Motivation

It took factories decades to stop building around one giant motor. AI is at that same moment with knowledge. I believe knowledge belongs with the person at the counter and the mom-and-pop shop making ends meet, not in a few headquarters. Here's why. It starts in a factory.

01One motor in the middle

Picture a big American factory at the end of the 1800s. One giant steam engine or water wheel turns line shafts that run the length of the building, and belts drop from them to every machine. The belts swirl dust and grease through the air and can maim or kill anyone caught in them. To change one department, you shut down the whole mill.[1]

You'd think electricity would end that fast. It didn't. In 1899, electric motors made up less than 5% of the power driving American factory machines. It took two more decades to get just past half, four decades after the first central power station opened. And factories that electrified mostly hung the new motors on shafts and belts.[1]

They weren't fools. Farsighted engineers could see what was coming by 1900. But power was still expensive, the buildings and shafts still had years left in them, and few people had learned how to build the new kind of plant. Keeping the old arrangement made sense to the people who owned it.[1]

The real unlock was a motor at each machine, as motors got cheaper and smaller.[2] Then the shafts came down. Skylights went in where they'd hung. Factories went to one story and laid themselves out around the work. The system was modular, so you could rework one line without stopping the rest. Manufacturing productivity jumped in the 1920s, and the economist Paul David ties about half of it, statistically, to the spread of electric motors.[1]

I'd argue today's AI is still hung on the old shaft. When I lived in San Francisco, the hype was all about replacing roles: same company, same org chart, a bot where the sales rep used to be. I think that's wrong, and I think it wastes what AI could be.

The power still came from a station somewhere, and AI is like that today: Roast Harness runs on big models in big data centers. What moved was the motor. Who should own it is a separate question, and I'll get to it.

What it means for you: the know-how that used to sit with specialists comes to your counter, and it answers to you.

02At the counter

Cafés still matter. They're a wonderful place for a community to gather, and I understand how thin the margins are. About two in three American adults drank coffee yesterday.[4] I believe millions of them pay far more than coffee should cost, and so do the cafés serving them. Take a café that goes through 40 pounds a week at $12.85 a pound. On our estimate, buying roasted costs it $26,728 a year. Roasting its own with Roast Harness costs $15,145 a year once it's running, after about $4,841 up front, and it comes out ahead even after paying someone for every hour at the machine.[5]

I'm not blaming the small roastery down the street. Part of that gap pays for real work: the machine, the rent, the person running it. The rest, I'd argue, is the price of knowing how.

And knowing how doesn't need a head office. In Ethiopian and Yemeni communities, people often roast their own.

In 2008, Jennifer 8. Lee put the number of Chinese restaurants in America at about 40,000, more than three of the biggest fast-food chains combined.[6] Most of us in America lost that kind of knowledge. I believe AI is how it comes back.

So who does this put in power? Anyone who roasts their own, or drinks at a café that does, because they finally get a say over what's in the cup. And the small shops whose tools and ingredients, I'd argue, get taxed at every step by giant centralized businesses. If someone else owns the machine, the recipes, the beans and the service contract, that's still the motor in the middle with a nicer belt.

03The job question

I'll be straight. Someone still has to be at the machine, for the whole roast, with a hand near its power switch. They pour in the beans and drop the batch when the agent calls it, and if anything looks wrong they stop it. What they don't have to be is a trained roasting specialist. So doesn't that cost someone a job?

In the shops I'm talking about, mostly no, because that job never existed there. Most ordinary cafés can't justify hiring, training and scheduling roasting specialists and paying a consultant to set them up. On our estimate, that same café doing it that way ends five years $17,500 behind buying roasted, even counting potential bag sales.[5] So it keeps buying roasted.

Where it can bite is the big centralized roasteries. The Census counted about 24,000 people working in all of American coffee and tea manufacturing in 2023, more than half of them at just 50 plants.[7] Some of those jobs could go, and I won't pretend otherwise. The difference from a bot in the same org chart, as I see it, is who ends up with the money: the shop and the person drinking the coffee. I believe it empowers more people than it replaces. That's no comfort to someone losing a job, and I won't tell them it is.

04Why I believe it

At hackathons, for years, I built travel planners, food recipe tools and thought partners, and I never really saw one through. Now, on my own, I've built an agent that writes, rehearses and runs a roast, and published a benchmark for AI-written roasts, limits and all. I didn't get that much smarter. The tools got that much better, and I'm my own case study for what that does to the playing field. I believe AI is fundamentally a decentralizing force, and it only gets there if people take it there.

