Anysphere introduced Cursor 3 — an AI code editor with local and cloud agents

Anysphere presented Cursor 3 on April 2nd — a new version of its AI tool for programming. The company is restructuring the product around an agent mode: now Cursor should not only suggest code, but also coordinate multiple assistants that write, review, and refine changes.

New Cursor Interface

The main change in Cursor 3 is a new interface, built from scratch around AI agents rather than a classical editor window. A developer can describe the required function in natural language, select one of the supported models, including Claude, and get not only generated code but also a demonstration of how this code works. In essence, Cursor is increasingly shifting from the role of intelligent autocomplete to the role of a work panel where a person manages a whole set of automated executors.

"Cursor 3 is a unified workspace for creating software with agents."

Inside the new window are all active assistants: local, cloud-based, and those launched from other channels. Each agent has its own working context, and the user sees a general list of tasks in the sidebar. This is important for teams and developers who maintain multiple repositories simultaneously: Cursor 3 is designed specifically for parallel work, where one agent writes code, a second prepares edits, and a third shows results through screenshots and demos.

Local and Cloud Agents

Anysphere divided the work between two types of agents. Cloud agents receive more computational resources, so they are better suited for heavy and long-running tasks: for example, generating large code fragments or processing multiple requests in parallel. Local agents are slower, but they have a different advantage: the developer can immediately open the result on their computer, manually make changes, run tests, and quickly verify how everything behaves in a real environment.

An important detail of the release is the fast transfer of a session between cloud and local machine. You can start a task in the cloud, then transfer it to the desktop for fine-tuning. Or vice versa: send a local session to the cloud so the agent can continue working while the user is offline or switching to another task.

For such scenarios, Anysphere separately highlights Composer 2 — its own model for programming, which the company calls more economical for such workflows.

What Appeared in the Release

Cursor 3 adds not just one major feature, but several layers of convenience around agent-based development. Some are aimed at code generation and editing, others at quality control and accelerating the everyday cycle from request to ready change. In addition to the agents themselves, Anysphere updated the diff viewing interfaces, edit review, and task navigation so that developers can more easily not only run AI but also accept its work.

• chat interface where a task can be described in natural language
• switching between local and cloud agents without breaking the session
• design mode for quick editing of interface elements using text instructions
• step-by-step review of agent actions with explanations, errors, and screenshots
• faster viewing of changes before commit and rollout to production

Separately, Anysphere improved the response selection scenario: now the developer can send a request to multiple large language models at once and choose the best result. This reduces dependence on a single model and makes Cursor more of a shell for working with different AI than just another editor with an embedded chatbot. In practice, this approach is useful where you need to quickly compare implementation options rather than blindly accept the first generated answer.

What This Means

Cursor 3 demonstrates how rapidly the market for AI tools for programmers is moving away from the "suggestion in editor" format to the "agent orchestrator" format. Anysphere is betting that developers will increasingly not write every line themselves, but distribute tasks between local and cloud assistants, checking intermediate results. If this scenario takes hold, the next field of competition will no longer be individual models, but the convenience of managing an entire agent team.