The Efficiency Trap
AI didn't give you more time. It gave you more things to do with it.
Last quarter, a content team I know cut their campaign brief production time from a day and a half to about forty minutes.
So they started producing more briefs.
Not because anyone told them to. Not because their manager raised the quota. Because forty minutes felt like nothing, and there was always another brief that needed writing, and the tools were right there, and it felt like momentum.
By the end of the quarter, the team was exhausted. They were producing more than they ever had. They couldn’t explain why it felt worse.
There’s a name for what happened to them. It’s been around since 1865.
William Stanley Jevons was an economist trying to understand a paradox. Steam engine technology had improved dramatically. Engines were more fuel-efficient than ever. So why was Britain burning more coal, not less?
Because efficiency made coal cheaper and more practical. So more people used it, for more things, in more places. Every unit of efficiency got immediately converted into expanded demand. The savings never accumulated anywhere. They disappeared into appetite.
He called it the rebound effect. We’ve been calling it the Jevons Paradox ever since.
And it’s playing out right now in your task list, your calendar, and your Thursday afternoon.
Here’s what it looks like in practice.
AI makes content creation faster, so you produce more content. AI makes research quicker, so you research more things. AI makes communication easier, so you communicate more frequently.
Each individual task takes less time. The total volume of tasks grows without limit.
You now have ten things on your plate that didn’t exist eighteen months ago. Every single one of them started with “well, AI saved me time on X, so now I have time to do Y.” And then Y spawned Z. And Z spawned three more things that didn’t have names yet.
The time AI saved? Spent. Then borrowed against next week too.
This month, Harvard Business Review published a study called “AI Doesn’t Reduce Work — It Intensifies It.” Researchers Aruna Ranganathan and Xingqi Maggie Ye from UC Berkeley’s Haas School of Business spent eight months embedded at a 200-person tech company watching what happened when workers genuinely embraced the tools.
Nobody was pushed to hit new targets. Nobody was threatened with a performance review. Workers simply started taking on more because more felt possible. They managed several active threads at once, manually writing code while AI generated an alternative, running agents in parallel, reviving long-deferred tasks because AI could handle them in the background. Work slid into evenings and weekends. Not because anyone demanded it. Because it still felt doable.
One employee put it plainly: the expectation was that being more productive with AI would mean working less. In practice, they found themselves working the same amount or even more.
ManpowerGroup’s 2026 Global Talent Barometer surveyed nearly 14,000 workers across 19 countries and found something that should stop everyone in their tracks. Regular AI usage jumped 13 percentage points. Confidence in the technology fell 18%. Nearly two-thirds of workers, 63%, report burnout.
Then there’s the METR randomized controlled trial, which had experienced developers using actual frontier tools, Cursor Pro with Claude, completing real-world tasks on codebases they knew well. With AI, they took 19% longer. When asked, they estimated they’d been 20% faster.
They felt faster. They were slower.
Here’s the part of this story nobody talks about.
Every one of those studies is measuring organizations that handed their teams powerful tools and then stood back. No new frameworks for how to use them. No intentional constraints on what they should be used for. No redefinition of what “productive” actually means in an AI-assisted environment. Just capability, dropped into an existing system that was already running at capacity, and then watched as the system filled every inch of the new space.
That’s not an AI problem. That’s a leadership problem.
The Jevons Paradox tells us that efficiency gains always get consumed by something. In a vacuum, they get consumed by volume: more tasks, more threads, more output that may or may not connect to anything that matters. The antidote isn’t less AI. It’s intentional direction about what the expanded capacity is actually for.
The organizations getting this right are doing a few things differently.
They’re defining outcomes before deploying tools. Not “we’re rolling out AI to the content team” but “we’re using AI to cut brief production time so the team can spend more hours on strategy and less on execution.” The efficiency gain has a destination before it’s created. Without that, the hours don’t go to strategy. They go to more briefs.
They’re keeping humans in the loop at the decision points that matter. The HBR researchers described workers who felt like quality-control inspectors for an unreliable but prolific junior colleague. That’s what happens when AI handles execution and nobody owns judgment. The teams that avoid burnout aren’t using AI less. They’re being deliberate about where human thinking is non-negotiable and protecting that time explicitly.
They’re measuring the right things. If your only metric is volume, content produced, campaigns launched, tasks completed, AI will optimize for volume and your team will burn out hitting it. The signal you actually want is whether the output is moving the needle on something that matters. That requires a human deciding what the needle is and checking whether it’s moving, which requires protected time that doesn’t get eaten by the paradox.
They’re treating AI adoption as an organizational design question, not a tooling question. Which means someone with authority over how work gets structured has to own it. Not IT. Not the enthusiastic individual contributor who discovered ChatGPT. Someone who can say: here’s what we’re trying to accomplish, here’s where AI helps us get there, and here’s what we’re explicitly not going to let it do.
The burnout data isn’t telling us that AI makes work worse. It’s telling us that AI amplifies whatever structure, or absence of structure, was already there. Deploy AI into a reactive, volume-obsessed organization and you get a more reactive, higher-volume organization. Deploy it with clear intent and human oversight at the right points, and you get something different: a team that’s genuinely more capable, not just busier.
The Jevons Paradox doesn’t care whether you’re burning coal or burning through your people. The rebound happens either way.
The only variable is whether anyone decided in advance what the efficiency was for.