Harvard Business Review: how heavy AI use leads to "brain fry"

AI increasingly takes over not just our routine work but also part of the attention we used to spend on calm, sequential thinking. Against this backdrop, a new workplace effect has emerged: AI brain fry. It doesn't feel like ordinary end-of-day fatigue. It feels like a strange mix of mental noise, sluggishness and lost clarity. You seem to have done a lot: you queried a model, refined the task, got an answer, double-checked it, launched another tool. But at some point you notice that decisions come harder, your wording gets worse, and simple tasks start demanding far too much effort.

This is the phenomenon examined in a March 2026 Harvard Business Review article. Experts from BCG and the University of California, Riverside wrote it after surveying 1488 US employees at large companies. Their conclusion is uncomfortable but familiar to many teams: AI doesn't just save time; it can quietly drain cognitive resources. This isn't a clinical diagnosis. It's a working state that emerges where automation promises relief but actually adds a new layer of load. To get a good result, an employee can't simply "ask the AI." They have to frame the task precisely, pick the right tool, judge the quality of the answer, spot errors and fit the result into a real workflow.

The people most at risk aren't beginners but the most active users. The reason is that intensive AI work consists of a long chain of micro-decisions. Which model fits the task?

How much context should it get? Is the prompt worth rewriting? How do you verify a fact?

When do you stop iterating, and which of several versions counts as final? From the outside, this looks like acceleration. From the inside, it feels like constant dispatch work.

The more assistants, agents, code editors and text generators you have, the more likely you are to turn from a doer into the operator of a complex system. And if the system speeds up faster than you can make sense of it, your brain pays for it with overload.

The condition has recognizable symptoms. Thoughts break off midway more often, attention splinters into small fragments, and confidence in decisions drops. Sometimes people start trusting the model too much and miss errors because they're too tired to keep checking. Sometimes the opposite happens: endless re-checking, new prompts, comparing versions and an inability to call anything finished. Either way, the quality of the work suffers.

The effect is especially visible where work requires long chains of reasoning and mistakes are costly: analytics, software development, product work, research, marketing and management. AI is good at helping you get started, quickly assemble a draft or clear out routine tasks. But it can also fragment the middle of the process, the part where a person needs focus, critical thinking and the ability to keep the big picture in mind.

The main takeaway isn't that we should abandon AI. If anything, it's the reverse: the deeper AI embeds itself in daily work, the more usage hygiene matters. Teams and individual specialists will have to design not only their automation but also their own cognitive load.

It helps to decide in advance which tasks AI genuinely speeds up, where a single iteration is enough, and where it's better to think first without prompts. It also helps to limit the number of parallel tools, set quality criteria before running the model, and leave time for manual review without new queries. Real productivity in the AI era isn't measured by the number of prompts.

It's measured by whether you keep a clear head and end up making stronger decisions.