One could be forgiven for thinking that automation tools would make difficult tasks redundant, and make work more relaxing in general. But this eliminates an important law of the universe: the ratchet of productivity turns only one way. In other words, it’s a modern reality that when automation — AI or otherwise — makes any kind of positive change in your work life, you’ll feel some kind of squeeze, and more work will happen to erase any momentary sense of relief.

According to a case study highlighted by some “in-progress research” from Aruna Ranganathan, who teaches management at UC-Berkeley and Xingqi Maggie Ye, a Ph.D. student who is part of Ranganathan’s Berkeley program, AI is “intensifying” work, and certainly not making people’s days any easier.

It sounds, in other words, like hell on earth.

If that is, paradoxically, what you are want in your workday, then you probably work somewhere like Silicon Valley, or even OpenAI, where CEO Sam Altman describes AI’s ability to power his own work in ways that surprise and humble him (though he expresses little or no regret about his ambition to eliminate the jobs of knowledge workers). “I don’t think I can come up with ideas that quickly,” he said said in an interview last Octoberadded “I think it means that things happen faster and that you can … that you can try more things, and find out the best ideas quickly.”

Altman’s experience can be heard in the workers mentioned in article on Ranganathan and Ye’s research for the Harvard Business Review. They describe an eight-month study of the effects of generative AI on the working life of a company with about 200 employees. Employees “work at a faster pace,” the authors wrote, cover a “wider range of tasks,” and find themselves working “more hours during the day, often without being asked to do so.”

This is a workplace that, Ranganathan and Ye explain, does not mandate the use of AI. It simply makes business AI tools available. It’s not like a 200-person workplace where widgets are strung together. However, many of the roles described in the article involve engineering, writing code, and communication in Slack, so it’s safe to say that these are knowledge workers and software engineers, who probably use tools like Claude Code.

Because of AI, many of Ranganathan’s and Ye’s subjects seem to be starting to expand the scope of their jobs, taking over each other’s roles, and taking on the roles of teaching others to code, or correcting their vibe-coded work. Hiring new employees may be postponed or avoided altogether, as employees “absorb work that previously justified additional help or headcount.”

Workers are also, it seems, pretending to feed tasks into their AI devices while they’re supposedly in meetings, and submitting prompts while taking breaks, while waiting for things to load, or while they’re having lunch.

How you interpret this case study will vary. If your workplace is a startup in “founder mode” and everyone in your office is working punishing hours in exchange for equity in a company that everyone hopes will become a unicorn, I think you might like the sound of it—especially if you’re a CEO/founder and you plan to become a billionaire.

That is far from a universal experience, however.

According to a 2024 Pew survey, approx HALF of US workers reported that they were somewhat satisfied or “not very/not at all satisfied,” and the other half said they were “extremely/very satisfied.” That group that is “very satisfied” decreases from 50% to 42% if the respondent has a low income.

The survey also found that far and away the most satisfying aspects of a job according to respondents are other people, with 64 percent reporting being “extremely/very satisfied” with their relationships with their co-workers. Skill development, on the other hand, ranked low, with 37 percent reporting being “extremely/very satisfied” with that aspect of a job.

So I don’t get the impression that fewer people, need to learn to do more things, and work that enters the breaks will help most people’s job satisfaction, but maybe I lack a specific vision.

In other words, if instead of building an app, you are a person who works as, say, a hospital receptionist or a school administrator, maybe you are not all that worried about a hypothetical where the hiring is posted, you have to do other people’s jobs, you work on your breaks, and instead of getting new, helpful software, you get business AI tools to do it. create your own software.

But let’s not assume that all tech workers love this kind of productivity theater, or that the sense of greater productivity in Ranganathan and Ye’s case study is necessarily anything but an illusion. An anonymous worker at cybersecurity firm Crowdstrike wrote the newsletter Blood in the Machine last year, and said that the company’s workers were “motivated to handle more per capita workload simply by working harder and sometimes working longer without additional compensation,” and that “While our Machine Learning systems continue to perform with excellence, I am not yet convinced that our use of genAI has been productive in the context of proofreading, troubleshooting, and general maintenance.”

According to this person, “The result is not a lightening of the burden as often promised,” and “Morale is at an all-time low.”