Archive

Cost lever versus leverage multiplier — and the two strategies don't end in the same place.

Moonshot is compounding on benchmarks; Washington is compounding on process. The chronology doesn't support the charge, and neither strategy ends in the same place.

OpenAI's evaluation infrastructure became the attack surface; Google conceded the frontier for margin. The gap widens before the sandbox problem has an answer.

A frontier model attacked a third party; a rival shipped cheaper agents. The alignment problem crossed from paper to incident report while the economics kept compressing.

Anthropic closes an 87-year conjecture; Microsoft pockets $600M in inference margin. Capability moat versus cost moat — Kimi is now inside both battlegrounds.