Why do so many people find that, after adopting AI, their output has gone up but they feel more exhausted?
A major reason is that knowledge work has always had high and low gears. High-gear tasks require constant thinking: reading difficult material, making trade-offs, evaluating approaches, checking whether AI outputs are usable. Low-gear tasks are more like manual work: formatting, moving data around, running commands, writing boilerplate code, fixing a trivial bug.
In the past, these two types of tasks would naturally interleave. After finishing a stretch of difficult judgment work, you might tidy up files, rename variables, run tests, or fill in some templates. These activities look slow and unsophisticated, but they have one benefit: you are still working, yet your brain doesn’t need to make difficult judgments continuously.
AI happens to eliminate these low-gear tasks first. So the workflow gets faster, but people also downshift less often. You aren’t doing more manual labor — you’re just spending more time in the state of reading, reviewing, judging, dispatching, and verifying.
As a result, the problem of rest in the AI era has become harder. In the past, switching tasks might naturally downshift you. Now, switching tasks often just means moving from one high-density judgment to another. The focus of rest, therefore, shifts from “do something else” to “first stop external input.”
Knowledge work was never a nonstop mental burn. When programmers write code, they aren’t designing architecture every minute. Much of the time is spent renaming things, filling in types, adjusting formatting, running tests. Researchers aren’t always reading the hardest passages in a paper — they pause to organize citations, archive materials, clean up notes.
These activities look trivial on their own. A manager looking at a timesheet might see them as inefficiencies. But for the human brain, they serve a purpose. Because during these moments, you don’t need to continuously absorb new information or make constant trade-offs. You’re still moving things forward, but the judgment system has loosened slightly.
This is also why many people used to recover just by switching tasks. Tired of writing? Go fix typos. Tired of designing? Go organize documents. Tired of meetings? Go clear a few quick emails. Strictly speaking, this isn’t real rest, but it does pull the brain down from a high gear.
What AI handles best is precisely these low-gear tasks. Retrieving information, generating boilerplate code, organizing meeting notes, rephrasing a paragraph, filling in a simple implementation — all of these can be done quickly now.
This certainly boosts output. But something else has happened too: people have lost many of the opportunities to naturally downshift.
The old rhythm looked like this: judge, then execute on your own for a while, then judge again. Now two common rhythms have emerged.
The first is micro-management. You ask AI to write a section, it gives you one; you read it, find problems, tweak the prompt; it gives you another version, you keep spotting issues. Less manual work, more mental gatekeeping.
The second is parallel work. You dispatch several AI tasks at once: one to look up references, one to write code, one to draft a proposal. On the surface, you’re waiting for them to work. In reality, you’ll soon be switching among several results, judging which one is reliable, which needs a redo, which can be merged. Parallelism reduces waiting — and concentrates the verification pressure onto the same person.
BCG and Harvard Business Review research on AI-related cognitive fatigue surveyed 1,488 US employees. The researchers call the fatigue from overusing or continuously monitoring AI tools “AI brain fry.” The label itself doesn’t matter much. What’s useful is that it identifies this: the most exhausting part is often not opening the AI tool itself, but constantly watching its output, checking for errors, and judging whether it’s usable.
This matches the experience of many heavy AI users. AI hasn’t simply reduced work. By speeding up execution, it has pushed human judgment further to the front.
This leads to a misconception: I switched tasks, so I rested.
Tired of writing reports? Switch to answering email. Tired of reading code? Switch to reading industry news. This used to help, because you might move from a hard task to an easy one. Not necessarily anymore. Email may require judging tone and risk. Industry news feeds you new concepts. AI summaries also demand you judge truthfulness and priority.
In this scenario, you’ve only switched windows. Your brain is still judging.
Scrolling on your phone is the same problem. It has no work objective, so it looks like rest. But you’re still reading, choosing, judging, reacting: tap this or not? Reply to that or not? Is this opinion right? Should I keep watching the next video? These micro-decisions all consume attention.
A study on smartphone use during lunch breaks summarized by the Association for Psychological Science points out that using a phone during a break may be just as mentally taxing as continuing to work, leaving you more tired after the break. A rest review in Frontiers in Psychology also emphasizes that recovery requires detaching from current goals. If a period of time still requires you to stay alert, process input, and make choices, it’s hard to count as rest.
