The Last Ten Percent

In short: AI has made the competent middle of craft nearly free, but it hasn't made judgment or care cheap. Craft is moving from the hand to the eye: taste, the last ten percent of a piece of work and caring about details nobody asked for. Because the eye was trained by the hand, organizations now have to build the apprenticeship on purpose.

Early in my career I spent most of an afternoon on four pixels.

A navigation bar sat slightly wrong against the headline below it. Nobody had complained. Nobody would have. I nudged the padding, rebuilt the line height, checked it at three screen sizes, and went home sure that something in the world was now a little more correct than it had been that morning. I couldn't have explained the value of that afternoon to a client. I could feel it.

That feeling is what people mean when they talk about craft. It isn't a style or a skill set. It's a relationship with the work in which you care about parts of it that nobody asked you to care about, and you keep caring after the point where anyone would notice.

I've been thinking about that afternoon a lot lately, because the thing I spent it on can now be generated in a few seconds. Not just the padding. The navigation, the headline, the whole page, the code underneath it, and a reasonable explanation of why it works. It arrives competent. It arrives fast. And it arrives without anyone having cared about it at all.

So the question people keep asking me, usually designers and developers I've worked with for years, is whether craft is dead. I understand why they're asking. I don't think the answer is the one they're afraid of, but I don't think it's the comfortable one either.

What Is Craftsmanship?

We tend to describe craft by its outputs. The hand-set type, the clean component, the interface that just feels right. That's where it shows, but it isn't where it lives.

Craft lives in the decisions nobody sees. A finished screen is the residue of a few hundred small choices, most of them invisible: this spacing instead of that, this word instead of the one that was almost right, this edge case handled instead of ignored. Any one of them is trivial. Together they're the difference between something that works and something that feels considered.

The sociologist Richard Sennett spent a book on this and landed somewhere I keep coming back to: craftsmanship is the desire to do a job well for its own sake. Not for the client, not for the review. For the work. The craftsperson cares about the back of the cabinet, the part that faces the wall.

That definition matters now, because it tells you what can and can't be automated. A model can produce the front of the cabinet beautifully. It has read millions of cabinets. What it can't do is care about the back, because caring requires something to be at stake for the one doing it. Fluency is not understanding, and the same gap runs through craft. A generated interface has the form of considered work without anyone having considered it.

There's a second thing craft was, and it's the one we talk about least. It was how people learned. The four-pixel afternoon wasn't only about the navigation bar. It was where I learned to see. Every junior designer who obsessed over a kerning pair, every developer who rewrote a function three times because the first two were ugly, was building an eye they would use for the rest of their career. Craft was the work, and it was also the apprenticeship.

What Has AI Taken From Craft?

The loss is real, and it isn't evenly spread.

AI didn't take the top of craft. It took the middle. The competent landing page, the serviceable component library, the clean but unremarkable article, the function that works and nobody will ever read. For years that middle was where most paid work happened, and where most people who called themselves craftspeople actually spent their days. It's now nearly free.

When the middle is free, the market stops paying for it. A client who can get a decent homepage in an afternoon will not pay for the four-pixel version, not because they can't tell the difference but because, most of the time, the difference doesn't change the number they care about. That's not cynicism. It's how markets have always treated craft once a machine could make the median version. Hand-weavers learned it two hundred years ago. Typesetters learned it forty years ago. We're learning it now, faster.

The harder loss is the one I named above. The middle was also the ladder. Junior designers learned to see by doing the competent work a hundred times until they could do it well. Junior developers learned architecture by writing the boring parts and having someone senior tell them why it was wrong. That work is exactly what AI now does first and does adequately. So the question isn't only whether craft survives. It's how anyone develops it when the work that used to train the eye is no longer handed to people who are learning.

I don't have a clean answer to that, and I don't trust anyone who says they do. It may be the most important open question in my field.

What Parts of Craft Survive AI?

The generated version can't tell you whether it's right.

Every competent output still has to be judged by someone. Is this the right page, or just a plausible one? Does this interaction respect how the person using it will actually feel at 11 p.m. when the payment fails? Is this code going to hold up when the third integration lands, or is it a confident sentence built on a detail nobody checked? I've watched a team nearly reposition a company around a competitor fact a model made up. The document looked finished. Finished is what we'd learned to trust.

That judgment is craft. It always was. We just used to exercise it with our hands on the work, so it was hard to see it as separate from the making. Now the making is cheap and the judgment is exposed, standing on its own.

Three things survive, and they're worth more than they were.

Taste. Knowing which of ten competent options is the right one, and being able to say why. Models are very good at producing the average of everything they've seen. Taste is the ability to see when the average is wrong for this person, this brand, this moment.

The last ten percent. Generated work gets you most of the way, quickly. The final stretch, the edge cases, the microcopy that sounds like a human, the motion that feels earned, the decision to cut a feature entirely, is still where most of the value sits. It used to be ten percent of the time. Now it's most of the job.

Care. The back of the cabinet. Someone still has to decide that the error state matters, that accessibility isn't optional, that the thing should be right and not just fine. A model will do it if asked. Someone has to want to ask.

Where Does Craft Go From Here?

So craft isn't dying. It's moving, from the hand to the eye. From making the thing to knowing what the thing should be and recognizing when it isn't.

That sounds like a promotion, and for some people it will be. The designer who used to spend a week on a component can now direct fifty and spend the week on the three decisions that matter. The developer who used to hand-write boilerplate can spend the time on architecture nobody will think about until it saves them. The work gets more strategic. In my world, that's not a small thing.

But there's a paradox in it I can't get past. The eye was trained by the hand. I can tell when a layout is off by four pixels because I spent years moving things four pixels at a time. My judgment is made of all that slow, unglamorous making. If the next generation never does the making, where does their eye come from?

I think the honest answer is that we have to build the apprenticeship on purpose now, because the work won't build it for us anymore. That means letting juniors make things by hand sometimes even when a model could do it faster, the way musicians still practice scales. It means senior people narrating their judgment out loud instead of just exercising it, so the reasoning behind the eye becomes visible and teachable. It means reviewing generated work the way a good editor reviews a draft: line by line, asking why, not just whether.

The businesses that do this will have people who can tell good from plausible. The ones that don't will have a lot of competent output and nobody who can say which of it is right.

Machines made the middle of craft cheap. They didn't make caring cheap. They made it rare. And in a world full of things that look finished, the person who still cares about the back of the cabinet might be the most valuable person in the room.

Frequently Asked Questions

Is craftsmanship dead because of AI? No. AI has made competent, median work cheap, but judgment about what's right, the final details and the care behind them are worth more than before.

What is craftsmanship? The sociologist Richard Sennett described it as the desire to do a job well for its own sake. In practice it's the hundreds of small, invisible decisions that make work feel considered.

How will designers and developers learn craft if AI does the basic work? Only if organizations build that learning on purpose: having juniors make some things by hand, seniors explaining their judgment out loud and teams reviewing generated work line by line.

Evan Prout is a strategist in Grand Rapids, Michigan, who writes about how organizations think and decide with AI.