Here’s what’s on our plate today:

• 🧪 How AI is fueling a new wave of workplace burnout.

• 📰 OpenAI's teen-safety scrutiny, India rejects Musk's Starlink claim, Microsoft's Surface Laptop Ultra.

• 💡 Roko's Pro Tip: change the target when you add the tool, or you're buying burnout.

• 🗳️ Poll: why are AI's biggest productivity winners the most burned out?

Let’s dive in. No floaties needed…

The Laboratory

TL;DR

AI tools were meant to lighten work, but the people using them most are the most burned out.

• The finding: Upwork research found that 77% of employees using AI say it added to their workload, and a 2025 follow-up found that 88% of workers reporting the biggest AI productivity gains are burned out.

• The conversion: companies kept their old targets, so the time AI saved turned into more output, more checking, and more retraining.

• The spillover: unchecked AI output, or ‘workslop,’ costs colleagues about two hours per case and a 10k-person organization roughly $9M a year.

• The test: 56% of freelancers, who set their own terms, don’t struggle with productivity demands, compared with 35% of full-time employees.

• The stakes: the pandemic forced new terms of work after people quit, and companies still have time to redesign AI work before burnout becomes the cost of adoption.

How AI is fueling a new wave of workplace burnout

When the pandemic shut offices in 2020, the laptop and the home internet connection became the whole workplace, and work carried on from spare rooms and kitchen tables. However, since that connection never switched off, the line between the working day and the rest of life faded under months of isolation and constant availability. People burned out, and the burnout eventually showed up as mass resignations and as a quieter disengagement among those who stayed. Employers kept the laptops and changed the terms of work around them, offering flexible hours, judging staff on results, and treating mental health as part of keeping people. The technology that had worn people down stayed where it was, and whatever relief came, it came partly through those new terms.

Office AI is now moving through the same cycle, beginning with a promise that software would take over dull jobs like drafting emails, summarizing documents, and writing code. The time saved was supposed to go to the work that needs human judgment. However, since most companies dropped AI into the same workloads, targets, and habits they already had, the time it saved has turned into more work, and a new kind of exhaustion has come with it. The burnout comes from how AI is being brought into work far more than from the tools themselves. A problem of that kind is bigger than any one survey, because it describes a management choice that any company adopting AI can repeat.

The clearest sign of that choice came from Upwork, the freelance hiring platform, whose Research Institute studies how work is changing. In 2024, it found that 77% of employees using AI said the tools had added to their workload, close to the opposite of what the tools were bought to do. A follow-up report from Upwork found something stranger: the workers reporting the biggest productivity gains from AI were also the most burned out, at 88%, and twice as likely to think about quitting.

If AI were simply bad at its job, the people getting the most out of it would be the least worn down, and the data shows the reverse. The tool appears to be delivering what it promised, and that pushes the question back onto what companies do with the time it saves.

Companies turn the time AI saves into more output and more checking

Kelly Monahan, managing director of the Upwork Research Institute, located the problem in the setup around the tool. “Our research shows that introducing new technologies into outdated work models and systems is failing to unlock the full expected productivity value of AI,” she said. The targets and routines of the old job stayed in place after AI arrived, so the time it saved was quickly spoken for by a request for more output.

Anyone who has worked alongside a fast but inexperienced junior hire will recognize what happens next. The new hire’s speed persuades the manager that the team can take on more, while every draft the hire turns in still needs someone to read it through. Upwork found employees spending their time in much the same way, producing more because AI makes more output possible on paper, reviewing what the tool produced, and learning tools that keep changing. The hour AI saves on a draft ends up spent on a higher target, on checking, and on retraining. And the checking doesn’t always stay with the person who used the tool, because a draft nobody checks becomes the problem of whoever receives it next.

Unchecked AI output moves the cost onto colleagues

Researchers at Stanford’s Social Media Lab, working with the coaching platform BetterUp, called that problem ‘workslop,’ their name for AI-generated work that looks polished but lacks substance. It turns up as slides, long reports, and code that a colleague then has to verify or fix. In a survey of more than 1k full-time U.S. desk workers, published in September 2025, 40% said they had received some of it in the past month. Each case took an average of about two hours to resolve, and the researchers put the cost for a 10k-person organization at roughly $9M a year.

