Efficient or Exhausted? The AI Trade-off nobody priced in

My laptop is barely twelve months old. Most days now it's flat by 11am. I've started plugging it in before lunch, something I never used to do. Nothing's wrong with the battery. I've just asked it to run flat out since 7am: two Claude Design tabs open, background AI tasks running, constant switching between decks and drafts.

I've mentioned this on stage a few times now, almost as an aside, and to friends outside it. Nearly everyone says the same thing is happening to them.

That's when I realised what my laptop had been showing me the whole time. AI has resulted in the intensification of the workday. My own (brain) battery was draining just as fast, on the same hours, doing what felt like the same work as always. Not a coincidence. It was the same story playing out on two different machines, and it matches what ActivTrak's Productivity Lab found when it analysed 443 million hours of work activity across 1111 companies for its 2026 State of the Workplace report: multitasking rose 12 per cent and collaboration surged 34 per cent, year on year.

Multitasking doesn't just feel effortful. It burns through glucose the same way a battery burns through charge, and every switch between tasks draws on the exact resource your prefrontal cortex needs to think clearly. My laptop was showing me, in real time, what my own brain had been doing all day. Neither of us was built to run flat out, uninterrupted, for eight hours straight.

Researchers at UC Berkeley's Haas School of Business, Aruna Ranganathan and Xingqi Maggie Ye, found the same pattern at scale, following employees at one US tech company for eight months. People didn't reduce their workload when AI arrived. They took on tasks they'd previously outsourced, squeezed in extra bursts of work on evenings and weekends, and spent more of the day supervising multiple bots running different jobs at once. At first, everyone felt energised. Then the workday began to stretch without anyone deciding it should. A prompt during lunch. One more query before leaving the office. Breaks got shorter, and nobody chose to cut them. The researchers call it workload creep. It accumulated, unnoticed, until it started showing up in the quality of people's decisions.

There's a second shift happening underneath the workload creep, and it's easy to miss because it looks like relief. AI takes the execution first: the drafting, the formatting, the first pass at the research, the parts of the job with a clear right answer. What's left for you is what AI can't do cleanly: judging whether the output is good enough to use, deciding which version to trust, working out what the data means, weighing a decision that doesn't have a single correct answer. These tend to be more cognitively demanding tasks. Researchers at Microsoft Research and the University of Edinburgh call this task-complexity polarisation. Easy tasks get easier. Hard tasks get harder, because the person doing them is now also monitoring, verifying, and correcting a second worker that never gets tired. What's left in your day is disproportionately the highest-order thinking a brain can do: analysing, synthesising, hypothesising. And there's more of it per hour than there used to be.

This is the Toggle Tax, and the brain pays it whether the switch is between two humans, two screens, or two chatbots. Gloria Mark's research puts a number on the switch itself: attention drifts after 47 seconds on screen, and a full interruption can cost more than 23 minutes to recover from. Boston Consulting Group has since put a number on where the tax turns punitive. Productivity climbs as people adopt their first one to three AI tools. At a fourth tool, it drops. Add a fifth system to supervise and you haven't multiplied your output, you've multiplied your toggling. The belief that more tools means more speed doesn't hold up under measurement either. METR found experienced developers using AI coding tools completed tasks more slowly in practice, even as they believed the work had sped up by nearly a quarter.

The same research team surveyed nearly 1500 full-time employees across industries and found 14 per cent of people who need heavy oversight of their AI tools report acute cognitive fatigue as a result. Workers described mental fog, headaches, slower decision-making, and a strange sense that their own thinking had become crowded. They call it AI brain fry, mental fatigue that sets in once interacting with AI exceeds a person's cognitive capacity. What's notable is why. AI didn't just reduce workloads. It expanded what the researchers call the sphere of accountability, so people ended up producing more, monitoring more outputs, and managing more information in the same number of hours.

The cost compounds the longer it runs. MIT Media Lab researcher Nataliya Kosmyna found that brain connectivity dropped by as much as 55 per cent while people used ChatGPT for a writing task, compared with doing the same task unaided. Michael Gerlich's research, published in the journal Societies and cited separately by McKinsey Health Institute, found a significant negative correlation between frequent AI tool use and critical thinking ability. Researchers at Carnegie Mellon put this to the test directly. After just ten minutes of AI-assisted problem solving, the people who then lost access to the AI performed worse, and gave up more often, than people who'd never used it at all.

McKinsey Health Institute's Jacqueline Brassey, the researcher behind the exhaustion figure I've cited before, published new work with her colleagues in July naming exactly this problem. Their argument: AI's value is constrained by human capacity, not technology. This is exactly what I explore in my keynote The Brain Health Dividend. Calibrate cognitive load, so intense work is interspersed with lighter work instead of stacking demand hour after hour. Protect recovery, because the brain clears metabolic waste far more effectively during sleep than at any point during the day. Build in protected, uninterrupted focus time. Keep the muscle of independent judgment in use, not outsourced by default. And design the culture so AI is introduced as a thinking partner (I call it your sparring partner), not a replacement for thinking (what I call a ghostwriter).

If your organisation is rolling out AI faster than it's protecting the brains running it, that's worth a conversation before the numbers make the case for you. Reach out if I can help your team to power up performance in the digitally intense workplace.

References

ActivTrak. (2026). 2026 State of the workplace report. ActivTrak Productivity Lab. https://www.activtrak.com/resources/state-of-the-workplace/

Becker, J., Rush, N., Barnes, E., & Rein, D. (2025). Measuring the impact of early-2025 AI on experienced open-source developer productivity (arXiv:2507.09089). arXiv. https://arxiv.org/abs/2507.09089

Bedard, J., Kropp, M., Hsu, M., Karaman, O. T., Hawes, J., & Rosen Kellerman, G. (2026, March 5). When using AI leads to "brain fry." Harvard Business Review. https://hbr.org/2026/03/when-using-ai-leads-to-brain-fry

Brassey, J., Goran, J., Krivkovich, A., & Smaje, K. (2026, July 13). Five principles for designing brain-powered organizations. McKinsey & Company. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-organization-blog/five-principles-for-designing-brain-powered-organizations

Budzyń, K., Romańczyk, M., Kitala, D., et al. (2025). Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: A multicentre, observational study. The Lancet Gastroenterology & Hepatology, 10(10), 896–903. https://doi.org/10.1016/S2468-1253(25)00133-5

Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), Article 6. https://doi.org/10.3390/soc15010006

Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task (arXiv:2506.08872). arXiv. https://arxiv.org/abs/2506.08872

Liu, G., Christian, B., Dumbalska, T., Bakker, M. A., & Dubey, R. (2026). AI assistance reduces persistence and hurts independent performance (arXiv:2604.04721). arXiv. https://arxiv.org/abs/2604.04721

Mark, G. (2023). Attention span: A groundbreaking way to restore balance, happiness and productivity. Hanover Square Press.

Mark, G., Gudith, D., & Klocke, U. (2008). The cost of interrupted work: More speed and stress. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 107–110). ACM. https://doi.org/10.1145/1357054.1357072

Ranganathan, A., & Ye, X. M. (2026, February 9). AI doesn't reduce work, it intensifies it. Harvard Business Review. https://hbr.org/2026/02/ai-doesnt-reduce-work-it-intensifies-it

Simkute, A., Tankelevitch, L., Kewenig, V., Scott, A. E., Sellen, A., & Rintel, S. (2025). Ironies of generative AI: Understanding and mitigating productivity loss in human-AI interaction. International Journal of Human–Computer Interaction, 41(5), 2898–2919. https://doi.org/10.1080/10447318.2024.2405782