Managing AI agents is still a full-time job

Does this sound familiar? You sit down to make something and end up refereeing AI agents or concurrent tasks. One agent is drafting the copy, another is turning your outline into slides, a third is writing code or pulling together the numbers you asked for. You didn’t produce any of it yourself. Your whole day is reading it over, sending it back, and catching the thing that’s wrong before it goes out and prompting for an update. By late afternoon you’re fried, and if someone asked what you actually made today, you’d have trouble answering.

Brian Fox, the CTO of Sonatype, wrote about this for Fast Company last week, and his thesis is an interesting one to consider. When the machine does the producing, you are not the maker anymore. You’re the person keeping an eye on the maker machine, and that turns out to be a job of its own.

The team that never goes home

The best comparison he uses is managing people. It’s tiring, sure, but at least it’s spread out. Someone asks you a question or you meet and plan a project, agree on a direction, and everyone goes and does their work. Meetings are on a calendar. The whole process has friction built into it, and the friction is where you get the space to think.

Agents don’t come with any of that. It’s closer to your entire team pulling their chairs up to your desk and asking for feedback every few minutes, forever. The work output got exponentially faster, but your judgment still runs at the speed it always did, and the little gaps between decisions, the ones where you used to catch your breath, are gone.

He likens it to listening to a podcast at 2x. It’s fast, it feels great, and after a while you can feel your brain warming up. You can keep that up for a stretch and tell yourself you’re being efficient. What you have actually stopped doing is thinking. You are mostly just reacting to whatever the screen puts in front of you next.

And it is oddly hard to walk away, because the interruptions are useful. When an agent hands you a finished draft or catches a real mistake before you do, that is not a notification you can wave off the way you might wait before replying to a Slack message. It arrives with a small feeling of progress attached, which is exactly why you keep iterating on it long after you should have stopped.

Your attention is the bottleneck now

For most of the history of knowledge work, making the thing was the hard part and paying attention was the cheap, abundant part. Now the machine makes the thing in seconds, and the scarce part is you: your ability to understand the problem, weigh the tradeoff, and decide whether what came back is any good.

An agent can produce the work, but as Fox notes, it can’t be the one who answers for it when it turns out to be wrong. Someone still has to know what’s going on and make the call. So the real limit on how much all those agents can do isn’t computing power. It’s how much machine-made mess one person can wade through before their judgment starts to wane.

If you work this way, that should change how you think about your own attention. It’s not a luxury for after the urgent stuff is handled, but rather one part of the job the machines cannot do for you. Everything they produce runs back through your judgement, supported by your attention. So it’s worth protecting.

The burnout that looks like a great quarter

The old kind of burnout was typically self evident. Long hours, a full calendar, the obvious grind. This version is sneakier, because from a distance it looks like a good week.

The dashboard is green. Output is up, tasks are being completed, drafts keep being created. Up close, though, your day was eight hours of half-finished thoughts, each one tugging at your sleeve. It is, in Fox’s phrase, burnout built out of open loops. A few of them isn’t an issue, but a few hundred of them back to back will leave you hollow by the end of the day, even though nothing about the day looked on the surface hard enough to cause that exhaustion.

And it can spill over into relationships. Someone running on empty has less patience for the human parts of work, the mentoring and the honest disagreement and the actually caring about how something turns out. That cost shows up eventually, usually somewhere you weren’t looking, like in your relationships with colleagues and employees.

What you can actually do about it

The advice he gives leaders is sound, and it mostly amounts to building some structure around the person doing the reviewing instead of leaving them as an always-open set of eyes and ears. Group reviews into batches. Keep creating, reviewing, and finishing as separate stages rather than one endless blur. And protect deep work more fiercely, not less, since more output from the machines means you need more time to make sense of it, not less of it.

Fox’s advice is aimed at bosses and companies, but is applicable across anyone’s workflow that outputs AI results that need human review. You can build your own personal version starting today, and the first move is concrete: decide when the agents get to interrupt you instead of letting them do it the second they finish.

Give yourself real review windows instead of living inside a long open loop. Let the agents run for a set stretch while you deliberately don’t look. Then open the door, work through the pile in one pass, make your decisions, and close it again. You set the rhythm, rather than a queue of eager bots setting it for you.

This is exactly the kind of fragmentation Freedom is built to manage. The idea behind it is simple: you get to decide where your attention goes, instead of being compulsively pulled wherever the loudest thing is happening. Freedom blocks the sites and apps that keep tugging at you, across all your devices at once, so for a set stretch the pull just is not there. Start a session and things go quiet. No Slack, no inbox, no “quick check” that somehow eats forty minutes. Those small interruptions are what attention fragmentation is actually made of, and a session takes them off the table long enough for you to do the slow, connect-the-dots thinking the review loop never leaves room for.

If anything, agents make the old advice about protecting your focus more urgent than it used to be. When you were the one producing, a distraction mostly cost you your own momentum. Now a distracted reviewer misses a bad element that a machine copied into a hundred placesh. The stakes went up, but the remedy did not really change: block the noise and give yourself some quiet.

A study in PNAS Nexus, later highlighted in The Washington Post, found that cutting off mobile internet roughly halved how much time people spent online and measurably lifted their focus and mood. Freedom users tell us they get back something like 2.5 hours a day. In a job spent supervising agents, it’s important to consider if that time is for producing even more output, or if it’s for the deeper thinking the quality of the work depends on.

Your real job now is protecting your attention

The story was never really about humans getting replaced by AI, but rather that human attention gets more valuable the more automated work piles up around it. Someone has to be awake and have enough mental energy to judge the pile.

That someone is you, and the agents genuinely do not care whether you’re rested or wrecked. They’ll keep producing either way. Whether any of it turns out to be worth keeping comes down to something unglamorous: whether you had enough quiet left in the day to think clearly about what they handed you.

Source: Brian Fox, “The coming burnout from managing AI agents,” Fast Company, July 21, 2026.

Frequently Asked Questions