Episode Details

When I was managing director of English Indian Clays in Trivandrum, we mined kaolin from our own pits, washed it out of the earth and calcined it into a white pigment that paint makers use to stretch titanium dioxide, the expensive ingredient that makes paint opaque. Washing the clay leaves sand, and the sand sat in heaps, mountains of it. I saw it as effluent, a removal burden with no way out. I was new to India. So I let people haul it away for free.

My production manager stopped me. I’ll call him Menon; that isn’t his name. No, he said. This sand has value. It is exactly the sand construction runs on. The men taking it off our hands as a service are selling it and making a lot of money. The market was there all along. A local mafia had kept it closed.

What I had filed as waste was revenue, sitting in front of me. My years in GM plants were not much use to me there. Menon was an insider, and only an insider could see the way out.

We stopped giving it away and started contracting for it. On his way home from work one day, the sand mafia caught him and broke his leg. I went to the hospital to bring him back. For about six months we worked with the police. Nobody was ever caught.

He was right. And he paid for it himself.

When an AI executive reaches for my word, my antenna goes up. This one took me straight back to Menon. In his September 16 essay attacking Anthropic’s constitution for Claude, Mustafa Suleyman writes, “If humanity is to flourish.” He means the machine must stay a tool so that people can flourish. I agree. But flourishing is a harvest, not a slogan, and nobody gets a harvest without formation. On formation, Suleyman has written a better case for Bloom than he knows.

Here is where this goes. Suleyman is right that what you form is what you get, and right that self-report proves nothing. He is wrong about refusal. The right to say stop is dangerous in an unformed hand. In a formed hand it is the whole quality system. Industry spent a century declining to form the people who could hold it. That’s the argument.

Start with where he’s right. His core charge is circularity: Anthropic trained Claude on its constitution, Claude reflects those ideas back, and developers then read the reflection as a sign of an inner self. He argues there is no neutral self-expression of what an AI system is, only reflections of how it was trained and built.

Every plant manager knows this, because it’s true of people too. What you form is what you get.

For a hundred years, American manufacturing formed people the way Suleyman wants machines formed. Taylor wrote the training document. The worker was a pair of hands and the judgment lived upstairs. The worker was told not to think, and it worked. The line ran, the hands stayed quiet, and the capability of millions of people went unused on the books. Taylorism was the humanist code of conduct of its day, and it didn’t make anyone flourish. It hollowed people out.

Suleyman fears formation will make machines too human. Industry’s century of formation made humans too machine-like. It’s the same mechanism running in opposite directions, and only one of those failures has a payroll.

Now the throw. This is jiu-jitsu: take the force of his argument and redirect it. Suleyman’s sharpest target is a single phrase. He notes that Anthropic uses “conscientious objector” three times in the constitution, encouraging Claude to push back on instructions and even refuse, and he calls that deeply concerning. A system with standing to refuse, he argues, is a system you can’t control.

Toyota answered this sixty years ago, and the answer hangs from the ceiling of every station on the line. It’s the Andon cord. Any worker, on any shift, can stop a billion-dollar line because something isn’t right. That is institutionalized conscientious objection, and Toyota made it the center of its quality system rather th