AI Didn't Give You Time Back. It Made You Volunteer More of It.
Nobody made me build the second version of the report. I just could, so I did, and then I told myself that was initiative.
That's the sentence I couldn't stop thinking about after reading Aruna Ranganathan and Xingqi Maggie Ye's study out of UC Berkeley's Haas School of Business. Ranganathan and Ye spent eight months embedded with roughly 200 employees at a US tech company as AI tools rolled into their daily work, and the finding — published in Harvard Business Review in February — wasn't the one the productivity narrative promised. AI didn't hand anyone their evenings back. It quietly moved the fence around what "my job" meant, and everyone on the inside of that fence agreed to the new boundary without being told to.
The Time-Back Promise Was Always a Category Error
Every AI productivity pitch since 2023 has run on the same premise: automate the boring parts, and the hours those parts used to eat are yours again. It's a clean mental model. It's also wrong in a specific, measurable way, and the Berkeley study is the first primary-source look at exactly how it breaks.
The mechanism isn't that AI fails to save time. It does save time — that part of the pitch is true. The failure is in what happens the moment after the time is saved. Ranganathan and Ye found that employees didn't bank the freed hours. They rerouted them, almost immediately, into scope nobody had assigned: a second draft nobody requested, a deeper analysis than the brief called for, a side project that "AI made possible now." The tool didn't lie about saving time. It just never said what would happen to the time once it was free, and it turns out what happens is that the job's boundary quietly moves to fill it.
The Boundary You Never Agreed To Move
Here's the part that should unsettle you more than the burnout headlines do: nobody in the study reported feeling forced. That's the whole mechanism. Coercion is easy to resist because you can see it coming. This is the opposite — it feels like agency the entire time it's happening.
Before AI tools, "my job" had a shape defined mostly by what was physically possible in a workday. That shape did real psychological work you weren't crediting it for. It gave you a legitimate, external reason to stop. AI removed the external constraint, and instead of an internal constraint stepping in to replace it, most people in the study filled the space with more of themselves — more output, more scope, more of what an ambitious, competent person would plausibly do with the extra capacity. The tool didn't ask you to work more. It just stopped telling you when to stop, and your own ambition filled the silence.
I've watched this happen to my own week. A task that used to take four hours now takes ninety minutes with the right tool chain. The other two and a half hours didn't become a walk outside. They became a second pass I'd never have had time for before, framed to myself as "since I have the time now." Nobody billed me for that time. I gave it up voluntarily, and I felt good about giving it up, which is exactly the design flaw.
Why This Reads as Virtue Instead of a Problem
The Berkeley researchers make a point that deserves more attention than it's gotten: absorbing freed-up time as new scope doesn't look like overwork from the inside. It looks like initiative. It looks like the thing every performance review says it wants. That's precisely why it's so hard to notice, let alone resist — the behavior gets rewarded before anyone, including the person doing it, has a chance to ask whether it should exist.
Compare that to the attention research showing how digital tools fragment focus — that story has an obvious villain, a device pulling your eyes away from what you meant to do. This one doesn't. Nobody's fighting you for the reclaimed hours. You're the one reaching for them, and the reach feels like ambition rather than a symptom, right up until the week you can't explain where all your time went even though every individual task got faster.
The Question That Actually Catches It
The study doesn't offer a tidy fix, and I'm not going to manufacture one it didn't earn. But it does hand you a genuinely useful diagnostic, buried in the mechanism itself: when a task gets faster, ask what you did with the difference before you decide the tool worked for you rather than on you.
If the honest answer is "I stopped there and did something else with my life," the tool gave you time back, and the promise held. If the honest answer is "I used it to do more of the same job," nothing was returned. The job just got a new, unspoken edge, and you're the one who drew it — which means you're also the only one who can move it back.
What the Old Boundary Was Actually Protecting
The uncomfortable implication of the Berkeley findings is that the old, slower version of your job wasn't purely inefficiency waiting to be optimized away. Some of that slowness was a boundary doing quiet, unglamorous work — the work of telling you when enough was enough, without you having to decide it yourself in real time, task by task, with no external signal to lean on.
AI didn't remove your workload. It removed the thing that used to tell you where your workload ended. Nobody's going to hand that boundary back to you. The study makes clear that it doesn't get reinstated by default — it only comes back if you put it there yourself, on purpose, before the next task finishes faster than you expected it to.
Cover photo by Faye Tsui via Pexels.