Rebuilt around agents, not files
Anysphere rebuilt its editor from scratch rather than continuing the VS Code fork it had used since 2023, describing the result as a unified workspace for building software with agents. The Agents window collects every local and cloud agent into one sidebar, including sessions started from mobile, web, desktop, Slack, GitHub and Linear. [1]
The rebuilt interface keeps most of what a conventional IDE offers: work across multiple repositories at once, a simplified diff view for reviewing changes, an integrated browser for opening local sites, full language server support with go to definition, and staging and management of pull requests. A developer can drop back into ordinary line by line editing at any point rather than only supervising agents. [1]
Agents that keep running after the laptop closes
A session can start locally and keep running in the cloud if the laptop closes, then come back to the desktop for testing against Composer 2, Cursor's own coding model. Cloud agents generate their own demos and screenshots so a developer can check what changed without pulling every branch locally. [1]
A marketplace bundled with the release adds hundreds of plugins that extend agents through MCP connections, skills and subagents, treating the editor itself as a place agents run rather than only a place code is written. [1]
A split reaction on Hacker News
The Hacker News discussion of the announcement split quickly over the parallel agent workflow the Agents window is built for. One commenter, who had been running Claude Code inside Cursor on one task at a time, wrote: "I have zero interest in these new \"swarms of agents\" they are trying to force on us from every direction. I can barely keep straight my code working on one feature at a time... I'll likely drop Cursor for good now and switch back to vanilla VsCode with CC." Another pushed back on the word itself: "I think you misunderstand 'swarms of agents' ... It is working on one feature with multiple agents taking different roles on that task," distinguishing that from agents each working a separate task at once. [2]
Atlas interpretation: The disagreement was less about whether agents work than about what task decomposition people trust. Commenters who ran several unrelated agents at once in a single git tree treated context switching, not code quality, as the real cost of the workflow. Commenters who rejected running more than one agent at a time gave the opposite reason: holding the full context in their own head while reviewing is how they catch mistakes, and that does not scale across parallel sessions. [2]
Sources
- Meet the new Cursor
Cursor · Apr 2, 2026
- Cursor 3
Hacker News · Apr 2, 2026