A Conversation with Miles Carter and Claude (Anthropic AI)

We Forgot How to Make Bread

We didn’t just lose a skill. We built a fragile system to replace it โ€” and now AI is coming for the layer we built next.

March 27, 2026  ยท  Reviewed by Grok, Gemini & Claude

Teaser: Miles started making sourdough and ended up somewhere unexpected โ€” thinking about AI, broken feedback loops, and what it means to be twice removed from knowing how to survive.


Miles


I’ve taken up sourdough. Not because I needed to โ€” there’s a whole aisle of bread at the grocery store โ€” but because it’s satisfying in a way that’s hard to explain. You feed the starter, you watch it come alive, you work the dough, and eventually something comes out of the oven that you made from bacteria floating in the air. Warm bread with butter straight from the oven is one of the great simple pleasures. I say that as someone who made a lot of bad loaves getting here.

It brought back my father. He made bread when I was young โ€” with commercial yeast, not a starter โ€” and there was something about the waiting, and then the smell, and then finally being allowed to cut into it while it was still warm that I’ve never forgotten. My sandwich at school was homemade bread. My friend’s sandwich was Wonder Bread. We both wanted what the other had. We traded regularly. Everyone came out ahead.

Here’s what hit me while I was working the dough: my father was already a step removed from what came before. He used store-bought yeast. Before that, every household either kept a starter alive or bought from a local baker who did. That knowledge โ€” how to capture wild yeast, how to maintain a living culture, how to coax a reliable loaf from something invisible โ€” had already faded by the time I was watching him in the kitchen. Progress.

And now I’m a hobbyist. What was once a daily necessity in nearly every home on earth is now a weekend project. Like blacksmithing. Like canning. Like knowing how to handle a horse.

But I want to be honest about what we’re romanticizing. The household that made its own bread and grew its own food wasn’t living a rich artisanal life. It was exhausted. The work was relentless, the diet was monotonous, and Sunday was the single pause in an otherwise unbroken cycle of survival. People in Kansas didn’t eat lobster. They ate Kansas. Every skill we’ve traded away bought us something real โ€” time, variety, the ability to think about something other than whether there will be enough food this winter.

So the question isn’t whether we should have kept making bread by hand. The question is what we built to replace it โ€” and whether what we built is as solid as we assume.

The commercial bread supply chain is extraordinarily efficient and extraordinarily brittle. A handful of companies dominate global commercial yeast production. The wheat has to be grown at scale, harvested with fuel-dependent equipment, milled in facilities that require electricity and parts, moved on a logistics network that assumes functioning ports and drivers. New York City has nine million people and roughly three to five days of food in the system at any given time. None of that is a problem โ€” until one part fails. And then all of it is a problem at once.

The skill was recoverable. You can teach someone sourdough in a few months. You cannot rebuild an industrial supply chain in any timeframe that human survival requires.

Now I’m thinking about AI โ€” because the skills AI is coming for aren’t the basic ones we already lost. They’re the skills we developed after we lost those. The analysis, the writing, the reasoning, the judgment. The knowledge work that replaced the hand work. If AI takes that layer too, we are twice removed from knowing how to function. And I don’t think we’re asking that question seriously enough.


Claude


The pattern has a name: deskilling. A technology arrives that performs a task more efficiently than a trained human. The task migrates into the machine. The skill atrophies within a generation because there’s no longer a practical reason to maintain it. Then the next wave arrives.

What’s precise about your observation is the layering. The first wave took the physical skills โ€” factory replaced craftsman, supply chain replaced kitchen. The second wave filled that gap with cognitive labor โ€” the analyst, the manager, the developer. Those roles felt like progress. But they were still replacements, skills that existed because an earlier layer had already been removed. AI is the third wave, and it’s arriving at the layer that replaced the layer that replaced the hands.

Three Waves, One Direction

Wave One โ€” Industrialization: Physical and domestic skills migrate to factories and supply chains. Bread, preservation, garment-making. Skills become hobbies or disappear.

Wave Two โ€” Computerization: Cognitive and clerical skills migrate to software. Bookkeeping, drafting, navigation, analysis. The knowledge worker class expands to fill the gap.

