Cursor, a San Francisco based AI startup, launched Cursor 3 on Tuesday, a unified development workspace that delegates tasks to autonomous AI agents running in parallel. The new environment shifts code work from manual editing to managed processes, giving developers a single control surface for multiple agents working at once.

Cursor 3 reframes how software teams coordinate AI assistance. By consolidating local and cloud agents into a single side panel, developers can start, stop, or move agents from a mobile app, the web interface, Slack, GitHub, or Linear without leaving their workflow. It's a logistical shift, turning code assistance from a tool into a persistent infrastructure layer.

Company engineers point to seamless task handoff as the core productivity boost. Sessions can shift from a cloud instance to a local machine for rapid testing, or remain in the cloud to continue work while the user's device is offline. The diff view added to the interface lets users review changes, stage files, create commits, and open pull requests without switching tools. Code review becomes a native step in the agent workflow.

Cursor enters a crowded field where Anthropic's Claude and OpenAI's Codex models anchor much of the market. According to Menlo Ventures, Anthropic's Claude Code tool has captured up to 54% of the market, while OpenAI has expanded unlimited access to its coding models. Cursor's earlier Composer 2 release drew scrutiny for its reliance on Moonshot AI's open source Kimi 2.5 network, a controversy the company has yet to fully address in public documentation.

Cursor plans to extend agent orchestration to more IDEs and deepen integration with CI/CD pipelines. Future updates will let teams define custom agent roles and automate larger portions of the software lifecycle. The aim is positioning AI not as an occasional tool, but as a permanent teammate, one that operates in the background, adapts to context, and scales with the scope of the project.