Why is AI multi-tasking making people more tired than ever?

I recently read a blog: https://steve-yegge.medium.com/the-ai-vampire-eda6e4f07163 :talking about AI making people more tired while more productive. We developers of AI tools and models, we actually feel somewhat related in our day to day work as well. But why? We dont really want to drill into the empirical aspect of this phenomenon, but we are very curious what is the science behind this? psychology and

Let’s break it down.

The Neurological View: Your Prefrontal Cortex is Literally Drowning in Glutamate

The most striking piece of science here comes from a 2022 study published in Current Biology by Wiehler, Pessiglione and colleagues at the Paris Brain Institute. They used magnetic resonance spectroscopy (MRS) to peer into the brains of people performing demanding cognitive tasks over the course of an entire workday. What they found was remarkable: after hours of sustained cognitive control, a neurotransmitter called glutamate accumulated to significantly elevated levels specifically in the lateral prefrontal cortex (lPFC) — the brain region that acts as your executive command center for planning, decision-making, and self-control.

Now think about what working with AI actually demands of your lPFC. When you’re vibe-coding, you’re not writing code the old way — line by line, in a flow state, with your fingers doing the muscle-memory work. Instead, you are constantly reviewing AI-generated output, deciding whether it’s correct, evaluating tradeoffs between different approaches, steering direction, catching subtle errors, and making judgment call after judgment call. Every single one of these actions is a cognitive control operation that fires up your prefrontal cortex. You’re essentially doing the hardest kind of brain work — executive oversight — nonstop, with almost zero of the easy routine work that used to give your PFC a break.

The Wiehler study showed that as glutamate builds up, it triggers a regulatory mechanism that makes the prefrontal cortex progressively more costly to activate. Your brain literally starts making cognitive control harder to mobilize — not because you’re lazy, but because it’s protecting itself from excitotoxicity (glutamate in excess is neurotoxic). The subjective experience of this? That crushing fatigue after a long AI session. The nap attacks Steve Yegge describes. The feeling that you just cannot make one more decision. That’s your lPFC hitting a biochemical wall.

A 2025 review by Pessiglione and colleagues in Trends in Cognitive Sciences further crystallized this: cognitive fatigue is fundamentally about depletion of control resources in the prefrontal cortex, observable through both fMRI and EEG. The brain has a finite daily budget for executive control, and AI-assisted work burns through that budget at an accelerated rate because it strips away the low-effort filler tasks and leaves you with a concentrated stream of high-effort decisions.

The Psychological View: Attention Residue, Decision Fatigue, and the Dopamine Slot Machine

Three psychological mechanisms converge to make AI-assisted work uniquely draining.

Attention Residue and Hyper-Switching. Sophie Leroy’s influential 2009 research at the University of Minnesota identified a phenomenon she called “attention residue.” When you switch from Task A to Task B, part of your cognitive attention remains stuck on Task A. The residue doesn’t clear instantly — it lingers, fragmenting your focus and degrading performance on the new task. Her research, cited over 500 times, showed this effect is especially potent when the prior task was left incomplete or felt unresolved.

Working with AI amplifies this to an extreme. In a typical AI coding session, you might prompt the model, review its output, spot an issue, redirect it, review again, then pivot to a different part of the codebase, prompt again, evaluate a completely different kind of output, and so on — dozens of micro-task-switches per hour. Each switch generates attention residue. The classic research on task-switching costs (Kiesel et al., 2010, Psychological Bulletin) shows that every switch incurs a measurable cognitive penalty in both reaction time and error rate — penalties that accumulate over a workday. With AI, you’re paying this switching tax at an unprecedented frequency.

Decision Fatigue on Steroids. The second mechanism is decision fatigue — the well-documented deterioration of decision quality after a long session of making choices. A 2015 study by Mullette-Gillman et al. in PLOS ONE demonstrated that cognitive fatigue destabilizes economic decision-making preferences. People begin making more impulsive, less rational choices as their decision-making machinery wears down.