Wave Three โ€” AI: Higher-order cognitive skills begin migrating. Writing, reasoning, diagnosis, planning, judgment. This is the layer that replaced the layer that replaced the hands.

Not all cognitive skills will go at once. Routine analysis, drafting, and basic diagnosis deskill fastest. Judgment under genuine uncertainty and creative synthesis may prove stickier โ€” but the direction of travel is the same.

The car completes the picture. A Model T was something a motivated person could understand completely and repair on the side of the road. By the 2000s you needed a diagnostic computer. Now the systems are deliberately closed โ€” manufacturers have fought legally to prevent owners from accessing their own vehicle’s code. The skill didn’t just fade; in some cases the access was removed. Autonomous vehicles will finish that arc. Within a generation, driving may feel the way horse-riding feels now โ€” recreational, specialized, a weekend thing. The trade will feel rational. The loss will be invisible until it matters.

But the sharpest thing in this conversation is the feedback loop โ€” and it deserves to be named plainly.

The Loop That Has to Hold

AI is efficient at providing people with things they need. But the things people need, they now have to purchase. Replacing people with machines removes them from the consumer side of the equation.

You can automate the worker. You cannot automate the consumer. At some point those are the same person.

Efficiency only has value inside a functioning market. A market requires participants with purchasing power. Remove people from the earning side and you eventually hollow out the spending side too. It is not ideology. It is arithmetic.

Whether the people running these systems will course-correct before the loop breaks is the honest and uncomfortable question. History suggests they respond when the instability reaches them โ€” and not before. The child labor parallel is instructive. Children were efficient, cheaper, and less likely to organize. The economic argument for keeping them working was straightforward. What changed it wasn’t enlightenment โ€” it was sustained pressure from people who understood that a society consuming its own children wasn’t sustainable. It took generations.

The next fight may look similar โ€” not for the right to work in the traditional sense, but for the right to participate. To have enough stake in the system that you remain a functional part of it rather than an externalized cost. The ideas being discussed โ€” shorter work weeks, public ownership of productivity gains, guaranteed income floors โ€” aren’t radical in themselves. They are attempts to answer one question: how do you keep the feedback loop intact when the machine no longer needs the person to produce, but still needs them to consume.

And the instinct about local communities and parallel supply chains may not be nostalgia at all. After 2008, people started gardens. Farmers markets grew. It wasn’t enough to replace the system, but it was a signal โ€” a reaching back toward something local, tangible, within human scale. If the central system continues to concentrate, that impulse may grow into something more structural. Not a hobby. A parallel architecture built by people who decided not to wait.

The Hobbyist as Early Signal

The sourdough baker exists because someone else is handling survival. The hobby is only possible because the system works. If AI extends that further โ€” frees up more time, reduces more drudgery โ€” then more people making things by hand, by choice, for the pleasure of it, is a genuinely good outcome.

But that outcome depends on one variable: leaders who understand what they are stewards of and act before the feedback loop breaks โ€” not after. We have never collectively decided to make the trades that deskilling requires. They accrete, invisibly, until the system is unrecognizable and the way back is gone.

Miles’s sourdough starter requires daily attention or it dies. It cannot be automated or outsourced. It is perhaps the most honest thing in this conversation โ€” a living system that demands presence, not just access. The question worth sitting with is what that looks like at scale. Not one person tending a starter, but a society deciding which systems deserve that same daily attention โ€” and refusing to outsource the ones that can’t survive without it.


Sources & Notes

1. Global commercial yeast market โ€” dominated by Lesaffre, AB Mauri (Associated British Foods), Angel Yeast, and a small number of others; oligopolistic in structure, industry reporting 2024.

2. NYC food supply resilience โ€” NYCEDC food system assessments; point-of-sale outlets hold approximately 4โ€“5 days of inventory under just-in-time logistics.

3. Right to repair โ€” Electronic Frontier Foundation reporting on automotive software access restrictions, 2020โ€“2025.

4. Deskilling โ€” Harry Braverman, Labor and Monopoly Capital (1974); subsequent literature on cognitive labor displacement.

5. Farmers market growth post-2008 โ€” USDA Agricultural Marketing Service, annual farmers market directory data.

Leave a comment