Back

I Think Anthropic and OpenAI Have Found Product-Market Fit

"Coding agents really did change everything"
Simon Willison article thumbnail captured from the source page

Recap

  • Opening signal: Willison starts with Anthropic profitability reports and companies reacting to larger LLM bills.
  • Personal usage check: He compares his own subscriptions with the much higher API-token cost his usage would have produced.
  • Enterprise pricing shift: Enterprise customers appear to be moving toward seat fees plus usage-based metering.
  • OpenAI Codex rate card: April 2026 Codex pricing moved toward token-based credit usage.
  • Product-market-fit claim: Coding agents are useful enough for companies to accept much larger per-user AI spend.
  • Consumer contrast: Consumer chatbot usage is massive but lower-value per user.
  • Hiring signal: OpenAI and Anthropic appear to be hiring heavily for enterprise sales and support.
  • Failure-story pushback: Cost complaints can signal waste, but they can also signal real adoption.
  • Frontier-lab economics: Labs need revenue streams that scale with actual work.
  • Caveat: Audited filings are needed to prove margins, retention, and durable profitability.

Context

Simon Willison is an independent developer and AI commentator. The source is a post on Simon Willison's Weblog published on May 27, 2026, titled "I think Anthropic and OpenAI have found product-market fit." The post argues that Anthropic and OpenAI may have found product-market fit through coding and general-purpose agents. Willison discusses Claude Code, Codex, token-based pricing, enterprise renewals, LLM bills, OpenAI's Codex rate card, Anthropic enterprise pricing, and job listings at OpenAI and Anthropic.

Technical Need To Know

  • Product-market fit: Repeated customer use plus willingness to pay.
  • Token: A unit of text or code processed by a model.
  • API pricing: Usage-based pricing, often per million input and output tokens.
  • Seat fee: A fixed per-user subscription that can hide heavy AI usage.
  • Coding agent: An AI tool that can inspect code, run commands, edit files, and help ship changes.
  • Enterprise renewal: A contract continuation point where new metering can appear.
  • Inference cost: The cost of running a model for users after training.
  • Return on investment: Whether agent spend connects to useful shipped work.

Nuanced Take

Agent usage has become a budget-line item. Painful bills can signal backlash, but they can also signal that the tools are doing enough real work to consume serious compute.