AI Didn't Make Me Faster. It Made Me Right More Often.

I'll tell you something that sounds like a humblebrag and isn't: AI hasn't really made me any faster.

My cycle time — from picking up a problem to shipping the answer — is about what it always was. What changed is how often the answer is right the first time. I used to make an educated guess, launch it, and find out. Now I can do so much more research up front, so much faster, that I mostly skip the guessing. If you'll forgive a baseball analogy: I went from a decent hitter to one who connects on nearly every swing. Same number of at-bats. Wildly different batting average.

That distinction — not faster, but better — turns out to be the whole story of the most interesting finding in this year's tech-worker survey, and most people are reading it exactly backwards.

The productivity paradox

On the surface, the AI numbers in that survey are glowing. 82% of tech workers say AI is making them at least moderately more productive; nearly half say "very much" or "extremely." Sixty percent feel confident or ahead of their peers on AI skills, against just 22.5% who feel anxious or behind. If you stopped there, you'd think everyone had found a superpower.

Then the researchers asked people to describe, in their own words, what "better at my job" actually meant. And "better" turned out to mean more and faster — not higher quality. Worse, a striking number described something breaking in themselves. "I'm amplified, but my brain is rotting." "I feel like I don't think hard enough anymore — I just follow Claude. I don't fully understand what I merge." "I miss feeling smart and having aha moments."

So which is it? Is AI a superpower or a slow lobotomy? The survey seems to say both, and that's not a contradiction. It's a fork in the road, and which branch you're on is up to you.

Not faster, better — if you spend the time right

Here's what I think the people getting genuinely better are doing, and it's the opposite of what the phrase "productivity gain" implies.

When AI hands you back an hour, you can do one of two things with it. You can ship more — take the hour as speed, produce more output, move on. Or you can reinvest it — pour that hour back into thinking, checking, researching the thing more deeply than you ever had time to before. The first path gives you more. The second path gives you better.

I'm relentlessly on the second path, mostly because I'm a skeptic by temperament. I check, and double-check, and let things sit. I look at a problem from three angles before I trust my read on it. That's exactly why AI hasn't made me faster on the calendar — I'm plowing the saved time straight back into judgment. But it's also why the ideas I ship now land the way they do. The tool didn't replace my thinking. It bought me more room to do it.

Why some brains rot and some don't

And this is where the "my brain is rotting" crowd and I diverge — not in whether we use AI, but in how.

The research on this is new and unusually clear. A 2025 study from Microsoft Research and Carnegie Mellon surveyed knowledge workers about hundreds of real AI-assisted tasks and found something worth tattooing on the wall: higher confidence in the AI predicted less critical thinking, while higher confidence in your own judgment predicted more. Same tool, opposite outcomes, depending on which one you trust. The paper describes the shift as moving "from doing tasks to supervising tasks" — and if you supervise without ever doing, the muscle goes. A separate 2025 study in the journal Societies found heavy AI use correlated with lower critical-thinking scores, with "cognitive offloading" — handing your thinking to the tool — as the mechanism in between.

Read those together and the brain-rot stops being mysterious. It's not what AI does to you. It's what happens when you let confidence in the tool stand in for your own judgment — when you follow Claude instead of arguing with it, when you merge what you can't explain. The people whose brains are rotting aren't using AI too much. They're thinking too little while they do.

The "I miss brainstorming with humans" complaint is the same story in a different costume. That's not AI's fault either. I've worked remote for twelve years, and staying connected to people you don't share a room with has always taken deliberate effort. Nobody pays you to keep your relationships alive, and nobody's going to hand you your aha moments. If the machine has quietly become the only thing you think with, that's a discipline you let slide, not a theft the tool committed.

Who owns the quality

Which lands on the uncomfortable question underneath all of this: whose job is it to keep the work good?

Earlier in this series I argued that whether you feel amplified or shaken by AI is largely about your environment — the room and the manager you're under. I'll hold that. But this is the other side of the ledger, and it points the other way. Your feelings about the work are shaped by your environment. The quality of the work is yours. You're a professional. Professionals own their output. If the tool is making your work worse and you can't see it, that's not the tool failing you — that's you handing off the part of the job that was never supposed to leave your hands.

Leaders make this harder, and plenty of them abdicate their end — they reward velocity, they never build the standards, they let "faster" quietly become the only metric. That's a real failure and I don't excuse it. But it doesn't transfer the craft off your desk. It just means you may have to defend your quality without much help.

So here's the move, and it's almost aggressively simple. When AI gives you time back, don't spend all of it going faster. Spend some of it going deeper. Stay skeptical of the confident-sounding output. Never merge what you can't explain. Treat the saved hour as a chance to think harder than you used to be able to — and the same tool that's rotting other people's judgment will quietly sharpen yours. (If you want the team-level version of this, I've written about what your AI coding tools are hiding and the verification gap that opens when developers stop trusting the code they ship.)