The work is about to change under us. The tools turn over every quarter, but that churn isn't it. The deeper change is in us: what our attention is for, and who is let near the work at all. Few are saying it plainly, so I will. The people best placed to see it have always sat at the human end.

Keep a human in the loop, I'm told. In meetings, in frameworks, on assurance calls, round it comes again. So I will. But the phrase puts us in the wrong place. It pictures a person inside the machine's loop, posted there to watch it, and that was never where we stood.

There are two loops.

One of them is the machine's. It makes, and keeps making, whether we're watching or not. The other is older than any machine: a person on the other end, who has to live with what's been made. Both turn on their own, and neither needs us to exist.

What needs us is the rhythm. Our job is to keep both loops turning in time, without fusing them into one.

Two loops, kept turning in time. Two separate hand-drawn loops side by side, one labelled the machine and one labelled the person, each with an arrow showing it turns on its own. A dashed line joins their facing edges across a clear gap, marked in time: the rhythm that keeps the loops in step without ever letting them touch. the machine the person in time
Two loops, each turning on its own. The work is keeping them in time. Close, never merged.

We can do that for a reason the machine can't borrow. We know, from the inside, what it's like to be on the receiving end. The machine makes the thing. It doesn't feel it land. We do.

Call it taste if you like. It's closer to a forecast: you run the made thing forward in someone else's head, and you feel the wince or the nod before it comes. It's a skill few people ever see done.

The machine has none of that, and it's working alongside us anyway. It's quick and tireless, and it's never once met the person it's making for. So it makes what it can make.

Nobody asked their government for something built before anyone understood the problem. People asked for a claim sorted, an appointment booked, a straight answer they could trust, and the machine builds it anyway, because building is the thing it knows how to do, and understanding is the thing it can't.

You feel the gap before you can name it. Someone comes to the service in the worst week of their year, and what comes back is warm and correct, and meant for nobody in particular. The reader can't say what is wrong. They only come away trusting the service less than they did a page ago.

We catch that, and fix it before it reaches anyone. It is hard work, and no one sees it. Done well it leaves no mark: the worried person never meets the wall, and never knows it was there.

This is what user-centred design has always done. Whatever our craft, research, content, service, interaction, we've sat at the human end of the loop. The machine's changed what fills that loop. The job still sits with us. It's taken some of the routine off our hands, and handed back the harder part.

And I know how easily that part gets skipped, because I skip it myself. I make things with these tools most days, a bit of guidance, a manual. It's easy to feel like I've kept a human in the loop. The human in the loop was only ever me, sat there nodding along at the screen. Then the work meets the people it was actually for, and I'm humbled, every time, by how much the two of us had just assumed.

The human in the loop was only ever me, sat there nodding along at the screen.

So this is a book about what that does to us. There are good books already on how to prompt the thing, and which one to reach for this quarter. I've learned from them, and I'm glad they exist. This one is about the slower, psychological change underneath: what our judgement and our care become, once a machine can draft almost anything. If it's prompts you came for, this isn't the book, and I won't be offended.

Two things sit under all of it. The first is that the parts of our work we've long called soft are turning out to be the hard ones. The feel for people, the reading of them. Those are the parts hardest to hand over.

The second is more hopeful: it lowers the technical toll that building software has always charged at the door. That toll fell hardest on the ones drawn more to people than to things, often those with the surest feel for the work. They can make for people now without first having to master the machinery.

The setting here is UK central government, where the stakes are plain and the standard is public. But the questions travel, to anywhere the work is regulated and the people on the receiving end have no real choice but to use what you've made.

The machine is starting to act on its own, where it used to only draft. You checked its work. Now you set it going and walk away. As it reaches further, the temptation grows to fold the human into the machine's loop and call it progress.

Do that and the human vantage goes with it. No one is left standing where the person stands, feeling how the thing arrives, and that absence costs more the more the machine does. So we keep the loops apart, and keep them turning in time. Close, never merged.

That is Loopwork. The oldest part of the job. The next ten years will lean on it hardest.