Pre-AI, a developer’s day had natural decision-light zones: writing boilerplate, running builds, browsing docs, doing mechanical refactoring. These were cognitively cheap — almost rest periods for the decision-making apparatus. AI has eliminated virtually all of that. What’s left is a pure, unrelenting stream of evaluation and judgment. Accept this code or reject it. Is this approach correct or not. Does this architecture make sense. Should I redirect the model or let it continue. These are all real decisions, and they never stop coming.

The Variable Reward Trap. The third mechanism is perhaps the most insidious. Yegge compares AI coding to a slot machine, and the neuroscience backs this up powerfully. AI outputs follow what psychologists call a variable ratio reinforcement schedule — the same reward pattern that makes gambling addictive. Sometimes you get brilliant code on the first try. Sometimes you get garbage. Sometimes you get something surprisingly creative that you never would have thought of. The unpredictability is the key.

Research by Volkow et al. (2011, PNAS) and more recently Clark & Zack (2023, Addictive Behaviors) has established that variable rewards trigger dopamine release in the brain’s mesolimbic pathway — the nucleus accumbens and ventral tegmental area. Each prompt you send to Claude or GPT is essentially pulling a lever. Your dopamine system fires in anticipation of the result. When the result is great, you get a hit of reinforcement. When it’s bad, you feel compelled to try again. This dopamine-driven loop keeps you in the chair long past the point where your prefrontal cortex is screaming for rest. You’re cognitively exhausted, but motivationally wired. That mismatch — fatigue plus compulsion — is exactly what makes it feel so draining and so hard to stop at the same time.

The Biological View: Cortisol, Your Autonomic Nervous System, and the Burnout Cascade

Beyond the brain itself, the body pays a heavy biological price for sustained AI-driven cognitive work. The story here centers on your hypothalamic-pituitary-adrenal (HPA) axis and your autonomic nervous system (ANS).

When you’re in an intense AI session, your body doesn’t distinguish between “reviewing Claude’s code output” and “being chased by a predator” as cleanly as you might think. High-stakes cognitive work — especially the kind where outcomes feel consequential (your code, your job, your startup) — activates the sympathetic nervous system. Your HPA axis fires, cortisol floods your system, your heart rate elevates subtly, and your body enters a low-grade fight-or-flight state. A 2015 study by Teixeira et al. in PLOS ONE demonstrated that chronic cognitive stress impairs autonomic nervous system reactivity and degrades cognitive performance — a vicious feedback loop where stress makes you worse at the thing causing the stress.

Gavelin et al. (2023, Biological Psychology) studied patients with clinical burnout and found that sustained mental activity produced abnormal autonomic responses — the body’s stress recovery system becomes dysregulated. The parasympathetic “rest and digest” system, which is supposed to counterbalance the sympathetic activation, stops doing its job effectively. In the context of the AI burnout phenomenon, this means that the recovery mechanisms that should kick in after work — helping you relax, sleep well, feel restored — are themselves compromised by the chronic cognitive overload.

There’s also the metabolic cost to consider. Your brain accounts for roughly 2% of your body weight but consumes about 20% of your metabolic energy. The prefrontal cortex, being the most metabolically expensive region, is disproportionately affected. Iodice et al. (2017, Scientific Reports) showed that fatigue modulates dopamine availability and shifts decision-making toward less effortful choices — your biology is literally steering you away from hard thinking as a protective response. The “nap attacks” that Yegge and others report after AI sessions aren’t a character flaw. They’re a physiological emergency brake. Your body is forcing a shutdown because sustained prefrontal activation has depleted local metabolic resources and accumulated potentially toxic byproducts faster than they can be cleared.

The chronic dimension is particularly concerning. Research on the HPA axis in chronic fatigue (Papadopoulos & Cleare, 2012, Nature Reviews Endocrinology) has shown that prolonged cortisol elevation eventually leads to HPA axis dysfunction — the system downregulates, producing abnormally low cortisol output. This manifests as persistent fatigue, cognitive fog, and reduced stress resilience. If AI-driven cognitive overload becomes the norm rather than the exception, we may be setting ourselves up for a widespread wave of HPA axis burnout that could take months or years to recover from.

