Author(s): The Smarter Way

Originally published on Towards AI.

I keep meeting people who tell me, with real pride, that AI has made them three times faster at their job.

I always want to ask the same follow-up and never quite know how to phrase it politely: faster at what, exactly?

Because here’s what I’ve started noticing. The people who sound most impressed with their own productivity gains are usually still doing all the same work they were doing before. Just more of it. More documents. More slide decks. More Slack threads that branch into sub-threads. More polished reports that nobody asked for and nobody will read past the executive summary, which they’ll skim on their phone in an elevator.

If that’s the win, I’m not sure it’s a win.

There’s a paper from METR that came out last year and got far less attention than it deserved. They ran a randomized controlled trial on experienced open-source developers using frontier AI tools — Cursor Pro, Claude Sonnet, the whole stack. Before starting tasks, developers forecast that AI would reduce their completion time by 24%. After completing the study, they estimated it had reduced their completion time by 20%. The actual measured result was that AI increased their completion time by 19%.

Read that again. The developers thought they were 20% faster. The stopwatch said they were 19% slower. The gap between perceived and actual productivity was almost 40 percentage points.

That’s not a small misread. That’s a fundamental disconnect between what AI feels like and what AI is doing.

And it points at something I think we’re all dancing around: most of the productivity story we’re telling ourselves about AI is measuring the wrong thing.

The Shift

The interesting question isn’t how fast does AI let me do this work. The interesting question is should I be doing this work at all.

This is the move that almost nobody is making, and it’s the one with all the leverage.

Think about what AI is genuinely, undeniably good at. It can draft a status update in eight seconds. It can summarize a forty-minute meeting in three sentences that capture more than your notes would have. It can turn a half-formed thought into a passable first draft of an email, a memo, a brief.

The naive read of all this is: great, now I can produce more status updates and memos and briefs. The smarter read is: if all of that is now this cheap to produce, most of it didn’t need to exist in the first place.

A status update that AI can write from your commit history in eight seconds was never really communication. It was friction. You were performing the appearance of accountability for someone who didn’t have time to look at the underlying work. AI didn’t make the status update faster. AI just exposed that the status update was always theater.

The meetings are the same. There’s an Atlassian survey from 2025 that found 68% of developers now save more than 10 hours per week with generative AI, but over half of them lose just as much time to organizational friction. Ten hours saved. Ten hours immediately consumed by the same meeting culture, the same review cycles, the same Slack avalanche. The math nets out to roughly zero.

Look — the work isn’t actually getting done faster. It’s just getting absorbed faster by the work-about-the-work.

The teams I see pulling away aren’t the ones with the best AI tooling. They’re the ones using AI as a kind of x-ray, looking at their existing workflow and asking, repeatedly: does this still need to happen now that this exists?

What This Looks Like In Practice

A few patterns I keep noticing in the people who are actually getting leverage:

They kill recurring meetings the moment AI can replicate the function. If the weekly sync exists to share updates that an AI summary could push to Slack, the weekly sync is over. Not optimized. Not shortened. Ended. The hour goes back on the calendar as deep work. This is harder than it sounds because someone, somewhere, scheduled that meeting for a reason that felt important in 2022.

They stop writing artifacts whose only purpose was to prove they did the work. Status reports. Project recaps. The kind of document that exists because someone two levels up wanted to feel informed without having to ask. If the AI can summarize your week from your tickets and your commits, you don’t write the summary at all. You let them generate it on demand if they ever actually need it. Spoiler: usually they don’t.

They distinguish between code they wrote and code they shouldn’t have written. Generating three thousand lines of plausible-looking code in an afternoon is not a productivity win if two thousand of those lines were a problem you could have avoided with a different design. The senior engineers I respect most use AI to think up the stack — about whether the feature should exist, whether the abstraction is right, whether the simpler approach is actually simpler — before they let it write a single line.

They protect against AI’s gravitational pull toward more. AI defaults to producing. Ask it for an outline, you get a deck. Ask it for a deck, you get a deck plus a one-pager plus a follow-up email plus three appendix slides. The discipline is asking for less than the tool wants to give you, and then asking again whether you needed even that.

They measure subtraction. This is the one almost nobody does. At the end of a week, the question isn’t “what did AI help me produce?” The question is “what did AI help me not do?” Meetings cancelled. Reports retired. Reviews skipped because the change was small enough that the review was performative. If the answer is nothing, the AI is making you faster at the wrong things.

The Closing Thought

There’s a version of the next five years where AI doesn’t make us more productive. It just makes us busier with higher-fidelity busywork. Better-formatted reports nobody reads. Smarter-sounding meetings that still shouldn’t have happened. Code that ships faster but solves problems we shouldn’t have taken on.

The people who escape that version are the ones who stop using AI to do more, and start using it to see what was never necessary.

The most productive thing I’ve done with AI this year is delete a recurring meeting. The second most productive thing was delete the followup document that came out of that meeting. Neither shows up in any productivity dashboard. Both gave me back an afternoon a week that I now spend on work that actually compounds.

That’s the metric I’d track if I were grading my own AI use. Not how much I’m producing. How much I’ve stopped producing on purpose.

The faster path was never the answer. The shorter path was.

Everyone’s measuring AI productivity by what gets made. The real leverage is in what stops getting made at all.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI