The Horse Left. The Craft Stayed.

This week, an Anthropic researcher resigned and warned that AI companies are gambling with our lives. Another senior researcher publicly shared his own fear that the technology could cause human extinction. These are people close to the work.

I don’t get to wave that off. Maybe they’re right.

I’ve also spent the last six months writing about what this technology has made possible inside a small company and my own family. Those experiences don’t settle the questions in the headlines. But they do leave me with a responsibility to think carefully about what I build, and why.

An unlikely place helped me think about that this week: a company that started making horse harnesses in 1837.

I listened to the Acquired episode on Hermès. Six generations of one family, with a business built around leatherwork for horse-drawn transportation. Then the automobile arrived, and the future of that business changed.

The craft had somewhere else to go.

The third generation brought the zipper to France and found new uses for it in leather goods. The company moved into bags and other products, carrying its knowledge of materials and construction into things people now needed.

The detail that stayed with me was the saddle stitch. Two needles working a single thread through the leather by hand. A construction valued for how well it holds together, including when a stitch is damaged.

Care you might never notice, until something goes wrong.

That’s the part I kept thinking about after the episode ended.

I went back through six months of my own writing and found a pattern.

The knowledge base gave a twelve-person team one place to look for information that had been scattered everywhere.

The app for my daughter helped us prepare questions about her medications for her pharmacist and cardiologist. I built it with a boundary: it can surface concerns, but it cannot declare something safe.

And last week’s issue described an hour a rep used to spend assembling a presentation by hand becoming time they could spend with the client.

The payoff was capacity. Hours returned to people for work that needed their attention.

I didn’t plan that as a theme. It’s what happened.

In our business, some of the work is changing: searching for information, moving it between systems, assembling a presentation. The relationship still needs someone who listens, understands what matters, and follows through.

That’s the craft I want the tools to make more room for.

I’m also learning that the tools themselves need more care than a working demo can show.

I spent a year building in the wrong order before I understood how much depended on the data underneath. Now I pay more attention to whether each step receives the information it expects, whether a failure is caught before it spreads, and whether someone can understand what happened when a system gets something wrong.

Most people using the finished thing will never see those decisions. They’ll experience them as information they can rely on, a problem caught early, or a clear place to step in.

The boundary in my daughter’s app is one of those decisions. A confident answer would be easy to display. Keeping the medical judgment with her care team takes deliberate design.

This is where the saddle stitch makes sense to me. The care goes into the construction, before anyone needs to trust it.

I want to move quickly where I can, and recognize the places where I haven’t earned the right to.

Sometimes that means another check. Sometimes it means a smaller scope, a person reviewing the result, or a feature I’m not ready to release.

None of that tells us whether the next generation of AI systems should be built, or under what conditions. My experience at DGW cannot answer that question. The warnings deserve scrutiny on their own terms.

But I can be specific about the responsibility on my desk: what I connect these tools to, what authority I give them, what I test, and where a person needs to make the decision.

The horse is leaving again. In my corner of the world, that looks like fewer hours spent gathering information and assembling things by hand. I want the time we recover to go toward our clients, our team, and our families.

What the labs build next, I can’t control. How carefully I build with what already exists is work I’m responsible for.

The web service and its three schema decisions are next week. This week, I needed to work through why those decisions matter.

P.S. If you’ve been reading since March, reply with the issue that stuck with you. I’m going to use the answers to help figure out what to build next.

Sources: The Verge, reporting on the Anthropic researchers’ warnings (September 9, 2026); Acquired, “Hermès” (February 20, 2024).