Putting It All Together: The Perfect Storm

What makes the AI fatigue phenomenon so potent is that these three systems — neurological, psychological, and biological — don’t just operate in parallel. They feed into each other in a destructive cascade:

Neurologically, glutamate accumulates in your prefrontal cortex, making each subsequent decision physically harder for your neurons to execute. Psychologically, the variable reward schedule keeps your dopamine system hooked, overriding the fatigue signals that should tell you to stop, while attention residue from constant context-switching fragments whatever cognitive capacity you have left. Biologically, your HPA axis floods cortisol in response to the sustained cognitive demand, your autonomic nervous system stays locked in sympathetic overdrive, and your metabolic reserves deplete faster than they can be replenished.

The result is a person who is simultaneously more productive than ever and more exhausted than ever. It’s not a paradox — it’s basic neurobiology. AI didn’t just automate the easy work. It concentrated the hard work into a relentless stream and then made that stream addictive.

Yegge’s recommendation of a 3-4 hour workday isn’t just workplace philosophy — it’s remarkably well-aligned with what the neuroscience predicts. If the prefrontal cortex has a finite budget for cognitive control before glutamate accumulation forces a shutdown, and if the HPA axis can only sustain elevated cortisol output for so long before it begins to dysregulate, then the answer isn’t willpower or caffeine. It’s structural: shorter, more intense sessions with genuine recovery time between them. The science says your brain was never designed for 8 hours of pure executive decision-making. It was designed for bursts of intense cognition separated by periods of lower-demand activity — exactly the kind of easy work that AI just eliminated from your day.

References

Wiehler, A., Branzoli, F., Adanyeguh, I., Mochel, F., & Pessiglione, M. (2022). A neuro-metabolic account of why daylong cognitive work alters the control of economic decisions. Current Biology, 32(16), 3564-3575.

Pessiglione, M., Blain, B., Wiehler, A., & Naik, S. (2025). Origins and consequences of cognitive fatigue. Trends in Cognitive Sciences.

Leroy, S. (2009). Why is it so hard to do my work? The challenge of attention residue when switching between work tasks. Organizational Behavior and Human Decision Processes, 109(2), 168-181.

Kiesel, A., Steinhauser, M., Wendt, M., et al. (2010). Control and interference in task switching — A review. Psychological Bulletin, 136(5), 849-874.

Mullette-Gillman, O.D.A., Leong, R.L.F., & Kurnianingsih, Y.A. (2015). Cognitive fatigue destabilizes economic decision making preferences and strategies. PLOS ONE.

Volkow, N.D., Wang, G.J., Fowler, J.S., Tomasi, D., & Telang, F. (2011). Addiction: Beyond dopamine reward circuitry. PNAS, 108(37), 15037-15042.

Clark, L., & Zack, M. (2023). Engineered highs: Reward variability and frequency as potential prerequisites of behavioural addiction. Addictive Behaviors.

Iodice, P., Ferrante, C., Brunetti, L., et al. (2017). Fatigue modulates dopamine availability and promotes flexible choice reversals during decision making. Scientific Reports, 7, 4898.

Teixeira, R.R., et al. (2015). Chronic stress induces a hyporeactivity of the autonomic nervous system in response to acute mental stressor and impairs cognitive performance in business executives. PLOS ONE.

Gavelin, H.M., Neely, A.S., Aronsson, I., & Josefsson, M. (2023). Mental fatigue, cognitive performance and autonomic response following sustained mental activity in clinical burnout. Biological Psychology.

Papadopoulos, A.S., & Cleare, A.J. (2012). Hypothalamic-pituitary-adrenal axis dysfunction in chronic fatigue syndrome. Nature Reviews Endocrinology, 8(1), 22-32.