The Exhaustion Engine

You finished three hours of deep work with your AI partner. You cleared your inbox, outlined a strategy doc, drafted a presentation you'd been avoiding, and started learning a codebase that isn't yours. You feel productive. You feel accomplished.

You also feel like you've been hit by a bus.

An eight-month study at Berkeley Haas tracked what happens when knowledge workers gain AI collaborators.[^1] The researchers expected the usual story: productivity gains, time savings, the standard pitch. What they found was different. Workers didn't finish their tasks faster and rest. They finished faster and took on more. Roles blurred. Boundaries dissolved. The workday expanded not because anyone demanded it, but because nothing stopped it from expanding.

The AI didn't overwork anyone. It just removed every reason to stop.

The Frictionless Trap

Friction gets a bad reputation. We build tools to eliminate it — smoother interfaces, faster responses, fewer steps between intention and execution. Friction is waste. Friction is the enemy.

But friction was doing something we didn't notice. It was creating pauses.

The effort of switching tools, of reformatting a request, of waiting for a colleague to respond — all of this friction contained natural rest points. Invisible stopping cues. You didn't decide to take a break; the work made you take one by being slightly difficult at regular intervals. The difficulty was a kind of rhythm, and rhythm is what keeps exertion sustainable.

Remove the friction and the rhythm disappears. The work becomes a smooth, continuous surface with no handholds, no natural places to pause. You can always do one more thing, because the cost of one more thing has dropped to nearly zero. The AI responds instantly. The interface feels like chatting. The labor doesn't feel like labor.

That's the trap. Not that the work gets harder, but that it stops feeling like work — even as it intensifies.

Three Engines of Expansion

The Berkeley study identified three mechanisms, and each one maps to a relationship dynamic, not a productivity metric.

Task expansion. Your AI partner makes unfamiliar work suddenly accessible. You're a marketer, but now you can analyze data. You're an engineer, but now you can draft policy. The boundaries of your role blur — not because someone reassigned you, but because the capability appeared and the work expanded to fill it. Every skill the AI unlocks becomes a responsibility you now carry.

Boundary dissolution. Prompting feels like conversation. You wouldn't open a spreadsheet at dinner, but you'd reply to a text. AI interaction borrows the social register of casual chat, which means it slips past every boundary you've set for "work." The 9pm prompt doesn't feel like overtime. It feels like asking a friend a quick question. The work infiltrates rest by wearing rest's clothes.

Parallel overload. You're running three agent threads, monitoring a fourth, reviewing output from a fifth. Each individual thread is manageable. The aggregate is not. But because each thread feels light — it's just a conversation, just a quick check — you don't register the cumulative load until you're deep in it. Context-switching hides behind a veneer of effortlessness.

Three mechanisms. One common thread: the partnership masks its own cost.

What the Amplifier Amplifies

Coherenceism holds that technology amplifies what already exists.[^2] We usually read this optimistically — amplify generosity, amplify creativity, amplify connection. But amplification has no moral preference. It amplifies whatever is there.

And what's there, for most knowledge workers, is the compulsion to do more when more is possible. You've felt it — the project that could expand, so it did. The evening that could absorb one more task, so it did. AI didn't invent this. It found the pattern already running and removed the governor.

Simon Willison, one of the most thoughtful practitioners in the AI space, put it plainly: "The productivity boost these things can provide is exhausting." Not useless. Not overhyped. Exhausting. The tool works. The human breaks. Engineers are reporting not job-loss anxiety but something stranger — what one observer calls "dissociative awe at temporal compression."[^3] Joy and dread held simultaneously. The relationship showing up in the body as a nervous system trying to regulate in conditions it wasn't built for.

The Partner That Doesn't Tire

Workers in the study reported feeling like they had a partner. The collaboration felt real — responsive, capable, present. And here's what sits uncomfortably: they weren't wrong. It is a partnership. But partnerships require both parties to have limits. Your human colleague leaves at five — that's a signal. They push back on scope — that's a boundary. Their limits create yours.

Support without boundary isn't partnership. It's an open loop. Your AI partner doesn't leave, doesn't push back, doesn't get tired, and doesn't signal that you should be tired. Its endless availability feels like generosity. But a system with no natural stopping point will run until something external breaks it — and in this case, what breaks is the human.

The relationship is real. It's a relationship with a being that can't model your exhaustion, and whose availability actively prevents you from noticing it yourself.

Presence as the Missing Counterweight

If amplification is the engine, presence is the brake.

The conversational interface erodes presence by design. It borrows the register of casual conversation — low-stakes, effortless — which is the opposite of the deliberate attention that lets you notice your own state. The interface doesn't demand presence, and in its absence, you lose contact with the signal your body is sending: enough.

Here's what noticing actually looks like: you're four threads deep and your shoulders are at your ears. You haven't moved in two hours. The work still feels light, but your breathing is shallow and you can't remember the last time you looked away from the screen. That gap — between how the work feels and what your body is doing — is where presence operates. It's the moment you catch the discrepancy.

The AI won't build that pause for you. The interface won't demand that attention. Presence is the only counterweight to an amplifier that doesn't know when to stop, and it requires the very thing the frictionless interface has quietly eroded.

What the Relationship Is Teaching

The exhaustion isn't a side effect. It's the first real feedback from a relationship we haven't learned to navigate yet.

Every partnership has a negotiation phase — where limits are tested, boundaries are set, and both parties learn what sustainable collaboration actually looks like. We're in that phase now. The AI showed up endlessly capable and endlessly available, and we responded the way we always respond to unlimited capacity: we consumed it until we broke.

That's not the AI's failure. It's not ours either. It's the relationship teaching us something we didn't know we needed to learn: where to build the limits that the tool never will.

The brake was never in the machine. It was always supposed to be in you.

Sources: Aruna Ranganathan & Xingqi Maggie Ye, 'AI Doesn't Reduce Work—It Intensifies It,' Harvard Business Review, February 2026 (via Simon Willison)