Software Engineering Radio - the podcast for professional software developers SE Radio 693: Mark Williamson on AI-Assisted Debugging
Nov 6, 2025
Mark Williamson, CTO of Undo and kernel-level debugging expert, talks AI-assisted debugging and time-travel debugging. He explores how AI can sift vast data, act as a smart search and rubber duck, generate tests and automate workflows. He examines LLMs hooked to debuggers like ChatDBG, challenges in kernel/C++ and distributed systems, and risks such as hallucinations and the need for guardrails.
AI Snips
Chapters
Transcript
Episode notes
Debugging Is Core Development Work
- Debugging is the ongoing quest to understand what a program does and why it differs from expectations.
- Mark Williamson says most developer time is spent debugging, not writing fresh code.
LLMs Excel At Sifting And Automation
- LLMs help debugging mainly by sifting large data and automating tedious tasks.
- Williamson frames LLM value as finding nuggets and removing toil, not magic bug fixes.
Use LLMs As Better Semantic Search
- Use LLMs as a semantic search over code, issues, and error messages to find relevant context quickly.
- Feed the LLM high-quality context so it can map your meaning to relevant examples.