 2,080 hours a year, and when one of them leaves, what they knew leaves too.
A crew is scoped to the jobs it does and priced against the hours that work takes today. It is trained on your own past work before it touches anything live. It can work any of the 8,760 hours in a year, and what it learns stays put. The person running it reviews work, not people.
That comparison will not always favor the crew. Some roles need a person, and a badly chosen job for an agent costs more than it saves. But for the routine, repeatable half of most departments, the math has changed.
How to retrain someone into the role
Nobody needs a classroom for this. The best path runs on the job the person already does, in four stages:
- Pair. An agent works beside them on their own work. They see what it gets right, correct what it misses, and learn to write a clear brief.
- Review. The agent drafts and they approve. Their day shifts from doing the work to checking it.
- Direct. They set the limits and handle only the exceptions.
- Manage. They run several agents across the whole function, hand work between them, and own the results.
Each person moves at the pace the agents earn their trust. For more on the people side of the rollout, see our piece on training your team to work with AI agents, and on managing a team of people and agents once there is more than one crew.
Who makes a good agent manager
Not the most technical person in the room. The right person is the one who knows the job best and can explain it: the one who trains the new hires well, notices when something is off before the report does, and cares how the work comes out. That person already exists in most departments, and giving them a crew is the clearest signal you can send that AI is there to make them more valuable, not to replace them.
It also gives them the skill every employer is about to hire for. Directing agents, checking their work, and making them better is becoming a core part of almost every job. The people who learn it first, on your work, are the ones who will build what your business looks like next.
Where to start
Pick one role you were about to hire for, and one person who knows that work well. Scope a small crew around the most repetitive part of the job, train it on your own past work, and let your person run it through the four stages.
We lay out the whole picture, department by department and with a closer look at engineering, on our agent managers page. The crews themselves run on our AI workforce platform, and we train the people who run them as part of every build through our AI training program. If there is a role you were about to post, book a call and bring it with you.



