Strong piece, and the operating advice holds up. Own your learning loop. Measure against your own outcomes, not public benchmarks. Treat models as swappable parts. That’s right, and most firms still don’t believe it.
One question the essay walks around: where does the loop actually run?
You can swap the generalist model without losing your company veteran. Fine. But the evals, the RL environment, the memory store, the orchestration all sit on someone’s stack. Move the moat up a layer and you haven’t escaped lock-in. You’ve relocated it from the model to the platform.
The framing offers two futures: a few models eat everything, or value flows broadly to every firm. There’s a third. Value concentrates on the rails the loop runs on. Swappable trains, one set of tracks, a toll on every trip.
Good advice. I’d just ask who collects rent on the ground it all sits on.
@Kinza1278 I dated a girl who was messy like this. That was an instant -2 points. She hits -3, I run. I run as fast as I can in this situation.
Hygiene is super important to me.