What I learned about power from LBJ and Robert Moses

Dwarkesh Patel made a video essay on Robert Caro's biographies of Robert Moses and LBJ, built as a warm-up for a Caro interview. No model release, no pricing move, no capability claim. A history lesson.

The essay's core argument: both men built durable power by controlling the machinery everyone else ignored. Moses wrote himself into bond covenants, the fine print of how a public agency's debt is structured, so that removing him would have required a constitutional convention. Johnson asked what formal powers the Senate majority leader actually held and got one answer: the right to speak first on the floor. He turned that nothing into the second most powerful office in the country by owning committee assignments, bill scheduling, and the flow of who-wants-what information.

There is a parallel to AI infrastructure, though it's my read, not Patel's. Power in the stack you depend on doesn't sit in the model weights. It sits in the covenant-equivalent fine print: the data-retention clause, the deprecation notice, the rate-limit tier. That fine print decides whether you can switch providers in a weekend or get locked in for a year. Right now OpenAI, Anthropic, Google, and the cloud vendors underneath them write that fine print. Most buyers sign it without reading it.

Moses and Johnson figured out where control actually lived before anyone else thought to look. The buyers who figure that out in AI infrastructure are going to be in a much stronger position than the ones who signed the covenant without reading it.

No prediction this week.

This one is a history lesson, not an AI story. Dwarkesh Patel made a video essay on Robert Caro's biographies of Robert Moses and Lyndon Johnson, built as a warm-up for his Caro interview. The whole thing is about how two men built and held political power. There is no model release, no pricing move, no benchmark, no capability claim. Nothing that changes a tool you use or a bill you pay.

So I'm not going to pretend otherwise. No council. No prediction. Here is the one thread worth pulling for an AI operator, and then I'll get out of the way.

The essay's core claim is that Moses and Johnson built power out of nothing by controlling the machinery everyone else ignored. Moses wrote near-permanent authority into bond covenants, the fine print of how a public agency's debt is structured, so that removing him would have needed a constitutional convention. Johnson asked the Library of Congress what formal powers the Senate majority leader had and got one answer: the right to speak first on the floor. He turned that nothing into the second most powerful office in the country by owning committee assignments, bill scheduling, and the flow of who-wants-what information.

The parallel to AI, if you want one, is about where control actually sits in the stack you depend on. It isn't the model weights. It's the covenant-equivalent fine print: the data-retention clause, the deprecation notice, the rate-limit tier, the thing that decides whether you can move providers in a weekend or get locked in for a year. Whoever writes that fine print holds the Moses position. Right now that's OpenAI, Anthropic, Google, and the cloud vendors underneath them, and most buyers sign it without reading the covenant.

That's a real lesson. But it's mine, bolted onto a history essay, not something the episode argues. Dwarkesh Patel never mentions AI here. If you came for his lab-founder interviews, this isn't one.

No high-conviction prediction this week.