Simon Willison Says It Out Loud: Coding Agents Are the Product-Market Fit

Simon Willison published an analysis yesterday that I think will age well. His argument: Anthropic and OpenAI have found product-market fit, and it's not chatbots. It's coding agents.

The numbers back him up. Anthropic's rumored Q2 2026 revenue hit $10.9B, up from $4B in August 2025. A SpaceX S-1 filing revealed a $1.25B/month cloud services agreement with Anthropic through May 2029. Both companies shifted enterprise plans from flat-rate seats to per-token usage pricing in April 2026. That pricing change is the tell. When you move from "pay per seat" to "pay per token," you're saying the heavy users are your best customers, not your most expensive ones.

Willison identifies November 2025 as the inflection point. That's when GPT-5.1 and Opus 4.5 shipped, and both companies' agent harnesses got reliable enough for daily professional use. I buy this timeline. I use Claude Code every day in my personal projects, and the jump from Opus 4.0 to 4.5 was the moment I stopped thinking of it as a tool and started thinking of it as a coworker. The 1,114 HN comments on this piece tell me I'm not alone.

What Willison gets right that most analysts miss: the people paying for these tools are professional software engineers at companies with budgets. A $200-$2,000/month tool that makes a $200K/year engineer 30% more productive is the easiest ROI calculation in enterprise software. The tool sells itself to anyone who can do arithmetic.

The shift to usage-based pricing is a double-edged sword, though. It means revenue scales with adoption, but it also means costs are unpredictable for buyers. Story #2 shows what happens when that unpredictability hits a large org.

What builders should do: If you're selling developer tools, the pricing model just changed. Per-seat is dead for AI coding tools. Build usage-based billing into your product from day one. If you're buying these tools, start tracking per-engineer token spend now, before your CFO does it for you.