AI Burnout at Work Starts With a Decision Nobody Made

A COO I advised called me on a Friday evening, still at her desk. Her company had rolled out AI tools to every team six months earlier. The internal survey said people were saving about four hours a week.

“So where did those hours go?” she asked. “Nobody looks less busy. And three of my best people just asked about leave.”

I had no answer that night. I do now. AI burnout at work rarely starts with the technology. It starts when saved time arrives and no one decides what it is for.

Why AI Burnout at Work Is a Capacity Problem

Time saved by a tool is capacity, and capacity behaves like cash. If nobody assigns it, it gets spent by whoever reaches it first. Those claimants are usually urgent requests, not important ones.

Cyril Northcote Parkinson wrote in The Economist in 1955 that work expands to fill the time available for its completion. He was joking about civil servants. I now watch the same law run at software speed.

The strongest evidence I have seen comes from the Berkeley Haas School of Business. In February 2026, Harvard Business Review published an eight-month study by Aruna Ranganathan and Xingqi Maggie Ye. They followed about 200 employees at a US technology company and interviewed 40 of them in depth.

The company never required anyone to use AI. It simply gave everyone a subscription. Even so, people worked faster, took on a wider range of tasks and stretched their hours. The researchers call the pattern workload creep, and they reported rising burnout, anxiety and decision paralysis as the months passed.

Nobody in that study was forced into anything. That detail matters, because it removes the easy villain. No cruel manager, no impossible target. Just a tool that made one more task feel cheap.

When I looked at the COO’s team, I asked to see one senior analyst’s calendar. Her four saved hours had become two extra client deliverables a week and a standing 7 am check-in she had added herself. She called it “getting ahead.” Her body called it something else.

Does AI Reduce Workload? What the Research Shows

Not by default. Research from the Berkeley Haas School of Business found that employees with AI tools worked faster but also took on more tasks, longer hours and more multitasking. Saved time was reinvested in extra work, not rest. AI lowers the effort of each task, yet total workload rises unless a leader decides otherwise.

That answer surprises executives, because the pitch for AI was always about relief. The relief is real at the level of a single task. A draft that took an hour takes ten minutes.

But a person does not experience tasks one at a time. They experience a day. And a day with cheaper tasks is a day that invites more of them.

The Three Ways Saved Time Disappears

The Berkeley researchers described three mechanisms. I have seen all three in the teams I advise, and each one hides in plain sight.

The first is task expansion. When AI makes unfamiliar work feel easy, people reach into other roles. A product manager starts writing code. A marketer starts building dashboards. It feels like initiative, and often it is.

Still, the cost lands somewhere else. The engineer who now reviews AI-assisted code from a colleague has gained a job nobody wrote down. One person’s expansion becomes another person’s review queue.

The second is the loss of natural breaks. Before AI, a hard question meant a pause. You walked down the hall, waited for a reply or stared out of the window. Now an answer arrives in seconds, so people fill lunch, the gap between meetings and the first ten minutes of the evening with quick queries.

Those pauses were doing work. They were where people recovered, and where half-formed ideas finished forming.

The third is multitasking. Cheaper tasks mean more tasks in flight. Sophie Leroy, a researcher at the University of Washington, described “attention residue” in 2009: part of your mind stays on the previous task after you switch. Gloria Mark of UC Irvine has found that people need over 23 minutes, on average, to return fully to a task after an interruption.

Do the arithmetic on a day with twelve small AI-assisted tasks. The time saved on each one is spent many times over on the switching.

What Causes Burnout From AI at Work?

The World Health Organization describes burnout as an occupational phenomenon with three parts: exhaustion, growing mental distance or cynicism toward the job, and reduced effectiveness. It does not call it a medical condition. It ties it to chronic workplace stress that has not been managed.

That last phrase is the one I return to. Christina Maslach and Michael Leiter spent decades showing that burnout follows mismatches between a person and their job in six areas: workload, control, reward, community, fairness and values. AI can strain several of these at once.

Workload rises through creep. Control drops when the pace of the tool sets the pace of the day. Community thins when a person asks a chatbot instead of a colleague. Fairness suffers when some people absorb the review work that others create.

None of this requires a bad tool or a bad employee. It requires only that the gains from AI are spent by default, rather than by design.

Why Teams Rarely Push Back on the Extra Load

There is a quieter reason the pattern holds, and it took me a while to see it. Nobody wants to be the person who says the tool did not help.

