Why utilization metrics will kill your AI program — and what to track instead
You'd never bonus a translator on words-per-minute — you'd bonus them on closed deals.
AI deserves the same treatment. Stop counting tokens; start counting what shipped. Activity is easy to count. Value is the only thing worth counting.
shipped in 2 days
by Anthropic's Head of Claude Code — without writing a single line of code himself. Measure outcomes, not activity.
Activity is easy to count. Value is the only thing worth counting.
Three pieces from the last 30 days that reinforce this analogy. Forward to the executive who's still calling it hype.
Run-rate revenue, not token count. Closed customers, not API calls. Anthropic itself measures outcome, not activity.
Read the original →Benioff's framing: the spend is justified by specific customer-facing capability shipped — not by volume of prompts.
Read the original →The translator-by-the-word problem in reverse: count features shipped, not prompts written. The model produced; the human directed.
Read the original →The catalog is free. The live session is where Nobody Pays a Translator by the Word gets applied to your specific stuck conversation — built for boards, ELTs, and founder-CEO 1:1s. Half-day, $7,500.