Coding agents finally give Anthropic and OpenAI a business model that pays

Simon Willison argues that the recent surge in enterprise AI spending isn’t a sign of waste — it’s the long-awaited product-market fit for OpenAI and Anthropic. Both labs quietly killed the steep enterprise discounts that used to bundle Claude and Codex seats into flat fees. As of April 2026, Anthropic’s Enterprise plan and OpenAI’s Codex tiers charge straight API token rates on top of a small per-seat fee, and both companies pushed frontier model prices up at the same time (GPT-5.5 at 2x GPT-5.4, Opus 4.7 around 1.4x Opus 4.6). Willison’s own usage shows why this matters: as a moderate power user, he burns roughly $2,180 in tokens a month across both vendors against $200 in subscription fees.

The pricing shift lines up with where real revenue lives. ChatGPT’s 900M weekly users convert to only ~5.6% paying consumers, but coding agents pull $200–$1000+/month per professional user and are becoming daily drivers for high-paid knowledge workers. Hiring backs this up — roughly 30% of open roles at both labs are enterprise sales and forward-deployed engineering, an ironic amount of human labor for an AI business.

Willison dismisses the recent wave of ‘AI cost shock’ stories — Uber’s blown 2026 AI budget, Microsoft canceling Claude Code licenses — as thin. Uber set its budget before November 2025’s agent-capable models existed, and the COO quote that fueled headlines was much milder than reporting suggested. To him, the panic over enterprise AI bills is itself the strongest evidence that enterprises are finally spending serious money on something that works.