Debug Mode is a new feature designed to enhance the debugging capabilities of coding agents. It incorporates runtime information and a human verification process to tackle bugs that often stump even advanced AI models. By analyzing the workflows of top debuggers, the team created a system that generates multiple hypotheses about potential issues in the code, rather than jumping straight to a fix. To start using Debug Mode, you select it from a dropdown menu and describe the bug in detail. The agent then scans your codebase, creates hypotheses, and adds logging statements to test these ideas. This step allows the agent to collect data on variable states, execution paths, and timing when the bug occurs. With this information, it can identify the root cause and propose a targeted solution, often requiring just a few lines of code instead of extensive rewrites. After the proposed fix is in place, you need to reproduce the bug again. If the issue is resolved, you mark it fixed, and the agent cleans up the added instrumentation. If the bug persists, the agent refines its logging and hypotheses, working alongside you until the problem is genuinely addressed. This interactive process emphasizes human judgment in the final verification, ensuring that the fix is not only functional but also appropriate for the context.