Does AI Fry Workers’ Brains, or Does the Way We Deploy It?

Deloitte treats AI “brain fry” as a problem of interaction design. The evidence says AI contributes, and the arrangement around it decides how much.

Is brain fry something AI does to people, or something organizations do to people once AI is in place?

Deloitte’s recent piece on human-AI interaction (Deloitte Insights, 15 October 2026) reads it mostly as the first. I would argue the answer lies less in the tool considered on its own than in the arrangement it enters. AI can be a critical contributor to brain fry. The arrangement decides whether that contribution becomes a burden.

What Deloitte proposes, and what it leaves in place

Deloitte’s four lessons are sensible on their own terms. Make AI reasoning visible, show fewer options first, build in “beneficial friction,” and ask for a human estimate before the AI’s. Each treats the moment of interaction between one worker and one tool.

The piece also contains, almost in passing, the other half of the story. AI makes it easier to get more work done, so “employees could find themselves being asked to do too much and could experience AI-related burnout.” In its close, Deloitte expects “time saved, tasks completed, and volume generated” to remain the core business metrics. Those two sentences describe the arrangement, and the remedies leave both untouched. A badge on an AI output does not change how much work lands on the person reading it, or what that person is measured on.

What the evidence says drives brain fry

The term comes from a BCG study of 1,488 full-time US workers, published in Harvard Business Review, which defines brain fry as mental fatigue from excessive use or oversight of AI tools beyond one’s cognitive capacity (Axios, 6 March 2026). About 14% of AI users reported it. Its drivers sit at three levels, and they connect.

The tool and the workflow

Some of the load is in the work AI creates. High-oversight workflows, where people review, correct and interpret model output, came with 14% more mental effort, 12% more fatigue and 19% more information overload than workflows where AI simply took a task over. Self-reported productivity rose with up to three AI tools and fell with four or more (Fortune, 10 March 2026). This is where Deloitte’s design remedies have real purchase, and where the claim that AI contributes holds.

What organizations add, measure and allow

The same study points beyond the tool. Its lead author notes that many companies add AI “on top of an employee’s already-established set of responsibilities,” and brain fry fell where managers gave training and support (Fortune). BCG advises that an AI strategy should not reward the quantity of AI use, finds an emphasis on work-life balance “highly protective,” and reports lower risk where teams built AI into shared team workflows and treated adoption as collective rather than individual competition (UNLEASH, April 2026).

Some employers do the opposite. Axios notes that more employers are making AI use mandatory and measuring it in performance reviews. Shopify’s CEO set AI use as a “fundamental expectation” in April 2025, with AI use to be counted in performance reviews and teams expected to show why they “cannot get what they want done using AI” before asking for headcount (CNBC via NBC Connecticut, 7 April 2025). The time AI saves is also being spent before anyone decides how to spend it. In Workday’s survey of 3,200 employees, roughly four of every ten hours saved went to fixing and checking output, and organizations struggling with AI put recovered time into more tasks (32%) or more technology (39%). Those seeing strong outcomes used it for deeper analysis and strategic thinking (57%) (UC Today, 14 January 2026).

What workers do on their own

The strongest objection to blaming the organization comes from an eight-month study of a roughly 200-person US technology company. Researchers at UC Berkeley Haas found that employees worked faster, took on broader scope and stretched work into more hours, “often without being asked to do so” (Berkeley Haas, 18 February 2026). No mandate was needed. The same researchers describe what happens next: higher output raises expectations, and those expectations drive further expansion. The pull may start with the worker. It holds because the organization absorbs the new pace as the new baseline.

Two caveats apply. BCG’s figures are associations, not proof of cause, and the Berkeley findings come from one firm. Together they still show that tool, organization and worker each move the outcome, and that none explains it alone.

A different reading: flip the sentence

The accepted sentence, in Deloitte’s piece and in most coverage, is “AI exhausts workers.” Flip it, keeping the same parties and the same relation and changing only which side acts: “Organizations exhaust workers through AI.” Read alone, it would be as one-sided as the original. What the flip shows is the work the accepted version leaves out. Someone reviews the output, refills the saved hours and meets the new pace, and the cost of that work falls on employees while the savings are booked by the firm.

Read this way, Deloitte’s remedies treat the symptom at the point where it shows, the interaction, and leave the condition in place. A worker who no longer second-guesses a badged output but is measured on volume, carries work added on top and sees saved hours refilled is less tired per output and does more outputs.

