AI is creating more work, not less

The pitch writes itself

AI saves time. It's the easiest sell in technology right now. Write emails faster, summarize meetings in seconds, generate reports that used to take hours. Every vendor deck leads with it.

The pitch works because it's partly true. AI does save time on individual tasks. What nobody's talking about is what happens to that time afterward.

Where the saved time goes

Researchers at the University of Lausanne tracked what happens when AI frees up hours in a workday. AI saved managers an average of three hours per week. But 36% of those managers wasted more than half of the time they saved. Among AI users broadly, 83% admitted to wasting at least a quarter of their newly freed hours.

"Wasted" is doing a lot of work in that sentence. What actually happens is closer to absorption. The freed hours don't convert into strategic thinking or better decisions. They get filled with more tasks. More emails drafted, more reports generated, more of the same work at higher volume.

A separate study using nearly two decades of American Time Use Survey data found the same pattern at a structural level. Workers in jobs with higher AI exposure aren't working less. They're working more. Moving from the 25th to the 75th percentile of AI exposure corresponds to an additional 2.2 hours of work per week. After ChatGPT's release, the gap widened: workers in high-exposure roles logged roughly 3.15 additional hours weekly compared to those in low-exposure positions.

The tool that was supposed to give people time back is, on average, taking it away.

The botholes problem

Bob Sutton, Stanford professor emeritus and organizational psychologist, has a name for something most knowledge workers have felt but couldn't quite articulate. He calls them "botholes": AI tools that make it effortless to waste other people's time.

The mechanics are straightforward. AI makes it trivially easy to produce polished output. A ten-minute prompt generates a document that looks thorough, reads professionally, and could have taken a full day to write by hand. The problem is that someone on the receiving end still has to read it, evaluate it, and figure out whether it actually says anything. The production cost dropped to near zero. The consumption cost didn't change.

Sutton describes this as the "weaponization of friction." Before AI, the effort required to create a document acted as a natural filter. If something took four hours to write, you thought about whether it was worth writing. That filter is gone. What's left is a flood of what Sutton calls "high-gloss workslop" landing in inboxes and channels across every organization that adopted these tools.

The $9 million tax

Stanford and BetterUp put a number on the workslop problem. In a study of over 1,100 workers published in September 2025, they found that 40% had received AI-generated output from colleagues that required significant rework. The average time spent dealing with each instance: one hour and fifty-six minutes.

That math scales badly. For an organization of 10,000 people, the workslop tax adds up to roughly $9 million in lost productivity per year. Not from AI failing at its job. From AI making it easy to create work for other people.

This is the number that should concern mid-market executives. Your teams are producing more since AI adoption. But more isn't better. It's often just more.

Volume is not quality

The dominant pattern across AI adoption is an increase in output volume, not output quality.

Look at what people actually use AI for. When Harvard Business Review surveyed the top 100 use cases in early 2025, production tasks dominated: writing emails, generating reports, drafting documents. The measure of success is "did something get created?" not "was the thing that got created any good?"

Stanford HAI co-director James Landay predicts that in 2026, "we'll hear more companies say that AI hasn't yet shown productivity increases, except in certain target areas like programming and call centers." His colleague Angèle Christin puts it plainly: in many cases, the real impact of AI is "some efficiency and creativity gain here, some extra labor and tedium there."

The organizations getting value from AI aren't the ones generating more output. They're the ones that recognized early that when the cost of producing something drops to near zero, the only question that matters is whether the something is worth producing.

The early-career signal

There's a harder data point in the research. Erik Brynjolfsson and colleagues at the Stanford Digital Economy Lab found that employment for workers in their early twenties declined by roughly 13% in jobs with high AI exposure after ChatGPT launched.

These are the entry-level positions where people learned by doing the work that AI now handles. First drafts. Research summaries. Data assembly. The work wasn't valuable because of the output it produced. It was valuable because of what it taught.

When AI absorbs those tasks and the organization measures success by total output, the short-term numbers look good. The long-term bench strength is a different story.

The question that matters

The research points in one direction. AI's time savings are real at the task level. What happens to those savings at the organizational level is less clear. Saved time mostly converts to higher-volume work, not higher-value work. AI-generated output creates new consumption costs that offset production gains. Workers in AI-heavy roles are logging longer hours, not shorter ones.

For companies in the middle of AI adoption, this reframes the conversation. The important question isn't "how much time can AI save us?" It's "what are we going to do with the time it saves?" And if the honest answer is "produce more of the same, faster," the follow-up question is worth sitting with: who is going to consume all of it?