I read a Harvard Business Review article this week titled “AI Doesn’t Reduce Work—It Intensifies It.”

The researchers studied a 200-person tech company over eight months after it gave employees access to generative AI tools. No one was forced to use them. There were no top-down mandates. People adopted AI on their own.

Instead of working less, employees ended up working more.

They moved faster, took on a broader scope of tasks, and let work spill into more parts of their day. What struck me is that this expansion wasn’t driven by management pressure. It happened because AI made doing more feel easy and accessible .

And that part feels very real.

When I use AI, starting something is easier. I don’t procrastinate on the blank page the same way I used to. I don’t delay research because I know I can get a decent first pass in seconds. The friction at the beginning disappears, and that changes behavior more than we probably admit.

But removing friction doesn’t just speed up existing work.

One of the main findings in the study was what they call “task expansion.” Product managers started writing code (recipe for disaster?). Designers took on engineering responsibilities. Researchers absorbed technical work they might previously have handed off .

AI filled in just enough of the knowledge gaps to make people feel capable.

I’ve noticed the same thing in my own workflow. I’ll take on projects that, two years ago, I would have outsourced without thinking twice. I’ll tell myself, “It’s fine, I can figure this out with AI.” And most of the time, I can...

There’s also a second-order effect the article highlights. When non-engineers start producing AI-assisted code, engineers spend more time reviewing and correcting it . So the work doesn’t disappear, it shifts, spreads, and sometimes multiplies.

I’ve caught myself doing exactly this. I’ll think I’m just refining an idea quickly, and suddenly I’m deep into revisions at a time I didn’t plan to be working. It doesn’t feel heavy in the moment, but over time you realize your day has fewer clean breaks.

The third pattern they found was more multitasking . Employees would run multiple AI threads in parallel, revive old projects because “AI can handle it,” and constantly check outputs while working on something else. It creates a sense of momentum, but also a steady layer of cognitive load.

That part resonates the most with me.

AI makes it extremely easy to have five projects partially moving at once. You’re drafting here, researching there, refining something else in the background. On paper, it looks efficient. In practice, your attention is fragmented across more surfaces than before.

Now, here’s where I slightly disagree with the dramatic framing of “intensification.”

I don’t think AI is inherently making work more oppressive. I think it’s lowering the cost of action. And when action becomes cheap, we take more of it.

Before AI, friction acted as a filter. If something was hard to start, you were selective about what you chose to begin. Now that the barrier is low, the natural constraint disappears. You can start almost anything. So you do.

The problem is that depth still requires focus, judgment, and time. AI helps you get to a decent first draft faster, but it doesn’t decide which projects deserve your full attention. It doesn’t protect your calendar. It doesn’t tell you to stop.

So what looks like productivity can turn into scattered progress across too many threads.

The researchers suggest companies create intentional “AI practices” with pauses, sequencing, and structured norms . That makes sense at an organizational level.

Individually, though, I think this becomes a question of discipline.

And if anything, the real skill now isn’t prompting better. It’s deciding what not to prompt at all.