Anysphere has released Cursor 3, a major update to its AI code editor that reorganizes the product around autonomous coding agents rather than the traditional file-and-editor layout. The company describes the release as a new interface built from scratch, in which developers assign tasks to AI agents and review their work instead of writing most code by hand.
The update is available to existing Cursor subscribers at no extra charge. Cursor has become one of the most widely used AI development tools, and Anysphere has been valued in the tens of billions of dollars in recent funding rounds, placing it in direct competition with coding products from OpenAI and Anthropic as well as GitHub’s Copilot.
Agents Window
The centerpiece of Cursor 3 is the Agents Window, a view that lets a developer run several AI agents in parallel, each working on a separate task such as implementing a feature, fixing a bug or writing tests. Each agent operates in its own isolated copy of the codebase so that changes do not collide, and the developer can inspect, accept or discard each agent’s output. Coverage of the launch describes this as the shift from an editor with AI assistance to a workspace for supervising AI workers.
Agents can run locally on the developer’s machine or in the cloud, and Cursor 3 allows tasks to be handed between the two. A developer can start a job in a cloud agent from a laptop or phone, then pull the result into a local environment to finish it, or push a local task to the cloud to continue running in the background.
Design Mode for front-end work
For user interface development, Cursor 3 introduces Design Mode. Rather than describing a visual change in text, a developer can click on an element in the running application, such as a button or a layout section, and tell the agent what to change. The agent then locates and edits the relevant code. Anysphere positions the feature as a way to shorten the loop between seeing a problem on screen and fixing it, particularly for designers and developers working on front-end code.
A crowded market
The release arrives as AI coding tools move quickly toward agent-based workflows. Anthropic’s Claude Code, OpenAI’s Codex and GitHub’s Copilot have all added features for delegating multi-step tasks to agents, and the competition is increasingly about how well a product manages many agents at once rather than how good its autocomplete is. Cursor 3 is Anysphere’s answer to that shift, and early reviews from developers have focused on whether the parallel-agent model saves time in practice or simply moves the work from writing code to reviewing it.
