Cursor launched Cursor 3 Thursday, a new agent-first coding interface that lets developers spin up AI agents to complete entire tasks without writing code. The product, developed under the codename Glass, centers around a chatbot-like text box where users describe tasks in natural language, with a sidebar to manage multiple running agents. Unlike standalone products from OpenAI and Anthropic, Cursor 3 integrates directly with the company's existing AI-powered IDE.
This move signals how dramatically the AI coding landscape has shifted in 18 months. Cursor pioneered AI-assisted coding and became one of the biggest customers of OpenAI, Anthropic, and Google's models. But those same providers now compete directly with highly subsidized agent products like Claude Code and Codex that have attracted millions of developers. As Cursor's Jonas Nelle told WIRED, "our profession has completely changed" and much of what built their success "is not as important going forward anymore."
The competitive pressure is real. OpenAI and Anthropic can afford to offer coding agents at below-market rates because they're selling their own models, while Cursor pays retail prices for the same underlying technology. This puts Cursor in the uncomfortable position of competing with their own suppliers, who have deeper pockets and can subsidize user acquisition.
For developers, this competition benefits everyone. Agent-first coding is becoming the standard, and having multiple polished options drives innovation. But Cursor's challenge highlights a broader issue: when AI labs control both the models and increasingly the applications, independent tool makers face an uphill battle unless they can differentiate significantly on user experience or specialized workflows.