The lost hours are only part of the price, because about half of those workers said they saw the colleague who sent the work as less creative, capable, and reliable afterward. The junior hire’s unchecked drafts are being passed down the line, and the damage reaches past the calendar into how people at work regard one another. All of these costs fall on employees working to targets someone else has set, so the fairest test is to look at people who set their own terms.

Workers who set their own terms feel less pressure from the same tools

Upwork’s 2024 study ran that comparison between freelancers and full-time employees, and found that 56% of freelancers did not struggle with the productivity demands placed on them, against 35% of full-time employees. Freelancers choose their own tools, run their own workflows, and are judged on outcomes, so nobody hands them a new tool and a higher target at the same time.

A freelancer works with the same fast junior hire as the employee, but gets to decide alone how much extra work its speed should buy. The same technology produced less difficulty with productivity demands under different terms, which places the burnout in the arrangement around AI. None of this makes the tools harmless, though, because a separate line of research suggests that leaning on AI heavily can wear down people’s thinking, whatever the arrangement.

Heavy dependence on AI raises a separate concern about thinking

Researchers at MIT’s Media Lab, which studies how people use technology, had 54 participants write essays using ChatGPT, a search engine, or nothing at all. Each participant wore an EEG headset, a cap that records electrical activity in the brain, and the ChatGPT group showed the lowest engagement across brain, language, and behavior measures. Teachers who graded the essays had their own words for the results, describing them as largely “soulless.”

The pattern deepened across sessions, and by the third, many ChatGPT users were only typing prompts and making light edits. When they were later asked to work without tools, they struggled to recall what they had written. The researchers call this ‘cognitive debt,’ the idea that thinking handed to a tool never becomes the person’s own. If that effect were driving the exhaustion, burnout would follow AI into any workplace, however it was managed, and the obvious answer would be to restrict the tools.

The worry holds for heavy dependence, although the MIT team itself concluded that careful use can support learning and that the risk lies in early or excessive reliance. Banning the tools hasn’t solved the problem either, since a 2025 survey by Anagram, a security training company, found that 45% of workers had used banned AI tools at work. And 40% said they would knowingly break company policy to finish a task faster. People reach for a banned tool when the pressure to produce outweighs the rule against it, which points straight back at the targets.

AI has not yet had its pandemic renegotiation

Office work went through this cycle once already, when the pandemic’s burnout ended in a renegotiation with the connection still running and the rules rewritten around it. AI is at the point just before that renegotiation, with the tools in place, the old targets unchanged, and the workers gaining most from it already twice as likely to consider quitting. The freelancers hint at what comes after it, since the same tools caused less difficulty with productivity demands once people set their own pace. The pandemic forced employers to rewrite the terms of work after people started leaving. AI is creating a similar pressure, but companies still have time to change the targets, review time, and expectations before burnout becomes the cost of adoption.

Roko Pro Tip

Bite-Sized Brains

• OpenAI faces teen-safety scrutiny: OpenAI is under renewed pressure over how ChatGPT handles vulnerable teens in crisis, intensifying the debate on safeguards for minors using AI chatbots.

• India rejects Musk's Starlink claim: India dismissed Elon Musk's claim of discrimination over Starlink's delayed launch, pushing back on his account of the holdup.

• Microsoft's Surface Laptop Ultra: Microsoft unveiled the Surface Laptop Ultra and a slate of hardware at its Windows event, doubling down on AI-first PCs.

Monday Poll

Meme Of The Day

The Toolkit

• Mirage: AI video generator that turns prompts or photos into cinematic clips with multimodal foundation models.

• Speechmatics: Speech-to-text engine built for accuracy across accents, noisy audio, and real-time use.

• Superhuman: AI-powered email client that drafts replies, summarizes threads, and gets you to inbox zero faster.

Rate This Edition