Nobody lets go of anyone's hand… oops, I meant AI's

This piece started as a LinkedIn post and got too long to fit there. So I brought it here, where there's room to talk it through more calmly.

I spent my whole career running away from micromanagement, and today I spend a good part of the day reviewing someone's work. The difference is who's on the other side: now it's AI. And keeping that close an eye on things is a lot more tiring than I imagined when I started.

The way I lead

I've always led by giving people room. I stayed close to check that everything was fine, but there came a point where I'd let go and say: go. Micromanaging was never how I worked with people.

That changed a bit when I started leading people from different countries and cultures. Many of them came to me to review every task and every screen before the final sign-off, a very different profile from what I was used to in Brazil, and it took me a while to understand why.

For them, the person who leads is the reference, and going to the leader means going to learn. Asking for my sign-off was a way of learning from me. So I taught, followed along closely and, over time, let go again. Reviewing everything was a phase, and the phase passed.

With AI, review is the process

With AI, human review is still part of the work, and on purpose. I work with human-in-the-loop, the model where the tool executes and a person validates before anything moves forward. AI speeds things up a lot, and the final judgment is still mine.

A single word read out of context can send an answer somewhere else entirely, and that's when the eye of someone who knows the project makes a difference. With the team, review was a phase of learning. With AI, it's part of how the process is designed, and that's where the quality of what I deliver lives.

That eye has a cost, though. Reviewing carefully is heavy cognitive work, and it almost never comes up when we celebrate how much faster AI has made everything.

The fatigue got a name

In March this year, researchers from Boston Consulting Group and the University of California, Riverside published a study in Harvard Business Review that put a name to this: AI brain fry. They define it as mental fatigue from excessive use or oversight of AI tools beyond a person's cognitive capacity.

The survey covered 1,488 full-time workers in the United States, and 14% of them reported it. Tasks with heavier oversight demands were associated with more mental effort, more fatigue and more information overload. And people who reported brain fry also reported 39% more major errors.

A caveat is worth making: the data is self-reported, from a survey taken at a single point in time, run by a consultancy that sells AI transformation. It shows association without proving cause, and the study looks at people who oversee agents, a case close to the reviewing I do every day.

Even so, the error number says a lot. Review exists precisely to catch errors, and in the study the people who are worn out say they make more of them. Here I'm going past what the data proves, but my reading is that human-in-the-loop only really works if the human is whole.

The study also separates brain fry, which is cognitive fatigue, from burnout, which is emotional exhaustion. People who used AI to replace repetitive tasks reported 15% lower burnout. What weighs is the volume of oversight, and the authors themselves recommend limiting how many agents one person oversees at the same time.

For anyone who leads, I think there's a reminder here: reviewing is work, and it needs to count in the time and energy of whoever does it, leadership included. I let go of my team when they were ready. With AI, I chose to hold on, and I'm learning to do it without wearing myself out.

A big hug, and see you next time!

---

If you want to go to the source:

• Bedard, J., Kropp, M., Hsu, M., Karaman, O. T., Hawes, J. & Kellerman, G. R. (2026). When Using AI Leads to "Brain Fry". Harvard Business Review, March 5, 2026 (read on HBR). Authors, affiliations and date. The body of the article is behind the paywall.
• Schreiner, M. (2026). Study warns of "AI Brain Fry" as workers hit cognitive limits overseeing AI agents. The Decoder, March 9, 2026 (read). Definition of the term, the sample of 1,488 workers, the 14%, the 39% more major errors, the 15% lower burnout, the recommendation to limit simultaneous agents, and the caveats about self-reporting and about BCG.
• The Register (2026). Article from March 9, 2026 on the study (read). Second source for the sample, for the 39% (self-reported errors) and for the 15%.