The AI Speed Tax: Why Faster Work Just Means Faster Expectations
You used AI and finished the report in two hours instead of eight. Your boss now expects reports in two hours.
The mechanism underneath that matters, because it's not cynical management but a structural feature of how efficiency gains get absorbed, and it's been happening since long before AI existed.
Parkinson's Law Has an Inverse
What happened with email is the clearest version of this. In the early days, a 24-hour response window was normal, but by 2013, more than three-quarters of workers expected a reply within four hours, and by 2018, more than 40% expected one within an hour, with no one declaring a new policy. Each improvement in reachability quietly became the new floor. That's Parkinson's inverse: when time compresses, the efficiency rarely flows back to the person who created it.
It all just happened. And that quiet, unannounced shift is exactly what we're watching with AI now, only faster.
A UC Berkeley Haas study spent eight months inside a 200-person tech company watching what happened after AI arrived in the workflow, and the finding wasn't that people worked less, it was that they worked more, faster, on more things, in a vicious cycle where increased capability leads to increased output, which leads to higher expectations, until "what was once extra effort becomes standard performance." By month six, burnout had spiked most sharply among the heaviest AI adopters.
The Expectation Ratchet
What makes this different from the email wave is that the gains are so quantifiable that the ratchet tightens faster. When you finish a six-hour task in one hour, there's data now, your output history, your commit cadence, your response times are all logged somewhere, and the new baseline isn't a vibe, it's a benchmark.
I feel like this is the part most productivity writing about AI refuses to say plainly. The Fed estimated AI saves the average worker about 5.4% of work hours, yet a survey of around 1,250 workers found 31% reporting increased workloads, and of those, 43% said the workload had at least doubled. Productivity grew more than 80% between 1979 and 2019 while typical worker compensation grew about 13%, and of organizations reporting AI-driven productivity gains in 2026, an EY survey found only 17% had reduced headcount.
Torsten Slok, chief economist at Apollo Global Management, applied the Jevons paradox directly: "When steam engines made coal more efficient, Britain didn't burn less coal, it burned more. The same pattern is happening for cheaper legal services, consulting services and financial services." Cheaper output enables more output, and demand expands to absorb the supply.
Who Gets to Decide What the Time Was For
I find Cal Newport's framing in Slow Productivity useful: the dominant model of knowledge work treats visible busyness as proof of value, what he calls pseudo-productivity, and it doesn't measure actual output, it measures motion. AI gives you more motion, faster, which means your performance floor just moved up.
The workers who come out of this better are probably the ones who name it explicitly before expectations harden, not as a confrontation but as a negotiation: here's what I can now do, and here's what I'd like to do with the margin. The Slack Workforce Lab found that when workers were asked how they'd use AI-freed time, "more admin" ranked first and skill-building ranked last, which isn't irrational, it's rational in an environment where visible activity is what gets rewarded. That conversation about margin becomes nearly impossible once the new pace is the standard.
I don't know if most people get to have it, and I'm not sure most organizations are structured to allow it. But absorbing the speed gain silently, watching the baseline rise, wondering why you're more tired while producing more than ever, probably isn't sustainable either. Perhaps the better question isn't whether AI makes us faster, but who that speed ultimately belongs to. We seem to be watching efficiency gains flow away from the people who generated them, and whether that ends somewhere good, I genuinely don't know.