- Employees using generative AI tools daily face a 37% higher rate of burnout than those who do not.
- Knowledge workers spend approximately 90 additional minutes per day on “the review tax” checking AI-generated outputs.
- Companies that cap output expectations and provide specific AI collaboration training report lower burnout rates.
Daily use of generative AI tools is contributing to worker burnout, according to the Slack Workforce Index, which found that employees using these tools daily report higher burnout rates than colleagues on traditional workflows.
The Hidden Cost of Infinite Drafts
When a tool can generate a 10-page report in 30 seconds, the bottleneck shifts from the writer to the reviewer. In many workplaces, output expectations have simply scaled up to match the speed of the AI. Employees who used to produce one polished presentation per week are now expected to produce five, but the human capacity for strategic thinking and final editing hasn’t moved at all. That mismatch is the core driver of the exhaustion the Slack findings describe.
According to the report, the survey spans finance, legal services and marketing, moving past early-adoption optimism to the messier reality of daily AI-augmented work.
Managing the Review Tax
Every paragraph generated by a large language model requires a human to check for hallucinations, tone problems and factual errors. The Slack report puts that process at roughly 90 minutes of focused attention per day. Original writing allows for creative flow; constant error-correction does not. It requires what the report calls “continuous partial attention”, a mode that blocks deep work and ramps up cognitive fatigue by mid-afternoon.
The burden is heaviest in sectors with low tolerance for error. In legal and financial services, time saved on initial drafting is often entirely consumed by the auditing required to ensure the AI hasn’t invented a case citation or a fiscal projection. For individual contributors, this can feel less like augmentation and more like being a caretaker for the output of a tool they didn’t ask for. Reduced autonomy over one’s work is a well-established precursor to burnout, and the June report indicates that sentiment is rising sharply among junior-level staff.
Microsoft 365 Copilot costs an additional $30 per user per month. Companies paying that premium want a return, which typically means higher quotas. When a manager sees that someone is “using AI,” the natural response is to assign more. The result is a workforce that is busier than before, doing work it finds less satisfying. The shift from creator to editor isn’t a transition everyone is ready to make, but the current software landscape offers little alternative. This dynamic sits at the heart of a broader debate about how Copilot is being positioned for workers across different business sizes.
Why More Tools Mean More Work
The average AI user often switches between multiple applications to complete a single task, fragmenting workflows rather than consolidating them and adding more “work about work” to the day.
The blank-page problem has been solved. The cleanup problem has taken its place. Prompting a tool to “summarise this meeting” is easy; the resulting summary is often a generic bullet list that misses the political undercurrents or unstated tensions of a real conversation. Fixing those gaps requires the worker to essentially relive the meeting while editing the text, a duplication of effort that rarely shows up in productivity metrics, which typically track only the speed of the initial draft.
Employees who received specific training on how to work alongside AI, keeping a human in the loop at defined points, report lower burnout levels, but such training is available at only a small share of companies.
Designing a Sustainable Workflow
Reading across recent product directions from Notion and Asana, there is a detectable shift in thinking: “more content” is not the answer. Newer iterations of these platforms are experimenting with summarisation layers and attention filters designed to shield workers from AI-generated noise. The aim is to move AI from content factory to a protective layer that flags only what genuinely needs a human decision.
For businesses, the more immediate fix may be recalibrating expectations. The Slack report highlights companies that cap output expectations rather than scaling them to AI speed, asking for one presentation that is better researched and better visualised, not five that are thinner on substance. That approach gives workers enough time to perform the review tax without burning out. It treats AI as a tool for quality, not volume.
Whether that becomes common practice is hard to verify from current public material. For now, the productivity trap remains the dominant experience for many knowledge workers. Without a change in how these tools are deployed, the burnout spike documented in the June Slack findings may prove to be an early data point rather than a peak. Explore more AI tools and tips in our Consumer AI section.
