I use AI all day long, and this is what it does to my mind

I often talk with people about the positive aspects of AI. With others, I focus on the downsides of AI in our work. Louis from werkvierentwintig falls into the latter category.

I really use AI a lot, but I also notice that my mind “empties” faster. This is what AI work intensification feels like in practice.

AI doesn’t free you up; it makes you busier

The promise is pretty simple: AI takes over routine work so you have more time for what really matters. This means you have to type less and can analyze faster, while code is available on demand. I can simply get more done and deliver more value in the same amount of time.

Sounds great, of course, but reality tells a different story.

Three studies, one uncomfortable conclusion

In February 2026, Aruna Ranganathan and Xingqi Maggie Ye of UC Berkeley’s Haas School of Business published an article in the Harvard Business Review: “AI Doesn’t Reduce Work, It Intensifies It.” For eight months, they tracked 200 employees at an American technology company, not through surveys, but through direct observation and more than 40 in-depth interviews. The company hadn’t made AI mandatory, but had made it available to everyone.

What they observed was consistent: people started working faster, took on more tasks, and extended their workdays. Starting earlier, finishing later. Without anyone having asked them to. One participant put it this way: “You’d think you might save some time, that you could work less. But then it turns out you’re not working any less after all.”

A month later, in March 2026, the BCG Henderson Institute published a second study, also in the Harvard Business Review. They surveyed 1,488 American employees across various sectors. Their findings introduced a new concept: AI brain fry. Mental fatigue caused by excessive use of or monitoring AI tools, which exceeds cognitive capacity.

At the same time, based on practical research at three Dutch companies, TNO described how GenAI delivers productivity gains but also intensifies work, increases mental strain, and reduces social contact in the workplace.

Three studies. Three different methods. Three countries. One pattern.

How AI leads to increased workloads: three mechanisms

The Berkeley researchers describe three forms of work intensification that reinforce one another.

Expansion of roles

Because AI makes tasks feasible that were previously beyond a person’s reach, colleagues are now taking them on. Product managers are now writing code. Researchers are taking on the role of engineers. This feels like empowerment—and it is—but it has a domino effect. Software engineers have to review the AI-generated code written by colleagues, which increases their own workload. The boundaries of “the job” are blurring. I’m certainly not opposed to this, but it’s nice to know where your work begins and where someone else’s work roughly ends.

Blurring of boundaries

AI is always available, accessible, and responsive. That makes it tempting to handle small tasks during moments that previously served as breaks: sending a prompt during lunch, fleshing out an idea in the evening, or quickly running an analysis before a meeting. The natural “breathing spaces” in the day disappear as a result. This is also because AI needs more time for certain tasks after you’ve entered the prompt. And as AI systems mature, this will only increase. And that brings us to mechanism number three.

Chronic multitasking

AI enables people to keep multiple threads active at the same time. Writing code manually while an agent generates an alternative version. Running multiple agents in parallel. Resuming long-delayed tasks because AI “can handle them in a flash.” The researchers describe this as “a feeling of constantly juggling tasks, even when the work felt productive.”

The result is what’s known in neuropsychology as attention residue: after every task switch, your brain needs time to fully switch gears. That time is no longer available.

I recognize all three of these to a great extent. I usually have about eight threads running simultaneously, which are often quite different in nature. That requires constant attention, coordination, and incurs continuous switching costs. My ADHD brain thrives on it, but it’s absolutely unhealthy and unsustainable. And I’d like to add a fourth mechanism to this.

Maintenance depth

If there’s one thing you learn when working with a development company, it’s that everything you build requires maintenance. And with AI, you tend to build a lot: smart, custom software that simplifies your work; sleek, streamlined AI processes that let you complete your routine tasks even faster. And every time you finish something like that, there’s room for new features and new automated processes. But everything you build also needs to be maintained, and that’s often forgotten. And the time and costs associated with that maintenance keep increasing as you automate more and more of your work.

The value of low-intensity work

AI reduces the costs of writing, analyzing, and coding. This drives more production, more tasks, and more output. I don’t want to do less in the same amount of time. I want to get more done.

At the same time, low-intensity work is disappearing from the workday. The routine tasks that required little cognitive effort but gave the brain a chance to recover. Because AI is taking over everything that is simple. Meanwhile, high-intensity work is piling up. Taylor already observed that people could handle physical strain for about 42% of a workday—back then, about four hours. Our brains have similar limits. And those limits haven’t shifted.

AI brain fry: when oversight costs more than it’s worth

The BCG study quantifies the cognitive costs. Employees who had to intensively monitor and correct AI tools (the standard mode for agentic AI) reported 14% more mental effort, 12% more mental fatigue, and 19% more information overload than those who used AI to take over tasks without intensive supervision.

