Why AI Makes You Work More, Not Less
AI was supposed to save us time. Instead, most heavy AI users work more than ever. What the productivity paradox actually is, why it happens, and how to handle it.
At an Anthropic-sponsored Claude Code event in Chiang Mai, the speaker asked a room full of AI engineers a simple question: "Raise your hand if AI should have reduced your workload, but you find yourself working more."
Every hand went up. Including mine.
That moment stuck with me, because it names something most of us feel but rarely say out loud. AI tools genuinely work. They make me several times faster. And I work more than I did before they existed. If you have adopted AI in your own work and feel busier than ever, you are not doing it wrong. You are experiencing the default outcome.
The paradox has a name
Economists call it the Jevons paradox: when something gets more efficient, we do not consume less of it. We consume more. When coal engines got efficient, coal use went up. When ATMs made bank branches cheaper to run, banks opened more branches and employed more tellers, not fewer.
Software is following the same curve. Building it used to be scarce and expensive, so most software that could exist did not. Make it ten times cheaper to build and you do not get the same software with less effort. You get vastly more software. The backlog was never the limit. Imagination was.
The same thing happens inside a single career. AI did not shrink my to-do list. It raised the ceiling of what one person can plausibly attempt, and my ambitions rose to fill it immediately.
Why it pulls so hard
Working this way daily, I notice three mechanisms that make the "more work" pull surprisingly strong.
Everything feels like a skill issue. When an AI tool fails, it rarely feels like the tool hit a hard limit. It feels like I set it up wrong: instructions not clear enough, context missing, workflow not clever enough. That framing is often true, which is exactly what makes it a trap. If failure is always fixable, there is always one more thing to optimize, and no natural place to stop.
The old excuse is gone. "I am one person, I cannot do everything" used to be a real constraint, and a quiet relief. It absorbed the gap between what you wanted to do and what you could do. With AI, the story becomes "one person can do almost anything with the right setup." The constraint that used to carry the guilt has been removed. The gap is still there. Now it is on you.
Capacity becomes a scoreboard. I have caught myself treating usage limits on AI tools as targets instead of guardrails, feeling like unused capacity is wasted opportunity. That is the tail wagging the dog. When the machine can work around the clock, the bottleneck flips from the computer to you, and "keeping the machine busy" quietly becomes your job.
What to actually do about it
I did not solve this by using AI less. The gains are real and I would not give them back. What helped was changing what the work is.
Shift from doing to building systems that do. After that event, I spent the rest of the weekend turning every task where I was still manually involved into a command my tools can run on their own: pulling analytics, transcribing recordings, generating images, managing my schedule. The work does not get less. But it moves up a level, from executing tasks to designing and reviewing the machine that executes them.
Choose the finish line deliberately. Efficiency gains get eaten by a longer wishlist unless you decide, on purpose, what done looks like. "What do I actually want to have finished this month" is now a more important question than "how much can I get done", because the honest answer to the second one is: there is no ceiling anymore.
Treat limits as protection, not failure. A usage quota, a working-hours boundary, a finished feature list. If the tools can always do more, the limits have to come from you, and hitting them is not underperformance.
If you run a business
This matters beyond personal workload. If you are adopting AI in your company expecting the same output from less effort, plan for what actually happens: the same effort producing far more output, and a new kind of work appearing, namely deciding what is worth producing at all.
The teams that struggle with AI are not the ones who fail to adopt it. They are the ones who adopt it and let the workload expand without anyone choosing where it should stop.
AI will not give you your time back on its own. It hands you leverage and leaves the throttle in your hand. Deciding where to set it is the one piece of work it cannot do for you.