We Got Faster, Not Freer

You could say that's just bad discipline on my part, and I wouldn't have much of a comeback. But I don't think it's only me, so I went looking.

Free time doesn't stay free

Here's roughly how it went.

The existing work finishes faster. Then you start to get a feel for where the line is. This one I can hand off. That one, not yet. This one I can set running and check back on while I do something else. You end up with a working sense of how to divide things up, and that's the actual change.

The trouble starts once you have it. Work you shelved because there was never time for it, and work you left out of the plan because it was never going to happen, comes back onto the list.

Writing this blog in both English and Korean is the obvious example. Not long ago, writing something like this at all would have been an indulgence. Now I get a draft down, spend a while on the voice, and most of what I meant to say is already there.

Add enough of those back and the total climbs, and the breathing room I thought I'd gained is gone again.

If this were only my own appetite, I could just stop. But when everyone around you speeds up and there's more coming out the other end, you end up matching the pace. Ricardo Bedin, a staff engineer at Stripe, put it this way in a piece he wrote in August:

AI somehow made things easier and raised expectations at the same time. There are fewer and fewer excuses to justify you not being able to land a project on time, or generating the impact you promised.

Ricardo Bedin, Being a Staff Engineer at Stripe in 2026

Easier and more expected, arriving together. He also describes a good part of his job as keeping track of N agents spread across various environments. The job isn't just building things anymore. It now covers managing what the agents are doing, much the way you manage people.

The numbers say the same thing

Linear published data from its own customer base. As of June 2026, pull requests opened per paid workspace were up 111% on a June 2024 baseline.

Did the time spent organizing and reviewing go down? Comparing June 2025 with June 2026 for engineering roles, time spent creating and triaging issues went from 24 minutes a month to 28. Time on comments went from 35 to 40. For founders the swing is wider: 40 to 57, and 39 to 64.

Output roughly doubled. The time spent sorting it out didn't shrink; it went up a little.

Before leaning on any of that, though, here are the caveats Linear put on the same page.

We have no way of knowing whether this increased output led to positive business outcomes. We count PRs opened rather than merged, and an opened PR says nothing about the value of the change. What we can't see is AI usage that happens outside Linear, so this is a picture of adoption inside our own customer base, not the market at large.

Linear, AI Usage Report

I don't think a PR count settles anything on its own either. What I do find telling is how many memes there are about this number. Think of the old one about how a ten-line PR gets ten comments and a five-hundred-line PR gets an LGTM. A rise in PRs opened is saying several things at once. It might mean more got built. It might also mean the chunks nobody really looked at got bigger.

Busy means something different now

What actually feels different isn't the volume.

When someone said they were busy, it used to mean they had a lot of work. A stack of tasks that mostly related to each other, or one task big enough that you had to sit with it for a long time.

Being busy now isn't a pile of tasks. It's several objectives sitting above the task level, all wanting attention on the same day. I look at card game balance in the morning, fix something on the site after lunch, and go through text for the next build in the evening. Each one needs a different set of background loaded back into my head.

The tiredness that comes with it is different too. It isn't from sitting with one thing too long, it's from switching. I finish days with a long list behind me and nothing I went deep on. I'm not sure yet whether that's a good trade.

Who looks at the extra?

If output goes up, review has to go up with it. Review doesn't scale on its own.

Think about a factory that brings in automation. Fewer people end up standing on the line, and more end up on inspection and record-keeping. If you don't create those positions on purpose, the work doesn't disappear. It quietly lands on whoever was already in the habit of tidying up.

On a team of four this shows up fast. If you haven't decided who looks at the extra, it becomes one person's problem. We haven't settled it properly either. For now, each of us looks closely at our own, and we try to give each other real feedback on the rest.

Less replacement, more supervision

For a while the story was that AI would replace people. That isn't entirely wrong. What I've seen is a little different.

The work changed, the things we'd put off came back, and instead of people disappearing there was simply more to do. Work that had been postponed and work nobody counted as work both grew, and someone still has to point those systems in the right direction and check what comes back.

So output goes up. Which makes me think that when this stretch is over, we come out somewhere further along, with more to show for it. On that much I'm optimistic.

One thing does change. What matters is how quickly someone responds to change, and how fast they get comfortable with a new tool, a new setup, a new problem. Agility is going to be the thing you hire for. It's been a buzzword for years, but what it actually points at keeps shifting, and it's getting sharper.

Nobody has the answer yet

Bedin's piece ends like this:

It's one of those unique moments in time where there's a breakthrough happening, and no one really understands the best way to go about it.

Ricardo Bedin

I don't have the answer either. One thing I would say: if you bring AI in expecting that time is on your side and things are about to get comfortable, you probably won't get what you were after.

Start by asking how far it's going to spread, and settle at the same time who manages and reviews whatever it spreads into. Look at the ripple an AI workflow is going to send through the place before it lands, and bring it in with the changes to roles and org chart already planned for.