AI and burnout: what the research says
So far, research doesn’t show that AI tools cause burnout by themselves. It does show, fairly consistently, that they change work in ways linked to burnout: more work instead of less, more reviewing and switching between tasks, and fewer real pauses. A 2026 review of 43 studies found that whether AI goes with more or less burnout depends on what it does in the job: taking routine work off people tends to help, adding oversight and demands tends to hurt. For developers, the one peer-reviewed study we found (442 developers, 2026) links generative AI to more burnout through higher job demands.
- The studies at a glance
- AI adds work instead of saving it
- Overseeing AI is tiring in its own way
- Developers: more demands, more review
- Feeling faster than you are
- How many are burned out?
- When AI helps
- What this means day to day
- Reading the headlines
- Questions
The studies at a glance
| Study | Who they asked | What it found | What it doesn’t show |
|---|---|---|---|
| Fang, Han and Ma, Frontiers in Psychology, October 2026 | 43 peer-reviewed studies | Assistive AI mostly goes with less burnout; monitoring and control systems with more; generative AI is mixed | Mostly cross-sectional studies, so little about cause |
| Feng, Afroz and Sarma, ICSE 2026 | 442 developers from open-source communities | Generative AI raises burnout by raising job demands; resources and a positive view of the tools soften it | A one-time survey: a model of cause, not an experiment |
| Ranganathan and Ye, UC Berkeley Haas, in HBR, February 2026 | Eight months inside a 200-person tech company, 40+ interviews | AI widened jobs, crept into breaks and had people running several AI threads at once | One company; no burnout measure |
| BCG, in HBR, March 2026 | 1,488 US workers | 14% reported “brain fry” from heavy AI oversight; offloading routine work went with lower burnout | Self-reported; the authors treat brain fry as separate from burnout |
| DORA, Google Cloud, 2025 | Nearly 5,000 technology professionals | People using AI more had similar burnout to matched peers | An association, not an effect |
| METR, 2025 and 2026 | 16 experienced open-source developers, then 57 | In 2025, 19% slower with AI while believing they were faster | Speed, not burnout; the 2026 follow-up is, by METR’s own account, unreliable |
| Faros AI, 2025 | Telemetry from 10,000+ developers | More tasks and pull requests on high-AI teams, and review time up 91% | No wellbeing measure; a vendor’s data |
| Upwork Research Institute, 2024 and 2025 | 2,500 people a year, 625 of them full-time employees | 77% of AI-using employees said AI added to their workload | Self-report, commissioned by a vendor |
AI adds work instead of saving it
The most detailed look at what AI does to a working day comes from Aruna Ranganathan and Xingqi Maggie Ye at UC Berkeley’s Haas School, who spent eight months inside a 200-person US tech company where staff could use generative AI freely. Think of buying a dishwasher and then inviting twice as many guests. They describe three ways work grew rather than shrank: people took on tasks that used to belong to others; prompting slipped into lunch, the minutes before meetings and the evening, because it felt quick; and people ran agents in the background while reviewing code or sitting in meetings, sometimes several at once. The result, in their account, was constant switching and less recovery, which they warn can impair judgment and increase errors.
Other data points the same way. Anthropic’s study of its own engineers (132 surveyed, 53 interviewed, December 2025) found that they estimated 27% of their Claude-assisted work wouldn’t have been done at all without it: new work, not saved time. DORA’s 2024 report found less time spent on valuable work as AI use rose, and its authors suggested freed time simply fills with other tasks.
Overseeing AI is tiring in its own way
A Boston Consulting Group survey of 1,488 US workers, published in Harvard Business Review in March 2026, named a separate kind of strain: “AI brain fry”, mental fatigue from using or overseeing AI tools beyond what one can handle. About 14% said they’d experienced it. As reported by The Register and Fortune, it was worse with heavy oversight of AI output and with more tools in use, and it went with more self-reported errors and more intent to quit. The same survey found that people who used AI to offload routine work had lower burnout scores. The authors present brain fry as cognitive fatigue, distinct from burnout, and the full article is paywalled, so treat the secondary figures with care.
Developers: more demands, more review
The strongest developer-specific study is by Feng, Afroz and Sarma at Oregon State University, presented at ICSE 2026. Surveying 442 developers, they found that generative AI adoption raises burnout by raising job demands, and that job resources and positive attitudes to the tools buffer the effect. It’s a single survey, so the causal direction is the authors’ model rather than an experiment.
Much of the extra demand is review. In Stack Overflow’s 2025 survey of about 49,000 developers, the top frustration with AI, named by 66%, was solutions that are almost right but not quite, and 45% said debugging AI-generated code takes more time. Faros AI’s telemetry from more than 10,000 developers found teams with high AI adoption completing 21% more tasks and merging 98% more pull requests, while review time rose 91% and there was no measurable improvement at the company level. Two 2026 preprints with software professionals add the less visible costs: worry about being accountable for code you didn’t write, and a weaker sense of craft. We go through these signs in developer burnout symptoms.
Feeling faster than you are
In METR’s 2025 randomized trial, 16 experienced open-source developers worked on 246 real issues in their own repositories, with AI allowed or not at random. With AI they took 19% longer, yet afterwards they believed it had made them about 20% faster. That gap matters for burnout because it invites taking on more. METR’s 2026 follow-up with 57 developers pointed toward a speedup with late-2025 tools, but METR itself calls that data an unreliable signal, partly because many developers declined to take part since they didn’t want to work without AI. Headlines that METR “reversed” its finding, or that developers are “still 18% slower”, both misread it.
How many are burned out?
The numbers vary with who’s asked and how, and none of them show that AI is the cause:
- Upwork Research Institute, 2024: 77% of employees using AI said it had added to their workload, mostly through reviewing AI output and learning the tools, and 71% of full-time employees described themselves as burned out. In its 2025 survey, 88% of the workers reporting the biggest AI productivity gains said they were experiencing burnout.
- LeadDev, 2025: 22% of 617 engineering leaders and developers showed critical levels of burnout on a Maslach-based scale, with no link to AI reported.
- State of Devs, 2026: 62% of about 5,300 developers said they’d had burnout at some point in their career, the same as the year before. Asked whether AI tools had a positive effect on their mental state, about 46% disagreed and 23% agreed.
All three rely on self-report, and Upwork sells freelance work, which suits a story about overloaded staff.
When AI helps
The same research has a hopeful side. The 2026 review in Frontiers in Psychology, covering 43 studies, sums it up best: AI’s links with burnout “depend on its functional role, implementation, and employee appraisal.” Assistive AI, mostly studied in healthcare, went with lower burnout; AI used to monitor and control people went with higher burnout. BCG found lower burnout among people who used AI to take routine work off their plates, a small German study found generative AI used as a resource linked to less burnout, and a 2026 field study (a preprint) found that using one AI mode at a time lowered perceived workload while mixing them in one task reduced the benefit. DORA’s 2025 report found similar levels of burnout in heavier and lighter AI users.
What this means day to day
If the problem is work that seeps into pauses and too many threads at once, the fixes are about boundaries more than tools:
- Keep pauses free of prompts. The Berkeley study’s clearest finding is that “quick” prompts ate the breaks. A break where you check on agents isn’t one.
- Limit what runs in parallel. Brain fry rose with oversight load. Run as many agents as you can actually review; how many sessions is too many has a rule of thumb.
- Give AI the routine work and keep the judgment, which is where the studies found AI helped.
- Break on a rhythm. The UK Health and Safety Executive advises short, frequent breaks from screen work, 5 to 10 minutes an hour rather than 20 minutes every two hours, and a 2022 meta-analysis found micro-breaks help with fatigue but that tiring work needs longer ones. See Pomodoro with AI coding agents.
- Set an end to the day, because an agent can always run one more task.
More on what helps is in why agent work wears you out.
Reading the headlines
- Survey, study or telemetry? Most figures are self-reported surveys, often by companies selling AI, analytics or talent. Telemetry measures activity, not wellbeing.
- Cross-sectional means correlation. People in overloaded jobs may use AI more, rather than AI overloading them.
- Brain fry isn’t burnout, and slower isn’t burned out. Check what was measured.
- Microsoft’s “infinite workday” (interruptions every two minutes for the most-interrupted users, more late meetings) predates agents, and Microsoft presents AI as the fix, not the cause.
Where Meowdoro fits
We built Meowdoro, a focus timer around the MacBook notch for people who work with Claude Code and Codex, around the problem the Berkeley study describes: during focus it shows which agents are working, done or waiting, and during a break it shows none of them, only the cat and the time, with one summary when you’re back (features). It won’t fix a job with too much in it; it does protect the break. Try it and tell us what’s missing.
Questions
Does AI cause burnout?
Not by itself, as far as research shows. What it does show is that AI often adds work, review and switching, which are linked to burnout, and that it can also lower burnout when it takes routine work off people. How it’s used matters more than whether it’s used.
What is AI brain fry?
A term from a 2026 BCG survey for mental fatigue from using or overseeing AI beyond one’s capacity. About 14% of the 1,488 workers asked reported it. The authors treat it as distinct from burnout.
Are developers who use AI more burned out?
One peer-reviewed study of 442 developers found AI adoption raised burnout through higher job demands. DORA’s 2025 report found similar burnout in heavier and lighter users. The evidence is mixed and mostly correlational.
Does AI make developers faster?
In METR’s 2025 trial, experienced developers were 19% slower with AI while feeling faster. Later data points to a speedup with newer tools, but METR calls it unreliable.
How often should you take breaks when working with AI?
Short and often: 5 to 10 minutes an hour is the UK HSE’s guidance for screen work, with longer breaks after tiring work. A Pomodoro rhythm of 25/5 or 50/10 fits that.
Related guides
Sources: Fang, Han and Ma, Frontiers in Psychology, 2026; Feng, Afroz and Sarma, From Gains to Strains, ICSE 2026; UC Berkeley Haas on Ranganathan and Ye’s study and HBR; HBR on BCG’s brain fry survey, with coverage by The Register and Fortune; DORA 2025 and 2024; METR’s 2025 study and 2026 update; Faros AI; Stack Overflow Developer Survey 2025; Anthropic; Upwork Research Institute and Workplace Intelligence, 2024 and 2025; LeadDev; State of Devs 2026; UK HSE; Albulescu et al., PLOS ONE, 2022; Microsoft WorkLab. Read on October 30, 2026.