I advised the founder of a forty-person consultancy who announced an “AI-first” year at a company retreat. Three months later he told me his team seemed thrilled. I asked one of his senior consultants, off the record, how she felt. She said the tool was excellent and her week was worse. Then she asked me not to repeat it.

She was afraid of looking slow. Everyone around her was producing more, so producing the same amount felt like falling behind. That is how a ratchet works: each person’s new pace becomes the next person’s baseline, and nobody ever votes to turn it back.

Psychological safety decides whether you hear the truth in that moment. If your team cannot say that AI made their workload heavier, your data will tell you the opposite. The survey will say people save four hours a week. It will not say what they paid for them.

So the first signal is rarely a complaint. It is a change in how people talk about their evenings, their lunch breaks and the work they used to enjoy. I listen for that, and I ask about it directly, in one-to-one conversations, before any survey goes out.

Why Executives Miss the Warning Signs

Gallup’s 2018 study of nearly 7,500 full-time employees found that 23 percent reported burnout at the top two levels of its frequency scale, and another 44 percent reported it sometimes. Burnout was already common before generative AI reached the office. Microsoft’s 2023 Work Trend Index, which surveyed 31,000 people in 31 countries, found that 64 percent struggled to find the time and energy to do their job, and 68 percent lacked enough uninterrupted time to focus. Layering a pace multiplier on top of an exhausted workforce is not a neutral act.

So why do leaders miss it? Because the dashboard rewards what burnout produces, at least at first. Output goes up. Response times fall. Everyone looks responsive.

I worked with a CEO who told me, with real pride, that his leadership team now answered messages within minutes at any hour. I asked what he would do if one of them stopped. He did not have an answer, and I think he heard the problem himself.

Leaders are also the heaviest users. When you are the one with the most access, the most context and the most appetite, your own pace becomes the unwritten standard. People copy it. They do not wait for permission to be exhausted.

A Calendar Test for Your Own Load

Here is a test you can run this week. Open your own calendar and count the commitments you added since AI arrived: the new check-ins, the extra reviews, the side projects that now feel small enough to say yes to. Then ask whether you would have accepted that load a year ago.

Most executives I ask go quiet for a moment. They had not counted. They had only felt the days getting tighter, and assumed that was the price of ambition.

It is not. It is a series of small yeses, each one made cheaper by a tool, and it can be reversed one decision at a time. A leader who removes one standing meeting and says why gives the whole team permission to do the same.

How Can Leaders Prevent AI Burnout?

Treat the saved time as a budget, and spend it on purpose. I give executives four moves, in this order.

First, make the capacity decision out loud. When a team reports hours saved, decide the split before the hours get spent. One version I like: a third goes to quality and review, a third to recovery and learning, and a third to new work. The exact ratio matters less than the fact that someone chose it.

Second, count the invisible work. Ask who now reviews, corrects or cleans up what AI produces. If task expansion has created review queues, put them on the roster and give them time.

Third, protect the pauses. Batch notifications. Set windows when nobody expects a reply. The Berkeley researchers recommend intentional pauses and coordinated batches of work for exactly this reason, and I have seen it help on teams that tried it for a single quarter.

Fourth, change what you measure. Output and speed will flatter you. Track after-hours messages, error rates, review load and how many of your best people ask for leave. Those numbers move before resignations do.

Each of these moves costs little. The founder with the AI-first year adopted the first two within a week. His consultant told me later that the meeting where he asked where the saved hours had gone was the first time she felt the tool was working for her rather than the reverse.

None of this is anti-AI. I use these tools every week. A tool that makes work cheaper also makes it easier to take on too much of it, and only a leader can set the limit for a team.

A Better Question for Monday

When my COO client finally ran the numbers, her team had not lost four hours. They had spent them, and then borrowed some from their evenings. She changed one thing: every team lead now states, in the weekly review, where their saved time went and who agreed to it.

Within two months the leave requests stopped. Output dipped slightly, then recovered, and the errors that had crept into client work fell.

So here is the question I would put to your next leadership meeting. If AI gave your team back four hours this week, who decided what happens to them?

If you want a practical way to bring AI into your workflow without letting it quietly set your team’s pace, I wrote The Practical AI Playbook with that tension in mind: https://content-flair.com/the-practical-ai-playbook/