Anthropic's Claude Code vs Cursor

Anthropic launching Claude Code marked the moment the model supplier started taking the customer workflow for itself. Cursor had been selling an AI native IDE on top of Claude, but Claude Code moved Anthropic into the terminal, where developers actually run commands, edit files, test changes, and debug repos. That turned Anthropic from infrastructure vendor into direct application competitor, while also giving it first party usage data and a higher margin product.

• Before Claude Code, Anthropic powered much of the AI coding stack indirectly. By March 2025, about 80% of Anthropic revenue came from API usage, with Claude serving as the default model inside tools like Cursor, Bolt.new, and Windsurf. Claude Code was Anthropic moving up from model supplier to owned coding surface.
  1 sacra 2 sacra
• The product shift mattered because the interface changed. Cursor lives inside a VS Code based IDE, while Claude Code was built as a terminal agent that can autonomously edit, test, and debug code. That pushed the market from autocomplete and chat in the editor toward longer running agent workflows across the full repo.
  1 sacra 3 sacra
• This was the same play OpenAI later pursued with Windsurf. The reason is simple. Owning the coding client means owning distribution, subscription revenue, and the stream of code edit feedback that helps train better models. In this market, the app is also a data collection machine for the model layer underneath.
  3 sacra 4 sacra

From here, coding tools keep collapsing into model companies and model companies keep climbing into products. The winners are likely to be the ones that control both the model and the developer surface, whether that surface is the IDE, the terminal, or an agent that spans both.