New debug mode in Cursor 2.2

Cursor version 2.2 introduced a debug mode that takes a different approach than typical AI assistants.¹ Instead of immediately generating code when encountering an error, the agent first formulates multiple hypotheses about the cause and then automatically instruments the code with logging.

The workflow is interactive. I describe the bug, the agent suggests causes and adds logging. After reproducing, it analyzes the runtime data and generates a fix. Human-in-the-loop for verification, systematic analysis by the agent. Whether this works as smoothly in practice as in the marketing video remains to be tested.