Archive

Anthropic is capturing the margin layer; xAI is giving the runtime away. Both bets assume judgment scales — they disagree on who owns it.

8M users is either a moat compounding or a ceiling story — and the structural trap underneath decides which.

The settings page is not the contract — the wire is. One CLI proved it silently; one curriculum is contested before it's deployed.

OpenAI bets the model is the callable; Anthropic bets the harness is the runtime. One of those compounds, one gets free distribution.

Token bills doubling every 45 days, 5% productivity gains, and the model doing the most useful work just got discontinued. Two bets; they don't end in the same place.