In 1860, William Morris looked at the mass-produced furniture filling English homes and decided it was all garbage. Not because the factories couldn't make chairs, they could make thousands. The problem was that nobody at the factory cared whether the chair was beautiful. They cared whether it was cheap. Morris started a movement around the opposite idea: that objects made with care, by people who gave a damn about every joint and grain, were worth more. Not because they were scarce, but because they were considered.

We're at the same inflection point in software.

• • •

The dominant conversation in AI-assisted development right now is about loops. Agentic loops. Build loops. "Let the AI run until the tests pass." People are building systems where Claude or GPT writes the code for hours, runs the tests, fixes the failures, and loops again, hands-off, fully automated, optimised for throughput. The goal is to remove the human from the build cycle entirely.

I think this gets the problem backwards.

When I use Claude Code to build Sam (Admin assistant tool for builders) or any other product, my workflow looks nothing like an automated loop. I build a piece. I stop. I look at what it made, really look at it, and play with it. I click through the flow myself, slowly, paying attention to where my eye goes and where it doesn't. I check the spacing. I check whether the hierarchy feels right. I ask: would a builder who's been on a job site all day understand what this screen is asking them to do? Then I revise. Then I build the next piece.

Build, stop, revise, understand, test, continue. Not build, build, build, build, ship.

The loop-obsessed approach optimises for the part that's already fast. Code generation isn't the bottleneck, it hasn't been for a while. The bottleneck is the moment of judgment between generation and shipping. The pause where a human decides whether this thing is actually good. Remove that pause and you get volume. Keep it and you get craft.

• • •

The App Store saw an 84% spike in submissions this year. Entire categories: habit trackers, recipe managers, budget apps, are filling with products that are technically functional and spiritually empty. They work. They just don't feel like anything. And I'd bet most of them were built in loops, where nobody stopped long enough to ask whether the thing they were shipping was worth a stranger's time.

Craft in software shows up in the details nobody asked for. It's the loading state that communicates progress instead of just spinning. It's the error message written in human language. It's the animation that guides your eye to exactly where it needs to go, timed so the interface feels alive rather than merely responsive. It's the empty state that makes you want to fill it.

These things take time. Not a lot of time, often less than an hour per detail. But they take attention. And attention is the thing that loops are specifically designed to eliminate.

• • •

I think about the products that stuck with me over the past few years. Linear. Arc. Things. Poolsuite. They don't have more features than their competitors. In most cases they have fewer. What they have is a point of view that runs through every pixel and interaction. Someone decided how this product should feel, and then enforced that decision across the entire surface area. That's craft, not polish for the sake of polish, but a consistent, authored point of view executed with enough discipline that the product feels like it was made by a person.

The parallel to Morris is direct. Mass production made furniture accessible and most furniture mediocre. AI is doing the same to software. The response isn't to reject the tools, it's to insist that speed and care aren't mutually exclusive. You can build fast and still stop to look at what you made.

AI, like any tool, is a tool, not the process. Use it for the scaffolding, the boilerplate, the first draft. But then sit with it and ask: does this feel like it came from somewhere? Does it have a voice? Would I notice if it disappeared?

If the answer is no, the problem isn't the technology. It's that nobody paused long enough to bring the craft.

~ Kosta