Authority centralizes decisions with whoever has the title. Alignment distributes them to whoever has the context, which is almost never the same person.

Primacy bias, recency bias, sycophancy, and anchoring are predictable distortions in how LLMs weight information. Understanding them changes how you prompt, evaluate, and trust model outputs.

Concentrating decisions at the top discards the intelligence of people closest to the work. Marquet’s leader-leader model distributes authority to where information lives, producing organizations that outlast their leaders.

How do you avoid killing something too early, or celebrating too early. And last, how do you know when to kill a dud?