AI is being sold to organisations as a productivity revolution. For a growing number of high performers, it is becoming something else entirely: a trap that exploits the very qualities that made them exceptional, and leaves them burned out, error-prone, and quietly wondering if they have permanently damaged their capacity to think.
AI is being sold to organisations as a productivity revolution. For a growing number of high performers, it is becoming something else entirely: a trap that exploits the very qualities that made them exceptional, and leaves them burned out, error-prone, and quietly wondering if they have permanently damaged their capacity to think.
Researchers now have a name for it: AI Brain Fry. And the data behind it should give every leader pause.
The High Performer's Dilemma
Consider the profile of the people most likely to embrace AI tools with genuine enthusiasm. They are achievement-oriented, results-driven, action-oriented, and decisive. They see AI not as a threat but as a superpower, a way to compress weeks of multi-team work into hours of focused prompting.
For a while, it is. The early phase of AI adoption for people like this is often genuinely euphoric. Consumer insights that previously required specialist teams can now be modelled before lunch. Competitive intelligence that once took days arrives in minutes. The barrier to starting new projects collapses.
This is precisely where the problem begins.
Because the same psychological qualities that make AI feel so powerful, the achievement orientation, the results focus, the inability to experience 'good enough' as a natural resting place, are the same qualities that make it almost impossible to stop. The working day expands. Then the evenings. Then the weekends. Projects multiply because starting them has become frictionless. The person has more ideas than ever and increasingly less capacity to execute any of them cleanly.
The Research Evidence Is Stark
A UC Berkeley longitudinal study tracking 200 employees over eight months reached a conclusion that should be required reading for every leadership team: "AI makes it easier to do more but harder to stop." The researchers identified three primary mechanisms at work.
Task expansion occurs when work that previously required specialist teams suddenly feels achievable solo, leading professionals to absorb far more than is humanly sustainable. Boundary dissolution describes how the ease and portability of AI tools erodes the separation between work and personal time, prompting sessions extend into evenings, commutes, weekends. Multitasking escalation follows naturally: the speed of AI-assisted output creates constant pressure to run multiple workstreams simultaneously.
A major BCG study of 1,488 workers, published in January 2026, put numbers to the resulting damage:
- 14% of all workers surveyed showed symptoms of AI Brain Fry
- 26% of workers in marketing roles were affected, the highest of any sector
- AI Brain Fry is associated with a 39% increase in major errors
- Affected workers are 39% more likely to be actively seeking new employment
These are not marginal effects. A near-doubling of serious error rates and a substantial increase in attrition risk, concentrated in the functions most central to people-intensive organisations, represent a direct and measurable threat to performance.
Workslop: When AI Output Masquerades as Quality
Alongside Brain Fry, a related phenomenon has emerged. Stanford researcher Jeff Hancock and colleagues at BetterUp have coined the term 'workslop' to describe AI-generated work that appears polished on the surface but is so flawed or inaccurate that it requires heavy correction, rework, or complete redo once it reaches colleagues.
A survey of 1,150 US desk workers found that 40% had encountered workslop within the preceding month, with the projected cost running to $9 million per year for a 10,000-person organisation.
Critically, this is not primarily a story of lazy workers cutting corners. Workers are often being directed to use AI without adequate guidance, support, or training. But there is a deeper issue: the study found a stark disconnect between management perception and worker reality. 40% of non-managers report that AI saves them no time at all, while 92% of senior executives say it makes them more productive.
This gap is itself a serious risk factor. Executives who believe AI is working across the organisation are unlikely to put in place the support structures that workers actually need.
The Automation Paradox, Revisited
None of this should be entirely surprising. In 1983, systems engineer Lisanne Bainbridge published a paper called Ironies of Automation that identified a fundamental tension in any system where humans are expected to monitor automated processes: the more reliable the automation, the less engaged, and therefore the less capable, the human operator becomes.
Bainbridge was writing about industrial control systems. But her core insight maps precisely onto what is now happening with AI in knowledge work. The more capable AI becomes at generating output, the more the human's role shifts from doing to reviewing. And sustained attention to reviewing work you did not produce, at the volume AI enables, is cognitively exhausting in ways that doing the work yourself is not.
The deepest irony is this: the very people organisations cannot afford to lose, the high performers who embraced AI earliest and most fully, are the ones most at risk.
What Organisations Must Do Now
The response to AI Brain Fry cannot be to slow AI adoption. The competitive pressures are real, and the productivity gains, where properly managed, are genuine. The challenge is to design the conditions under which people can work with AI sustainably.
From a psychological standpoint, this requires several things.
Cognitive load governance. Organisations need explicit policies about the volume and complexity of AI-assisted work any individual is expected to manage. The collapse of the barrier to starting projects does not mean humans have unlimited capacity to run them. Someone needs to hold that boundary, and it will not be the AI.
Quality verification structures. The BCG research is clear that AI Brain Fry dramatically increases error rates. This is not a performance management issue, it is a systems design issue. Verification checkpoints, peer review, and structured output validation need to be built into AI-assisted workflows, not treated as optional.
Psychological safety around limitation. Perhaps the most important cultural shift needed is one that makes it genuinely safe for high performers to say 'I am at capacity.' The psychological profile of the people most vulnerable to AI Brain Fry, achievement-oriented, results-driven, status-conscious, makes them the least likely to volunteer this information without explicit structural support.
Rebalancing the value equation. The BCG finding that 92% of executives believe AI is delivering productivity gains, while 40% of non-managers say it saves them no time at all, points to a fundamental misalignment in how AI value is being measured. If the gains are flowing upward while the costs are being absorbed by the people doing the work, the system is not sustainable.
The Deeper Question for Transformation Leaders
AI is not the first technology to promise to transform how organisations work, and it will not be the last. But it is arriving at a moment when many organisations are already stretched: post-pandemic, structurally leaner, with leadership pipelines that have had limited time to develop.
The question for leaders is not whether to adopt AI, that ship has sailed. It is whether to adopt it in a way that treats human psychological capacity as a fixed and finite resource requiring careful stewardship, or as an infinitely elastic input that can simply absorb whatever the technology enables.
The research is unambiguous about which assumption produces better outcomes.
Organisations that will navigate this well are those that bring the same analytical rigour to understanding their people's psychological experience of AI adoption as they bring to measuring AI's impact on output. That means mapping the psychological profiles of the individuals most at risk, designing workflows that respect cognitive limits, and building cultures where performance is measured by sustainable output over time, not by how many tasks a single person can absorb in a single day.
The organisations that get this right will not just avoid the costs of Brain Fry and workslop. They will build the kind of human capability that no AI tool, however powerful, can replicate: teams that think clearly, judge well, and remain genuinely engaged in the long-term mission of the business.
Matthew Bayfield