Anysphere released Cursor 3 with a complete interface redesign that treats developers as agent orchestrators rather than code writers. The new "Agents Window" lets users run multiple AI agents simultaneously across repositories, with local-to-cloud handoff capabilities and a plugin marketplace for extending agent functionality. Internal data shows the usage shift is real: Cursor's own engineering team now has 35% of merged pull requests written by autonomous cloud agents, and user patterns have flipped from 2.5x more tab completion users to 2x more agent users since March 2025.
This represents more than a UI change — it's a bet on what software development becomes when AI agents can handle entire coding workflows autonomously. I've covered Cursor's aggressive agent push since their $3B+ funding round, and this interface makes their vision concrete: developers managing fleets of AI workers rather than writing code directly. The local-to-cloud handoff addresses a real pain point I've seen teams struggle with — keeping agents running when developers go offline or context-switch between tasks.
The community reaction tells the real story. Reddit and Hacker News threads show sharp division, with many developers questioning whether Cursor is abandoning the IDE-first identity that made it popular. Some are calling it overengineering, others see it as inevitable evolution. The ability to switch back to the full IDE suggests Anysphere knows this transition won't be smooth for everyone. Meanwhile, the new "canvases" feature in version 3.1 lets agents create interactive dashboards and visualizations — pushing further into territory that feels more like agent collaboration than traditional coding.
For teams evaluating AI coding tools, Cursor 3 forces a decision about workflow philosophy. If your developers want enhanced IDEs with smart autocomplete, this might feel like unnecessary complexity. But if you're already running agents for substantial coding tasks, the parallel execution and handoff capabilities solve real operational problems. The plugin marketplace could be the deciding factor — teams that build custom agent workflows will find value, while others might stick with simpler tools." "tags": ["cursor", "agents", "coding-tools", "developer-workflow