Stop and breathe

8 April 2026

For most of my career, the work had two speeds.

Some days were breathless. In 2015 I sat in a hotel room on election day, editing a PHP file live over WinSCP on the hotel wifi, hoping it would save before the connection dropped. It made me feel sick and completely alive at the same time. That was the extreme, but a lot of days had a bit of it in them. An incident, a launch, a deadline that had already slipped once. You get a huge amount done on days like that, and the adrenaline is real. For a long time I thought those were the days that mattered.

Other days had long flat stretches. You'd push a change and wait twenty minutes for the build. You'd get a coffee you didn't need. You'd stare out of the window while CI ground through the tests, doing what looked a lot like nothing and felt a lot like skiving. On paper that time was doing nothing. Looking back, I think it was doing more than the busy hours ever did.

Two things stand out now.

One is that people need both speeds. The quiet stretches are when you come down from the morning, and the stress settles before it turns into something worse. A team that runs at full tilt all the time burns out. People get tired, and in the end they leave. Nobody ever scheduled the slow bits. They just happened, and they kept us sane.

The other is that good thinking rarely happens when you're under the cosh. When you're being pushed, you keep flogging the plan you've already got, spanner in hand, refusing to admit it's the wrong spanner. The moment where you realise you're solving the wrong problem tends to turn up later, in the shower, or on the walk to get the coffee, once you've stopped trying. With no slack in the day, those moments have nowhere to land.

The clearest example I have is from the campaigns I worked on in 2015 and 2016. The real wins didn't come on the night. They came afterwards, when we sat down and took stock of what we'd shipped, fixed the things that had buckled under load, and wrote up what we'd learned so the next campaign started further ahead. None of that happens unless you slow down enough to do it. Prompt a chat window all day and it simply never gets done.

The shape of the work has changed. On a good day I review ten pull requests, twenty minutes each, typing back into a chat window and moving straight on to the next one. The pressure to ship is higher than it's ever been. Some of it I put on myself and some of it comes from outside, but the message is the same, and it never knocks off early. Ship more.

The old hotel wifi feeling hasn't gone away, though. I caught myself getting the same jolt trying to review and safely ship ten fairly boring backend PRs before the end of the day. No election, no outage, nothing dramatic. Same racing heart, none of the payoff. What happened?

The flat stretches have gone. There's no waiting for the build now, and no natural gap between one thing and the next. So I keep asking what's happened to the two things those gaps were giving us.

I don't have a tidy answer. We're trying a few things.

We're getting more of the human work done before the review. Before an AI PR goes up, the author works through a longer checklist, so some of the mental load is carried up front instead of landing cold on a reviewer who's already nine PRs deep.

We're giving some of the speed back. If AI makes me three times quicker at writing PRs, I'd rather run at about one and a half, and put the rest of that time into thinking harder about the problem, or stepping away and coming back with a clearer head.

And we're talking about it honestly. We've started naming AI burnout as a team, and asking where the pressure actually comes from. A lot of the time it isn't the manager. "It's not me, guv." It's the pressure we put on ourselves. And yes, sometimes it is me, guv. I'm the manager.

The old rhythm isn't coming back on its own. This is an attempt to build the slow bits back into the day on purpose, now that the work doesn't hand them to us for free.

Stop and breathe.