When software was a fixed tool, seat-based pricing made sense. You bought chairs at a table. Everyone paid the same whether they logged in once a week or built their entire workflow around it. The misalignment was tolerable because the capability was static. But AI products aren’t tools anymore. They’re capabilities that scale with model intelligence. The gap between Claude Sonnet 3.x and 4.x shows why flat pricing becomes a trap: a user extracting 100x more value pays the same as someone checking in occasionally. Here’s the deeper tension: $20/seat becomes your cost ceiling. Every model upgrade has to fit inside that box. You either compress the intelligence you provide or don’t upgrade at all. It’s a scarcity framework baked into the business model. Usage-based pricing trades predictability for possibility. Customers think before they act. Finance teams lose their clean spreadsheets. But it’s the only model that lets you give customers the best intelligence available without worrying whether your margins can support it. An abundance mindset instead of artificial constraints. This isn’t about one pricing model winning. It’s about flexibility that traditional software never required. Different customers value different things. Some want predictable costs, others want maximum capability. AI companies need pricing architectures that can serve both without forcing an either-or choice. Anthropic, Cursor, and a few others nailed the hybrid: seat-based pricing for predictability and simplicity, with usage-based pricing layered on top for power users who want access to the absolute best models. The top students in the class get to push harder without breaking the whole structure. Most knowledge work companies will resist this for a few years. The cognitive load is real. The change management is hard. But as AI becomes the actual work instead of a feature, pricing has to follow capability. Outcomes matter more than seats. The best companies will shift from software as a tool to intelligence as a service.