Productive and Exhausted: The AI Workload Paradox

Two findings that keep appearing together and look contradictory: most professionals report AI making them meaningfully more productive, and a substantial share of the same people report feeling more drained than before.

Both are true, and the explanation is in what AI removes from a working day versus what it adds.

What it removes: the recovery

A working day isn't uniformly demanding. It has peaks — the difficult conversation, the decision under uncertainty — and troughs: formatting the deck, writing the routine update, cleaning the spreadsheet. The troughs feel unproductive, and they are, in output terms.

They're also where recovery happens. Low-demand work between high-demand work is how people sustain a day. Automate all of it and what's left is peak after peak, with nothing in between.

This is why "I got more done and I'm more tired" isn't a contradiction. Both statements describe the same change: the cognitive troughs are gone.

What it adds: continuous evaluation

The work AI creates is evaluative, and evaluation is more tiring than production at equal duration.

Producing something has natural rhythm — you build, you get absorbed, time passes. Evaluating is a sequence of discrete judgements with no flow state available: is this right, is this the tone, is this claim supported, is this the version to send. Each is small. There are now a great many of them, and they arrive faster than they used to.

Anyone who has spent a day reviewing other people's work rather than doing their own recognises this. AI turns a larger share of every professional's day into that.

What it also adds: expectation drift

The third piece is social. When a task that took an afternoon takes twenty minutes, the expectation does not stay at one per afternoon. It becomes several, and the standard rises to what's now achievable rather than staying where it was.

Nobody decides this. It happens because everyone's baseline moves at once.

Four things that help

Keep some low-demand work. Not out of nostalgia — as recovery. Choose deliberately which routine tasks you keep doing by hand, and treat that as legitimate rather than as inefficiency you haven't got round to fixing.

Batch the evaluating. Reviewing eight outputs in one sitting is significantly less draining than reviewing eight outputs scattered through a day, because you pay the context-switch cost once. This is the single most effective change available.

Set the standard before you generate. Most evaluation fatigue is deciding what "good" means while looking at each attempt. Written criteria turn a judgement into a comparison, and comparisons are much cheaper.

Budget the loop. Iterating with AI has no natural stopping point — there's always a slightly better version eight seconds away. Deciding in advance how many rounds a piece of work gets is a real intervention, because the alternative is stopping when you're tired rather than when it's done.

Léa — the day with no troughs

Léa is a marketing manager at a 70-person SaaS company. Six months into using AI heavily, her output was up substantially and she was going home flattened in a way she hadn't been before.

Her account of it: every task she used to do on autopilot — formatting, scheduling copy, first drafts of routine emails — was gone, and what remained was decisions. Her day had become an unbroken sequence of judgement calls with no shallow work in between.

She now keeps two things by hand: the weekly newsletter's opening paragraph, and her own meeting notes. Neither is efficient. Both are, in her description, the only parts of the day where she isn't assessing something.

Karan — the loop with no end

Karan is a consultant at a mid-sized advisory firm. His problem wasn't volume; it was that no piece of work ever felt finished. There was always a better version one iteration away, and he kept taking it.

He now sets a round budget before starting: two iterations for internal work, four for client deliverables. When the budget's spent, the work ships.

His quality hasn't dropped in any way clients have noticed. What's changed is that work now ends because it's done rather than because he ran out of energy.

The one thing

More productive and more tired isn't a paradox. AI removes the shallow work that used to provide recovery and replaces it with continuous evaluation, which is more draining per hour than production.

The counters aren't about using less AI. Batch the judging, set standards in advance, budget the loop, and deliberately keep some low-demand work — because a day of nothing but peaks isn't sustainable regardless of how much it produces.