What AI oversight is doing to your team's brain
May 20
Empty space, drag to resize
BCG studied 1,488 workers earlier this year. They found that the people most likely to burn out from AI tools were not the ones who resisted them. It was the eager adopters - the ones who took up the tools fastest and used them most.
More specifically, the workers at highest risk were those whose main job had become reviewing and correcting AI output. Checking. Catching errors. Staying alert for what the tool missed. BCG gave this pattern a name: "AI brain fry." Workers in this mode showed 33% more decision fatigue and 39% more major errors than those whose AI use looked different. Their intention to quit was 39% higher too.
More specifically, the workers at highest risk were those whose main job had become reviewing and correcting AI output. Checking. Catching errors. Staying alert for what the tool missed. BCG gave this pattern a name: "AI brain fry." Workers in this mode showed 33% more decision fatigue and 39% more major errors than those whose AI use looked different. Their intention to quit was 39% higher too.
Why the eager adopters burn out first
To understand why this happens, you need to meet Corty.
Corty is what I call the prefrontal cortex - the part of the brain responsible for focus, decision-making, and holding complex problems in mind at the same time. She's the part that makes you good at your job. She also has a limited energy budget.
When Corty is asked to monitor AI output - stay alert for small errors, evaluate whether the tool's answer is actually right, second-guess suggestions before they go out - she doesn't get to rest. She's running at full attention, all the time. This is not the same as using AI to replace a tedious task and doing something more interesting instead. This is using AI to create a new kind of tedious task. One that demands more of Corty, not less.
Corty charges when you stop, not while you're working. When her energy runs low, the first things to go are the capacities we rely on most: the ability to concentrate, to notice what matters, to make good calls under pressure. She doesn't crash dramatically. She quietly leaves the office. And once she's gone, the errors get worse, the decisions get shakier, and the day gets heavier faster than it should.
What leaders are measuring - and what they're missing
What leaders are measuring - and what they're missing
Most organizations rolling out AI tools are measuring adoption. How many people are using it. How many tasks it's being applied to. What percentage of the team has completed the onboarding. These are the visible numbers.
What they're not measuring is the cognitive cost of AI supervision. That cost is real, and it's showing up in the data. The World Economic Forum's 2026 report on brain capital found that the most in-demand skills - both today and in the future - are almost entirely what they call brain skills: analytical thinking, adaptability, creative thinking, resilience. These are Corty's domain. They're also exactly what gets depleted when she's in AI oversight mode all day.
There's a pattern in the BCG data worth paying attention to. Workers who used AI to replace repetitive tasks did not show the same burnout risk. It is not AI use itself that's the problem. It's the type of use. Replace tedious work with AI and the brain gets breathing room. Supervise AI output all day and the brain carries a new kind of load - one nobody designed the brain to carry at this scale.
Two things worth doing now
The question for a leader is not "are we using AI?" It's "what are we actually asking people to do with AI, and what is that doing to their capacity to think clearly?"
First, map how your team is actually using AI. Not how they're supposed to use it. How they are. If most of their AI time is spent reviewing, checking, and correcting outputs before they go anywhere, that is AI supervision - and it comes with a real cognitive cost. If AI is replacing work they used to do manually, the picture is different. This doesn't require a formal process. Ask. Spend twenty minutes with two or three people and find out what their day actually looks like.
Second, take Corty's charging needs seriously. She does not recover during back-to-back meetings. She recovers when there is nothing cognitively demanding happening - a walk, a proper break, real time between tasks. In a week where the AI oversight load is high, this matters more, not less.
One practical step: if you run a weekly check-in, add one question this week. "What was the most mentally demanding thing you did yesterday?" The answers are often different from what leaders expect.
BCG published their findings in March 2026. The headline got some attention. The finding underneath it - that the type of AI use matters more than the amount - got much less.
What would you find if you asked your team that question today?
— Maria