That's why it's disheartening to see so much of it aimed at things that hardly help the average person. I believe in the technology. What bothers me is how it's being developed and who has a voice in it. My aim as a solo developer is to make products that restore trust in AI.

05Why people don't trust it

They have good reasons.

• Data centers. 71% of Americans oppose an AI data center in their area, more than the 53% who oppose a nuclear plant there.[8]
• Layoffs. By one firm's count, employers cited AI for 121,401 announced job cuts through October, though the monthly number has fallen since May.[9] And 79% of Americans expect AI to cut US jobs over the next decade.[10]
• Mandates at work. Some big employers reportedly require AI use or tie it to promotions.[11]
• Slop. Merriam-Webster's word of 2025 was "slop": low-quality content made in quantity by AI.[12]

Only 27% of Americans trust businesses to use AI responsibly.[10] In San Francisco I saw billboards telling companies to stop hiring humans.[13] I found that appalling, and I can't think of a clearer sign of how out of touch this industry has become.

Most of us feel we have no say. Some 57% of Americans say they have not much control or none over whether AI is used in their lives, and 61% want more.[14] I don't think a better ad fixes that. I think you fix it one tool at a time: one where the roasts stay yours, that saves you real money and shows its work.

06Who keeps the tablet

Some of the oldest writing ever found is a record of beer: a clay tablet from around 3100 BCE, now in the British Museum, counting rations for workers. Tablets like it have been found discarded from large temples.[15] From the start, writing had food in it. Accounts, laws, math and recipes[16] are among the oldest things anyone wrote down.

Look at that tablet again. Whoever kept it knew what the whole temple had. Whoever drank the ration knew their own cup. I'd argue that gathering up knowledge and skilled people has been a source of power for thousands of years, long before any industrial revolution: the temple with its scribes, the court with its artisans and orators, the corporation with its specialists. AI could undo that. It could also lock it in.

We've seen this fork. In one working paper on American manufacturing, cheap electricity raised productivity quickly where firms were already big, with little new hiring, and the number of firms fell. Where firms were smaller, productivity took until 1940 to move, but output, jobs and the number of firms all grew.[3] Same electricity, different markets. It's history and it predicts nothing, but it's why I care whose hands AI lands in.

If AI arrives only in the hands of the companies that already hold the most skilled people, I believe they'll do what factory owners did with the motor at first: hang it on what they already have. Same org chart, a bot in the chair, fewer people. They'll keep their grip, and they don't have to be villains for that to happen. So it has to come from the ground up. I think it's time to reconsider who holds knowledge, because the more of what you need you can make yourself, the less anyone can charge you for not knowing how.

07Where it goes next

Coffee is what I've built. Clothing is what I think about next. Machines that knit a whole seamless garment already exist, though as of a 2018 paper they weren't widely used.[17] Put one within reach, add AI to design and fit what it makes, and I believe that for a sweater, at least, you won't need a factory. After that, maybe the small builder who can't keep an estimator and an engineer on staff. I haven't built that, and those fields have people in them too, so the same question comes with them.

Medicine is where I have to be careful, because the stakes aren't the same. Nobody should replace their doctor with a chatbot today. Bringing care home would take two halves, and neither is ready. The physical half, the equipment and robots, doesn't exist yet. The intelligence half isn't something to rely on yet: in a randomized study of 1,298 UK adults working through scenarios written by doctors, people using chatbots did no better than people using their usual methods.[18]

With more advances and proof, if trends continue, I believe care comes to the person, with undivided attention and at a fraction of the cost. I don't expect those who profit from today's arrangement to let that go easily. In one large study, hospitals with no rival charged privately insured patients about 12% more than hospitals with four or more.[19] I'm not blaming anyone on the ward. Centralization pays, and I believe it will defend itself wherever it does.

08What I'm asking

If you feel the same, write to me at jeff@nizamicoffee.com. That goes for anyone running a café who wants to roast their own, and anyone who backs or builds for small food businesses. And if you've been sitting on an idea, start your own AI project, alone or with a team. I'd rather see a thousand small motors than one more giant one.

They took the great shaft down and let in light; now bring the knowing to the hand that pours.