Cursor 3 Launches Agent-First Unified Workspace: The IDE Category Is Fragmenting Into Something Else

Anysphere shipped Cursor 3 (codenamed "Glass") on April 2, and it's not really an IDE anymore. It's an agent workspace.

The redesign is fundamental. Developers enter natural language task descriptions, select an LLM, and Cursor generates code plus a demo video. Local and cloud agents run in parallel and appear in a unified sidebar. The part that caught my attention: you can access it from mobile, web, desktop, Slack, GitHub, and Linear. The IDE broke out of the desktop.

This is a direct response to Claude Code and OpenAI Codex eating into Cursor's developer base. Both of those are terminal-first, agent-first tools that don't need a traditional editor. Cursor's answer is to abandon the traditional editor paradigm entirely and meet them on agent territory. Smart move, but risky. They're betting the entire company on this direction while facing pressure from subsidized competitor pricing.

I use Claude Code daily. It's terminal-based, keyboard-driven, and fits my workflow. Cursor 3 is betting that most developers want the opposite: a visual workspace where you describe what you want and agents figure out the how. I'm not sure which bet wins long-term, but I think they're both right for different types of work. Complex refactoring across a large codebase? I want Claude Code in my terminal. Prototyping a new feature with quick iteration? A visual agent workspace might be faster.

The bigger signal here is category fragmentation. "IDE" used to mean one thing. Now we have terminal agents (Claude Code), agent workspaces (Cursor 3), browser-as-IDE tools (Stagewise, YC S25, which turns the browser into a development environment where the coding agent has native console and debugger access), and cloud agent platforms (Codex). These aren't variations on the same product. They're different products solving different slices of the development workflow. The IDE as a monolithic category is splitting apart.

For builders, the actionable takeaway: don't commit to one tool. The category is moving too fast. Use Claude Code for deep work, Cursor for visual prototyping, and keep an eye on the browser-as-IDE pattern. The tool that wins will be the one that matches how you actually think, not the one with the most features.