Borrowed capacity

Why your output increased but your tank is empty

New research from UC Berkeley shows that workers who adopted AI most enthusiastically are burning out faster than those who resisted it. Engagement among heavy AI users has collapsed from 88% to 64% in a single year. Over 75% of workers now report burnout. The people who were supposed to be thriving are the first to break.

This seems like a contradiction. AI was supposed to reduce workload. It did increase output. So why are the biggest believers showing the earliest cracks?

Because AI is borrowed capacity. And borrowed capacity always comes due.

Two ways to do more

There are two ways to get more done.

The first is to build genuine capacity. This takes time. You develop skills, strengthen your attention, improve your recovery systems. Your ability to sustain effort increases because the underlying structure supports it. The output is real because the capability is real.

The second is to borrow capacity from somewhere else. This works immediately. Caffeine borrows alertness from your future self. Credit borrows money from your future self. Stimulants borrow energy from your nervous system’s reserves. AI borrows cognitive output from an external system.

Borrowed capacity is useful. Sometimes necessary. The problem comes when you mistake it for built capacity. When you operate at a pace your underlying structure cannot sustain. When you scale your output without scaling your ability to maintain it.

This is what’s happening to AI’s biggest fans. They’re operating on borrowed cognitive capacity. Their output increased. Their actual capability to sustain that output did not.

The expectations trap

Here’s the mechanism that makes this dangerous:

Once you demonstrate higher output, that becomes the new baseline. Your manager saw you produce at the AI-augmented level. Your clients experienced the faster turnaround. Your own standards calibrated to what you could do when borrowing capacity. The old pace now looks like underperformance.

You cannot return to the pre-AI level of output without consequences. The loan is locked in.

Meanwhile, something worse is happening: your actual capacity is atrophying. The skills behind the output are decaying because you’re not exercising them. Your attention span is fragmenting from context-switching between directing and reviewing. Your judgment is dulling because you’re not running the full cycle of creation and evaluation.

You’re now expected to maintain a pace that was only possible through borrowed means. And the capability that would let you sustain it is getting weaker, not stronger.

The enthusiastic adopters burn out first because they scaled the fastest. They’re furthest out on borrowed capacity. The gap between what’s expected of them and what they can actually sustain is the widest.

The hustle culture pattern

This isn’t new. AI just made the pattern more visible.

Hustle culture has always run on borrowed capacity. The entrepreneur on four hours of sleep and stimulants. The startup founder who’ll “rest when it’s over.” The high achiever running on adrenaline and optimization tools. They output more than sustainable capacity would allow.

For a while.

Then the crash. The burnout and the mysterious health problems that seem to come from nowhere. The inability to function at the level they used to handle easily.

This happens because the body keeps an honest ledger. You can run a deficit for a time. The nervous system will let you withdraw from reserves. But the withdrawal doesn’t disappear. It accumulates until the account is empty.

The AI burnout paradox follows the same pattern, just in cognitive territory. You can borrow cognitive output. The external system will let you produce beyond your actual processing capacity. But the shortfall accumulates. Your attention span decays. Your judgment atrophies. Your capacity to engage deeply erodes.

Eventually the loan comes due.

What rest doesn’t fix

Here’s what catches people off guard: the normal recovery strategies don’t work.

You take a weekend off. You feel the same. You take a vacation. You come back and the exhaustion returns within days. Rest that used to restore you now barely touches the fatigue.

This happens because the depletion runs deeper than tiredness. The capacity itself has degraded.

When you’re merely tired, rest restores you because your underlying capability is intact. Sleep recharges what was temporarily depleted. A weekend rebuilds what a week consumed.

When your actual capacity has degraded, rest doesn’t fix it. The fundamental structure needs rebuilding, not just recharging. This takes much longer than anyone wants to accept. Some traditions say genuine vitality takes a month or more of proper living to rebuild. You cannot weekend-retreat your way out of structural depletion.

The person who burned out through borrowed capacity isn’t just tired. They’ve been operating at a pace their system couldn’t sustain, and the system itself has been damaged. Recovery requires rebuilding the foundation, not just refilling the tank.

What actually builds capacity

The uncomfortable truth is that building genuine capacity is slow.

It looks like consistent practice rather than heroic output. Sustainable pace maintained over time. Investing in skills that take years to develop, rather than productivity hacks that work immediately.

Capacity builds when you do slightly more than you could do before, recover fully, then do slightly more again. This is the principle behind all genuine development. Small increments, full recovery, gradual expansion of what you can sustain.

Capacity does not build when you operate at borrowed levels, crash, operate at borrowed levels again. The heroic output followed by collapse does not average out to growth. It averages out to depletion. For a deeper look at why intensity itself is the trap, see that companion piece.

The 6/10 week with consistent sleep beats the week of 9s and 3s that averages to 6. The person who produces steadily within their capacity will eventually outpace the person who alternates between borrowed performance and crashes.

Not immediately. The borrower looks more productive at first. But the pattern inverts over time.

The real question

The fix isn’t to reject AI. Tools that extend your capability are valuable. The fix is to understand what AI actually is and isn’t doing for you.

AI is borrowed cognitive capacity. It’s useful for extending what you can do. It’s dangerous when you mistake it for an increase in what you can sustain.

The real question isn’t “how do I produce more?” It’s “what is building my actual capacity, and what is just borrowing against it?”

Watch yourself for these signs:

You’re doing the same hours but feeling more drained. This means you’re paying the cognitive cost of fragmented attention without getting the integration that creates flow.

You can’t tell if your output is good anymore. Your judgment is atrophying because you’re not exercising the full cycle of creation and evaluation.

You feel vaguely fraudulent about work bearing your name. Something in you knows you’re operating on borrowed capacity.

Rest doesn’t restore you like it used to. The depletion is at the structural level, not the tiredness level.

You don’t know how you’d function without the tools. Your actual capability has decayed to the point where the borrowed capacity is now necessary to maintain baseline function.

Building while borrowing

Borrowed capacity can be used wisely. The key is to keep building genuine capability alongside it.

Stay in the judgment seat. Use AI to accelerate parts of a process you still own and understand. Don’t outsource the thinking, just the production of what you’ve already thought through. Your skills stay active.

Maintain sustainable pace. If AI-augmented output requires unsustainable attention, you’ve borrowed too much. Scale back to what you can actually maintain. Let expectations recalibrate to reality.

Prioritize recovery. If you’re borrowing capacity, recovery becomes more important, not less. The deficit needs to be paid down. Sleep, rest, genuine disengagement from cognitive demand.

Build the skills behind the output. Whatever AI produces, maintain the capability to produce it yourself. Practice the craft even when you could shortcut it. Keep the judgment sharp by exercising the full cycle.

The enthusiastic adopters who are burning out made an understandable mistake. They thought increased output meant increased capacity. It doesn’t. Output can be borrowed. Capacity must be built.

The loan always comes due. The only question is whether you’re building while you borrow, or just running up a tab you’ll eventually have to pay.