The easiest thing to get wrong in the AI era is right here: changing the topic doesn’t mean the brain is resting.
The first principle of rest isn’t forcing yourself to stop thinking — it’s stopping the flow of new material into your brain.
There are roughly three types of input to stop. The first is information input: text, video, podcasts, group chats, news, long AI responses. The second is choice input: tap this or not, reply or not, believe or not, keep watching or not. The third is goal input: is this useful, can this be delivered, what’s the next step.
The activities that genuinely restore you are usually quite ordinary: walking without your phone, sitting with your eyes closed for a bit, showering, washing a cup, driving in silence, staring out the window. They may not be interesting, but what they share is low input, low choice, low goals.
Of course you might still think about work while walking. You might suddenly solve a problem in the shower. That’s fine. The brain organizing existing material on its own is a different thing from continuing to scroll and read AI output.
A micro-break meta-analysis in PLOS ONE shows that short breaks under 10 minutes are associated with higher energy and lower fatigue. However, for high-difficulty cognitive tasks, a few minutes is usually not enough to fully replenish resources. Microsoft WorkLab’s small 14-person EEG study can be understood similarly: leaving 10 minutes between meetings can stop stress from continuing to accumulate, but it doesn’t mean the brain has recovered enough to resume complex judgment.
So the value of short breaks is more like damage control. They keep stress from piling up further and help you exit the previous task. Truly restoring deep judgment usually requires a longer stretch of low-input time.
Coffee also deserves a mention here. Coffee can make you more alert, but not necessarily better at judging. Research in Nature Scientific Reports reminds us that complex cognitive performance isn’t just about alertness. Sleepiness is only one signal. Not being able to read, repeatedly switching windows, missing obvious errors, not knowing what to write next — these are stronger signs that you need to offload.
Mindfulness and meditation are often brought up in discussions of rest. By the “stop input” standard, they are indeed useful — but there’s no need to make them mystical.
The most down-to-earth value of mindfulness is that it gives you a clear action: sit down, close your eyes, place attention on your breath. In doing so, external information decreases. You don’t need to choose the next piece of content, and you don’t need to deliver anything. Like walking, showering, or washing a cup, it is low-input recovery.
The problem is that many people turn mindfulness itself into a task: did I check in today? How many consecutive days? Am I breathing correctly? Did I just drift off? Have I made progress? Once these metrics enter the picture, mindfulness becomes something to complete, optimize, and evaluate again.
For knowledge workers, the most useful part of mindfulness is often not training focus, but allowing yourself to temporarily not focus.
So “I need to practice meditation” can sometimes be too heavy. Better to replace it with a simpler phrase: “I’m spending 10 minutes stopping input.” This asks less and is easier to implement.
The logic isn’t hard. What’s hard is actually doing it.
When someone is used to short videos, group chats, news, and AI output, suddenly shutting off input usually doesn’t feel good in the first few minutes. You’ll feel bored, anxious, like you’re wasting time. Your hand will automatically reach for your phone. Your brain will remind you there’s still so much to do.
This doesn’t mean your rest has failed. It only means your system hasn’t yet come down from a high-stimulation state.
So don’t rely on willpower to decide not to scroll on your phone when you’re already exhausted. At that point, judgment has already declined, and you’ll probably lose to your default behavior. The better approach is to lay the path in advance.
Don’t keep your phone on you. Put it in another room, or at least somewhere you have to stand up to reach. Decide on your rest activity ahead of time too: go for a walk, wash a cup, sit with your eyes closed for 10 minutes, take a shower. Don’t wait until you’re tired to decide on the spot. When you’re tired, people naturally choose the easiest, most stimulating thing.
This approach can be very simple.
This isn’t about turning rest into yet another productivity system. It’s just about making fewer choices when you’re most likely to slide back into high input.
In the AI era, generation and execution are becoming cheap. What’s truly scarce is whether a person can still consistently make good judgments. Rest isn’t slacking off — it’s letting your judgment come back.