I think Anthropic and OpenAI have found product-market fit

The Leap from Tech Toy to Production Tool

Simon Willison's article stems from an interesting observation: he personally spends $200 monthly on AI subscriptions but receives over $2,000 worth of API calls. This made him realize that individual and enterprise users are experiencing completely different pricing realities. More crucially, rumors suggest Anthropic is about to have its first profitable quarter, while companies are surprised by ballooning AI bills. Underlying this is a fundamental shift: AI coding assistants are no longer optional experimental tools but essential components deeply embedded in core development workflows.

Unpacking the Business Logic Behind Pricing Strategies

The core change lies in the shift in pricing models. Previously, enterprise plans typically used a "fixed fee per person per month + included usage" model, similar to SaaS subscriptions. Now, both Anthropic and OpenAI have moved to a "base seat fee + API usage-based pricing" model. It's like switching from an "all-you-can-eat buffet" to "pay per dish."

Why would enterprises accept this seemingly more expensive model? Because the tools are genuinely too good to abandon. When developers become accustomed to using Claude Code or Codex for writing, debugging, and refactoring code, the productivity gains outweigh cost considerations. This reveals a deeper trend: once AI tools achieve stickiness, enterprises lose bargaining power and must accept new pricing rules.

Trend Insight: AI Repeating Cloud Computing History

This mirrors early cloud computing stories. Initially, enterprises were shocked by AWS bills, but eventually realized that on-premise infrastructure couldn't compete on cost and flexibility. Now, AI is following the same path: from "this is too expensive" to "we'll fall behind without it."

A key signal is that API revenue is becoming "less important" for these companies—not because it's unprofitable, but because direct enterprise subscription revenue is growing faster. This indicates mainstream enterprise clients have moved beyond pilot phases into full deployment.

Practical Value: How Should Developers and Enterprises Respond?

For individual developers, now might be the last window to lock in low-cost subscription plans. As enterprises begin paying per API, subsidies for individual users may become unsustainable.

For tech managers, immediate monitoring of AI tool usage is necessary—just like cloud resources. Establish internal cost allocation mechanisms, distinguish between "high-value tasks" and "abuse scenarios," and avoid situations like Uber's "AI budget explosion."

More critically, reassess the strategic position of AI tools. It's no longer an "R&D department experiment" but core infrastructure that needs inclusion in overall IT budgets and productivity assessments.

Counterintuitive: The Absence of AI "Failure Stories"

A keen observation in the article is that despite companies complaining about high bills, there are almost no stories of "we stopped using AI tools because they were useless." This indicates AI coding assistants' value has been universally validated; the issue is cost optimization, not value questioning. This is precisely the strongest evidence of product-market fit—users complain about prices while being unable to stop using the product.