Anthropic says the retail-price equivalent of its internal Claude usage has exceeded employee salaries, excluding stock compensation, since April. September token usage was 250 times January’s, and the company also says Claude writes more than 90% of new code. Crucially, the comparison uses customer-facing prices rather than actual cash spending or inference costs. AI adoption comparisons should separate token volume, internal costs, staffing and review time; the company’s productivity claims are not independently established causal effects.