Productivity was found to follow an inverted U-curve. With one to three AI tools, productivity increases. Starting at four tools, the increase levels out and a dip occurs. Not because the tools don’t work, but because managing them costs more than they deliver.

14% of AI users in the study experience brain fry. In marketing, where AI output is continuously evaluated and adjusted, that figure rises to 26%. The consequences are concrete: 39% more major errors, 33% more decision fatigue, and a 39% higher intention to leave.

One senior engineering manager put it this way: “It’s like having twelve browser tabs open in my head. I catch myself rereading the same things, doubting my own conclusions much more, and becoming bizarrely impatient. I’m working harder on managing the tools than on actually solving the problem.”

Read that again. “I’m working harder on managing the tools than on actually solving the problem.”

How this affects your brain in the long term

MIT researchers published a preprint on the long-term effects of regular ChatGPT use. People who wrote essays exclusively using ChatGPT for four months showed significantly lower brain connectivity than those who wrote without any tools or using a search engine. The researchers call this “cognitive debt”: when you habitually rely on external systems, they replace the cognitive processes you would otherwise develop for independent thinking. It’s neuroplasticity in reverse.

This is distinct from brain fry. Brain fry is acute overload.

Cognitive debt is insidious and cumulative.

Two sides of the same problem: excessive AI use tires the brain in the short term and diminishes it in the long term.

The focus time crisis that already existed

All of this is taking place in a work environment that, even before AI, structurally lacked sufficient space for deep, concentrated work.

Research by Worklytics shows that the median knowledge worker has 3.2 hours of focus time per day. Only when they exceed 3.5 hours do employees report themselves as productive. Gloria Mark of UC Irvine demonstrated that after a single interruption, it takes an average of 23 minutes for someone to fully refocus. Microsoft’s Work Trend Index 2025 states that employees are interrupted every two minutes during core hours, about 275 times a day. Asana’s research concludes that 60% of work time is spent on “work about work”: tracking statuses, searching for information, switching between tools.

AI adds to this landscape: more tools to monitor, more output to review, more threads to keep track of. Collaboration in meetings and via email has already increased by more than 50% over the past two decades. AI is now adding to that.

Subtracting instead of adding

So how do you do it right?

McKinsey puts it bluntly in its report Agents, Robots, and Us: Adding AI to existing work processes yields “at best marginal improvements.” Real productivity gains require redesign. Workflows in which people and AI each do what they do best.

Goulmy adds a behavioral psychology observation to this: people are structurally inclined to add something rather than remove it. In a classic experiment by The Behavioral Scientist, subjects were asked to straighten a crooked LEGO bridge. Most added a block. Only when it was explicitly stated that removing a block was free did 61% remove a block.

Organizations are doing exactly the same thing. Adding AI to existing processes, existing meeting culture, existing expectations about output. Without asking which blocks can be removed.

This actually calls for a radical redesign of your company and its processes. And as the cognitive load increases, rest periods must necessarily increase as well. Fewer working hours, more break time. Because working four days a week isn’t the solution. Five days, or better yet, four shorter days. I believe that’s the direction we need to head in.

What this requires of you and your organization

The studies point in the same direction. Ranganathan and Ye call it an “AI Practice”: deliberate organizational habits that counteract the natural tendency toward intensification. Specifically: built-in moments for reflection, structured work blocks to limit constant notifications, and space for human interaction that would otherwise be lost to AI-driven workflows.

BCG recommends clear boundaries for the use and oversight of AI tools, and warns that the tendency to keep adding more tools structurally increases cognitive overhead.

McKinsey also says: redesign the workflows themselves. AI is an opportunity to rethink how work is done, how skills are deployed, and how roles are defined.

Goulmy puts it most bluntly: stop adding things before you’ve removed anything. Systematically convert the productivity gains AI delivers into time savings, and ensure those gains benefit those who work with them. That increases adoption, reduces absenteeism, and makes growth sustainable.

The real question

AI increases what people are capable of. The question is where that extra capacity goes. Toward more output, higher expectations, and longer workdays? Or toward greater autonomy, better health, and smarter growth?

So far, most organizations, often unconsciously, have chosen the first option. I believe it’s time to make different choices.

Email hasn’t reduced our workload. The car hasn’t brought us closer together. AI won’t make the workday any easier, unless you actively and explicitly choose that path.

The default is always more. Unless you decide that enough is enough. So I’m heading out to the garden now to see how the grass is growing, because I’ve seen enough screens for today.

Thanks to Louis for the inspiration.