Cursor 3 launched with a new Agents Window — a standalone agentic chat interface that closely resembles Claude Code. A hands-on comparison tested both tools on two real bugs from the HTTPie open-source repo: one with a suggested fix, one without. Both tools resolved both bugs without additional prompting. Claude Code ran natively in the terminal, asked for permission before making changes, and was notably fast (54 seconds on the harder bug). Cursor's interface felt more familiar and IDE-like but couldn't run the test suite itself. The conclusion: the gap between the two tools has narrowed significantly, and the choice now comes down to personal preference and workflow style.
Sort: