Software Engineering Radio - the podcast for professional software developers SE Radio 693: Mark Williamson on AI-Assisted Debugging
15 snips
Nov 6, 2025 
Mark Williamson, CTO of Undo and an expert in AI-assisted debugging, dives into the transformative role of AI in debugging processes. He highlights AI's ability to analyze data, automate tedious tasks, and enhance debugging strategies. The discussion covers advanced tools like ChatDBG and time-travel debugging, which allow developers to rewind execution for deeper insights. Mark also emphasizes the benefits for developers of all experience levels and warns of potential risks, including hallucinations and security issues, stressing the importance of human oversight.
AI Snips
Chapters
Transcript
Episode notes
AI As A Smarter Rubber Duck
- LLMs make excellent 'rubber ducks' that suggest strategies and tools you might not consider.
- Williamson highlights preserving developer flow by letting agents switch tools without breaking concentration.
Be Cautious With Kernel-Level Agents
- Use LLMs for kernel-level work where documentation and mailing-list context exist, but add careful guardrails.
- Williamson says hooking an LLM to a kernel debugger could be powerful but is complex and risky.
Compilers Help Ground LLMs
- Compiled languages give LLMs guardrails via compilers, reducing hallucination in generated C/C++.
- But C/C++ projects often contain legacy patterns and memory bugs that remain challenging for AI.