Cursor receives 2.2 update with new Debug Mode and more

Cursor is rolling out version 2.2 of its AI code editor, putting a new Debug Mode, upgraded Plan Mode, multi agent judging and pinned chats directly into the daily workflow of its users. The release lands on December 10, 2025, as a standard product update for the desktop apps.

At the center of the update is Debug Mode, a fresh agent loop built around runtime logs and human verification. Developers select the mode, describe a stubborn bug, and the agent scans the codebase to propose multiple hypotheses. It then auto instruments the code with logging, waits for the user to reproduce the issue, and uses the captured variables, execution paths and timing data to propose a minimal, targeted fix, before asking the user to verify that the bug is truly gone and stripping the logging back out.

Plan Mode gains inline Mermaid diagrams, letting agents generate and stream architecture or flow visuals into plans, while a new option sends selected to dos into fresh agents for parallel execution. For heavier workflows, multi agent judging now reviews parallel agent runs, recommends a best solution and leaves a short rationale in a comment once all runs complete. Pinned chats keep priority agent threads locked to the top of the sidebar so long running refactors and investigations stay within one click.

Behind these changes is Cursor, an AI powered IDE from Anysphere that started as a fork of Visual Studio Code and now runs on Windows, macOS and Linux with deep model integration from several providers. Its pitch is clear: turn AI agents into first class participants in coding, covering autocomplete, codebase search, refactors and now structured debugging loops that mirror how experienced engineers actually hunt down heisenbugs.

Source