Cursor Launches Cursor 3 to Let Developers Manage Teams of AI Agents

San Francisco: Cursor has launched Cursor 3, a rebuilt coding interface designed around multi-agent management. The release is a direct response to mounting pressure from Anthropic’s Claude Code and OpenAI’s Codex, two tools rapidly capturing market share that Cursor once dominated.

Cursor 3 is not a new model. It is a new workspace. The company describes it as a “unified workspace for building software with agents.” Developers can spin up multiple local and cloud-based agents simultaneously. All agents appear in a single sidebar including those triggered from mobile, Slack, GitHub, and Linear.

The interface is built for multi-repo work. Developers can manage agents running across different codebases from one view. Cloud agents produce demo videos and screenshots of their work for human review. A developer can shift an agent session from cloud to local for editing and testing in a single step.

Design Mode is a notable addition. Users can select interface elements and describe changes in plain language. Agents implement the modifications automatically. A new /best-of-n command lets developers send a request to multiple LLMs and compare outputs before choosing the best result.

The competitive context is difficult to ignore. Claude Code now holds approximately 54% of the AI coding market, according to Menlo Ventures data. OpenAI has begun offering unlimited Codex access to pull in users.

Cursor launched Composer 2 last month and immediately faced criticism when the community discovered it relied on the Chinese open-source model Kimi K2.5, something Cursor had not disclosed upfront.

Cursor developed Cursor 3 under the codename Glass. Jonas Nelle, Cursor’s Head of Engineering, described the shift clearly: the profession has changed dramatically in just the past few months. Cursor is now betting that the next phase will be led by those who can manage agents at scale not those who simply write the most code.