In December, I wrote about the AI Productivity Paradox--the difference between the gains businesses were promised from generative AI and the messy reality. But new research reveals a more subtle problem. Even when AI works well, it doesn't necessarily make work lighter. It can make work more intense.

In a study published last month, researchers from UC Berkeley spent eight months studying how generative AI changed work habits at a tech company with about 200 employees. The company didn't mandate AI use but paid for 'pro' subscriptions. The result? Nobody went home early. They just did more at a faster pace. As one engineer put it: “You had thought that maybe because you could be more productive with AI, then you save some time, you can work less. But then really, you don’t work less. You just work the same amount or even more."

The Treadmill Effect

The researchers identified three patterns. First, task expansion: because AI fills knowledge gaps, workers started doing things that used to belong to other people. Product managers wrote code. Researchers took on engineering tasks. People absorbed work that might previously have justified additional hires.

Second, blurred boundaries between work and free time. Because prompting AI feels more like chatting than working, people slipped tasks into lunch breaks, meetings, and evenings. Work became something that can be nudged forward with a simple chat message to your friendly AI coworker.

Third, more multitasking: workers managed several AI-assisted threads at once, creating constant switching between a growing pile of open tasks.

These patterns created a self-reinforcing cycle where AI accelerated tasks, which in turn raised expectations for speed. Higher speed made workers more reliant on AI, which widened the scope of what tasks and projects were assisted by AI. This, of course, generated more work. As the Red Queen told Alice, you have to run as fast as you can just to stay where you are. For some, AI has become that Red Queen who is constantly raising the baseline for what counts as productive.

I’m on the Treadmill Too

I certainly recognize this in myself. Having invested the time to get genuinely great results with AI, I haven't used my AI skills to work less. I’ve used it to build new software, develop 3D model designs, and revive old projects I’d abandoned years ago because I didn’t have the programming skills to pull them off. That’s exciting—but it’s also more work, not less. But there's a big difference between an old slacker-hacker like me choosing to take on more side projects and an entire workplace where 'doing more all the time' becomes the new normal.

The larger problem is that innovation requires slow thinking: talking through a problem with a colleague, sketching ideas on a whiteboard, letting your subconscious chew on it overnight. The danger of heavy AI use is that it moves everything into the 'fast' pile. When you can generate a draft in seconds, the temptation is to ship it and move on, and skipping the slow thinking where better ideas can emerge.

Getting Off the Treadmill

So what do we do? The Berkeley researchers suggest leaders build what they call an 'AI practice' describing how the employees will use AI and, just as importantly, when they'll stop. Pause before barreling ahead to make sure you're still working on the right thing. Batch AI outputs instead of reacting to each one the moment it pops up. And protect time for human conversation.

I'd add one more: look at what you're rewarding. If the person who takes on the most tasks gets the most praise, you're feeding the treadmill. If you reward judgment, quality, and knowing when to stop, you're building something that lasts.

AI has already changed how we work. The question is whether you'll shape that change on your terms or let the treadmill set the pace.