Cursor 3 Reimagines Coding as Agent Management With $3B War Chest

Anysphere, the startup behind the widely used AI code editor Cursor, unveiled Cursor 3 on April 2 — a radical reimagining of its product that abandons the traditional code-editing paradigm in favor of what the company calls an "agent-first unified workspace for building software." Backed by more than $3 billion in funding from Nvidia, Google, and other major investors, the release represents one of the most ambitious bets yet on the future of AI-powered software development, and a direct challenge to Anthropic's Claude Code, which currently commands roughly 54% of the AI coding market according to data from Menlo Ventures.

Developed internally under the codename "Glass," Cursor 3 is built from the ground up to let developers manage fleets of autonomous AI agents that handle feature implementation, testing, and deployment. Rather than writing code line by line with AI assistance, programmers effectively become engineering managers, delegating tasks to agents using natural language. The platform supports a multi-workspace architecture enabling parallel execution of both local and cloud-based agents across multiple repositories simultaneously — cloud agents tap into greater hardware resources for speed, while desktop agents allow for local code editing and testing.

Among the most notable additions is a Design Mode that lets users select UI elements and describe desired changes in plain English, with agents handling implementation automatically. Cursor 3 also ships with a new Marketplace featuring hundreds of plugins that support Model Context Protocols and subagents, while agents can now be triggered from mobile, web, desktop, Slack, GitHub, and Linear — extending the development workflow well beyond the traditional IDE.

Powering much of the experience is Composer 2, Anysphere's proprietary frontier coding model. However, the launch has not been without controversy: it was recently revealed that Composer 2 is largely based on Moonshot AI's open-source Kimi 2.5 model, a fact that was not initially disclosed to users and has raised transparency concerns within the developer community. Cursor 3 also supports third-party large language models, including Anthropic's Claude family, and introduces a multi-LLM shortcut that sends the same request to multiple models simultaneously, allowing developers to compare and select the best response.

The competitive landscape Cursor 3 enters is fiercely contested. Anthropic's Claude Code has established a commanding lead with its 54% market share, while OpenAI's Codex 5.3 continues to iterate on its own agent-driven approach. Co-founders Michael Truell and Sualeh Asif framed Cursor 3 as a response to increasingly powerful coding models that "unlock new interaction patterns" — a signal that the industry is shifting from AI-assisted coding, where humans write code with AI suggestions, to AI-delegated coding, where humans orchestrate agents that write the code themselves.

The scale of Anysphere's fundraising underscores how high the stakes have become in the AI development tools market. With over $3 billion raised, the company has more capital at its disposal than many publicly traded software firms, and investors are clearly betting that the IDE of the future looks less like a text editor and more like a command center. Whether Cursor 3 can convert that capital into market share against deeply entrenched competitors remains the central question.

For the broader software industry, Cursor 3's launch accelerates a transformation that began with GitHub Copilot's arrival in 2021 but is now entering a fundamentally different phase. The notion that a single developer could manage a fleet of AI agents building features in parallel across multiple codebases would have seemed speculative even a year ago. With billions of dollars and multiple frontier AI companies now competing to make that vision real, the role of the software developer — and the tools that define it — is being rewritten in real time.