Six facets, one condition

Seen through how an organization’s parts fit together, brain fry sits where six facets meet. Technology sets the oversight load: how many tools, and how much review each output needs. Metrics decide whether that load is rewarded or contained. A target on volume of AI use turns every saved hour into expected output, and process then adds AI on top of existing duties instead of redesigning them. Culture determines whether people slow down or keep pace. Deloitte itself cites a survey in which more than half of US employees had knowingly passed along low-quality AI work, partly because of social pressure. Talent shows up through managers: where they train and support, brain fry falls. Structure decides who owns any of this. Deloitte’s 2026 Human Capital Trends survey found 66% of leaders saying the intentional design of human-AI interaction matters, and 6% saying their organizations are making great progress. Moderna’s response was structural: it merged HR and IT under one Chief People and Digital Technology Officer to plan “the flow of work” across people and AI agents together (UNLEASH, 27 June 2025).

The facets work through one another. Interface fixes touch the technology facet. Unless metrics and process change with them, the gains are absorbed.

What leaders should do

CIO

• Before rollout, cap the number of AI tools any one role must oversee, and treat four or more as a design risk, given BCG’s productivity drop at that point.
• Publish, alongside every AI business case, where the saved hours will go: to deeper analysis, to capacity released, or to more volume. Make that a decision rather than a default.

CTO

• Design workflows so that AI takes tasks over where it reliably can, rather than adding a human review step to every output. BCG’s data ties the heaviest strain to oversight-heavy work.
• When adopting trust cues, favor sources a reader can check over confidence labels. In a 2025 experiment, explanations raised reliance on wrong AI answers as well as right ones, while sources reduced it (Kim et al., CHI 2025). A badge that lowers checking without raising accuracy moves the risk instead of reducing it.

Transformation Leaders

• Remove AI use counts from performance and adoption scorecards, and measure quality of output and decisions instead, as BCG advises.
• Redesign roles when AI arrives: take duties out as AI work comes in, and track brain fry separately from burnout in pulse surveys.

Executive Team

• Name one owner for how work is designed around AI, with authority across technology, HR and the business. Today the 66-to-6 gap suggests that ownership sits nowhere.
• Ask in every AI review what was taken off people’s plates, not only what was added to their output.

So, does AI fry workers’ brains?

Partly, yes. Oversight-heavy workflows and too many tools carry a real cognitive load, and design can lower it. The larger share depends on the arrangement: work added on top, use rewarded by volume, saved time refilled, and a pace that becomes the baseline whether or not anyone asked for it. Deloitte’s lessons improve the moment of interaction; the arrangement decides whether the improvement survives. Which leader in the organization owns the arrangement, and what is that leader measured on?

Sources

1. Deloitte Insights, “4 lessons from behavioral economics that could help improve human-AI interaction,” David Mallon, Brad Kreit, Timothy Murphy, Sarah Szpaichler, 15 October 2026. https://www.deloitte.com/us/en/insights/topics/talent/improve-human-ai-interaction.html
2. Axios, “AI overuse could spark ‘brain fry,’ new research finds,” 6 March 2026 (reporting BCG and UC Riverside research in Harvard Business Review). https://www.axios.com/2026/03/06/ai-chatgpt-claude-jobs-brain-fry
3. Fortune, BCG “AI brain fry” study coverage, 10 March 2026. https://fortune.com/2026/03/10/ai-brain-fry-workplace-productivity-bcg-study/
4. UNLEASH, “Your employees have AI brain fry and it’s affecting your business: BCG on fixing the problem,” April 2026. https://www.unleash.ai/artificial-intelligence/your-employees-have-ai-brain-fry-its-affecting-your-business-bcg-on-fixing-the-problem/
5. CNBC via NBC Connecticut, “Shopify CEO says staffers need to prove jobs can’t be done by AI before asking for more headcount,” 7 April 2025. https://nbcconnecticut.com/news/business/money-report/shopify-ceo-says-staffers-need-to-prove-jobs-cant-be-done-by-ai-before-asking-for-more-headcount/3537432
6. UC Today, “Workday Reveals Why Most Companies Are Leaving AI Value on the Table,” 14 January 2026. https://www.uctoday.com/productivity-automation/workday-reveals-why-most-companies-are-leaving-ai-value-on-the-table/
7. Berkeley Haas News, “AI promised to free up workers’ time. UC Berkeley Haas researchers found the opposite,” 18 February 2026. https://newsroom.haas.berkeley.edu/ai-promised-to-free-up-workers-time-uc-berkeley-haas-researchers-found-the-opposite/
8. UNLEASH, “Why Moderna merged HR and IT to better architect the flow of work,” 27 June 2025. https://www.unleash.ai/artificial-intelligence/why-moderna-merged-hr-and-it-to-better-architect-the-flow-of-work/
9. Kim, Vaughan, Liao, Lombrozo, Russakovsky, “Fostering Appropriate Reliance on Large Language Models,” CHI 2025 (arXiv, 12 February 2025). https://arxiv.org/html/2502.08554v1