The pitch from every AI tool on the market is the same: do more in less time. So why is everyone working more?
UC Berkeley researchers spent eight months embedded inside a 200-person company watching exactly what happened when teams started using AI tools voluntarily. What they found was concerning — people were doing more, in more time. Not because they were asked to. Because AI made it effortless to start the next thing.
Researchers observed three patterns consistently. Tasks expanded — AI made unfamiliar responsibilities feel accessible, so workers quietly absorbed work that once belonged to other roles, or would have justified additional headcount. Nothing was removed. Every role just grew. Boundaries blurred — with no friction at the entry point of a new task, work slipped into lunch breaks, meetings, and the five minutes before leaving the desk. It didn't feel like working. Over time, breaks stopped providing recovery. Work became ambient. And depth eroded — AI introduced a rhythm of managing multiple threads simultaneously, running parallel agents, constantly monitoring outputs. Workers described feeling productive. What they were actually experiencing was continuous attention-switching and a growing stack of open loops that never fully closed.
AI didn't reduce cognitive load. It redistributed it — and then quietly raised the ceiling on what "enough" looks like.
Think about what happens when a farmer uses technology to plant faster and harvest more. Without rotating crops, without letting fields rest, without replenishing what's being taken out, the soil degrades. The land that once produced abundantly — can't.
People work the same way.
Cognitive capacity isn't a resource that refills automatically. It's built through recovery, through depth, through work that has a boundary. The problem isn't that AI reduces friction in your operations — done well, that's exactly what it should do. The problem is when AI removes the natural pauses that signal this task is done, this day is over, this is enough. Without those signals, there is no built-in replenishment.
The neuroscience here is not soft. When you're not actively focused on a task, your brain shifts into what neuroscientists call the default mode network — not going offline, but doing essential maintenance. Replaying what you learned. Making connections. Processing the emotional texture of the day. This is literally how you consolidate work into meaning. But it only happens during genuine unstructured time. The moment you check an AI output, nudge a task forward, or pick up your phone — even for two minutes — that maintenance stops.
AI makes starting tasks almost frictionless. That's the value. But the friction of starting was also doing something useful: it was a natural speed limit. It was the reason you didn't check in on a project at 7pm, or draft something during lunch. When that friction disappears, work expands to fill every available gap — not because anyone asked it to, but because the brain now sees those gaps as opportunity. What used to feel like a break starts to feel like wasted time.
The depletion builds quietly. Because nothing feels hard — the AI is doing the heavy lifting — nobody notices the soil thinning until the yield drops. Workers describe feeling productive right up until the point where they don't. By then, the ability to learn, to regulate mood, to think creatively — the cognitive functions that require downtime to happen at all — have already been compromised.
When "enough" is defined by what the tools make possible rather than what the people can sustainably give, the goalpost keeps moving. Every quarter, the baseline resets higher. And the people carrying it have no structural way to say so.
The problem is not AI. The problem is designing AI into your work without designing recovery back in. Friction-free tools need friction-full boundaries — clear signals that this thread is closed, this day is finished, this is enough. Without those boundaries built deliberately into how work is structured, the brain will fill every frictionless gap it's given. And it will do so quietly, invisibly, right up until the point where it can't.
Good farmland doesn't just produce. It's managed.
The most productive operations don't run at maximum capacity — they run with intention. They know what to harvest, when to rotate, and what conditions the land needs to keep giving.
If your team has adopted AI tools but you haven't redesigned how work is bounded, paced, and recovered — you haven't optimised your operations. You've just planted faster.
Cotempo works with Canadian businesses to build the operational infrastructure that makes AI sustainable — not just fast. That means designing the boundaries, workflows, and systems that tell your people this is enough for today — and mean it structurally, not just culturally.
