The people burning out on AI tools aren’t the ones who refuse to use them. They’re the ones who are best at it.
A BCG survey of nearly 1,500 workers found that the most enthusiastic AI adopters, the power users and evangelists, are the first to hit what researchers are calling “brain fry.” Thirty-four percent of them want to quit.
I read that stat and felt personally targeted.
The Jevons paradox, but for your calendar
There’s an old economics concept called the Jevons paradox. When steam engines got more fuel-efficient in the 1800s, coal consumption didn’t drop. It skyrocketed. The engines were cheaper to run, so people ran more of them. Efficiency didn’t reduce demand. It multiplied it.
AI tools are doing the same thing to knowledge work. ActivTrak’s data from post-AI workplaces is grim: email time doubled. Messaging volume jumped 145%. Deep focus sessions, the actual thinking work, fell 9%. The tools made individual tasks faster, so people just packed more tasks into the day. Nobody got their afternoon back. They got more afternoon.
The UC Berkeley Haas study tracked about 200 employees over eight months and found three patterns: task expansion, blurred work-life boundaries, and constant multitasking. Workers weren’t being forced into this by management. They did it on their own. The AI made “doing more” feel like flow state, so they kept going.
An NBER study found that workers in AI-exposed jobs now work roughly three extra hours per week. Leisure dropped by the same amount. The productivity gain went straight back into more work.
The quiet part, out loud
A Dun & Bradstreet executive told Fortune something that should’ve been scandalous but barely made a ripple: “I got the eight hours to two hours, but now I can get 20 hours of work.”
The efficiency gain didn’t benefit the person doing the work. It benefited the person assigning it. The eight hours compressed to two, and the reward was ten times the workload.
This shouldn’t surprise anyone. When has a productivity gain ever been handed back to workers as time? Four-day work weeks are earned and fought for, not given. The spreadsheet got faster in the ’90s and nobody went home at 3pm. They just built more spreadsheets.
What this looks like at a two-person agency
I run Victoria Garland with my wife. I’m the entire engineering department. When my tools got faster (and they got meaningfully faster this year) the workload didn’t shrink. My threshold for “too complex for our size” just moved.
Work I would’ve scoped out six months ago, I now say yes to. A Shopify app that would’ve taken a week, I can scaffold in an afternoon. That sounds great on paper. In practice, it means I’m doing a week and a half of work in five days and wondering why I’m fried by Friday.
And I’m doing it to myself. Nobody’s standing over me with a quota. Claude Code knocks something out that would’ve taken me a day, and I don’t take the afternoon off. I open the next ticket. It feels productive in the moment. The tools make everything feel like momentum. You’re shipping, you’re closing issues, the dopamine loop is tight. Then Friday hits and you realize you’ve been sprinting all week toward a finish line that moved every time you got close.
The BCG survey found that people using four or more AI tools actually reported lower productivity than people using fewer. More tools, more cognitive overhead. The machine handles execution faster, but the judgment, review, and coordination? The parts that actually tire you out? Those intensify. The bottleneck shifted from “can I build this” to “should I build this, and did I review it carefully enough, and is this the right priority.” That’s more exhausting, not less.
I’m not going to pretend I’ve solved this
Bloomberg ran a cover story in February calling it a “productivity panic”: companies racing to ship faster with AI coding agents while nobody knows how to price the gains. Google says 50% of their code is now AI-written and they’ve got “well over 10% velocity gain.” But velocity gains just raise the shipping cadence. Nobody’s going home early.
I don’t have a fix for this. The tools are too good to stop using and the incentives all point toward doing more. I’m not going to pretend I’ve found balance or cracked some productivity framework that makes it all sustainable. I haven’t.
The least I can do is name it. Productivity gains don’t get handed back to you as time. They never have. The steam engine didn’t give coal miners a shorter day. The spreadsheet didn’t give accountants longer weekends. And AI isn’t going to give developers free afternoons. The bar just moves.
If you’re an AI power user and you’re exhausted, it’s not because you’re doing it wrong. It’s because the system is working exactly as designed.
Shameless plug: At Victoria Garland, we build Shopify apps and integrations for high-growth merchants. If you need serious infrastructure from a team that sweats the details, even when the AI is doing the typing, we’d love to talk.