San Diego C++ Meetup #84: March 2026 Edition – UNDO – Agentic Debugging Using Time Travel

Hello everyone!

Last Tuesday, March 10th, we held our 84th San Diego C++ Meetup. This was a special “super-fast” edition where we broke our usual virtual routine to meet in person at Qualcomm. It was a packed house with great pizza and even better technical discourse.

Our New Home on Luma

As mentioned in our previous sessions, we have officially moved away from the Meetup.com platform. To stay updated with our future events and sync them to your calendar, please follow our new dedicated page.

Join our community calendar here: https://luma.com/sdcppmu

At this time, we are no longer seeking additional sponsorships, but we are excited to focus on growing our community through this new platform.

The Main Event: Agentic Debugging Using Time Travel

We were honored to host Dr. Greg Law, co-founder and CEO of Undo, who flew in from England to share his insights on “Agentic Debugging.” Greg is a programmer at heart with over 25 years of experience in the software industry, and his talk focused on how time travel debugging is evolving alongside AI.

The event page on Luma: https://luma.com/8lwifeef

Why Time Travel Debugging?

Greg opened with a classic Brian Kernighan quote: “Debugging is twice as hard as writing the program in the first place.” If you are as clever as you can be when you write it, how will you ever debug it?

Traditional debuggers tell you what is happening now, but they don’t tell you what happened in the past. Time travel debugging treats a program’s execution like a “core file for every instruction,” allowing you to reason backward from a failure to the root cause.

Key Highlights from the Talk:

• Doom as a Debugging Sandbox: Greg demonstrated the power of time travel by using a recording of Chocolate Doom. He showed how to find the exact line of code where a zombie was killed simply by tracking the “last” time a specific pixel value changed on the screen and following the data back to the health decrement.
• The implementation of “Magic”: We went under the hood to see how Undo works. Unlike naive state-saving, it uses a JIT (Binary Translation) to record only non-deterministic events (system calls, thread switches, signals, and shared memory accesses). This allows for high-performance recording that can be replayed deterministically.
• The AI Revolution (Agentic Debugging): The most exciting part of the talk was seeing how Large Language Models (LLMs) like Claude can be given “tools” to drive the debugger. Greg showed an agent diagnosing a subtle reference-counting bug in CPython that had previously taken human maintainers weeks to solve.
• Backwards Reasoning: AI agents are remarkably good at tool-use when constrained. By allowing an agent to “reverse-step” and “trace” values through a time-travel recording, we can move from manually hunting bugs to simply asking the agent to “find where this door opened” or “explain why this pointer is null.”

Recording

If you couldn’t make it to Qualcomm or want to re-watch Greg’s live demos of the AI agent in action, the full recording is available below:

Special Thanks

This meetup was a collaborative effort, and I want to extend several thank yous:

First, a huge thanks to the UNDO team for coming to San Diego. Specifically, thank you to Greg Law for the fantastic presentation and Tony for his collaboration and help in making this session a reality.

I also want to thank Qualcomm and Brian for orchestrating the venue and hosting us.

Finally, thank you to our long-term supporters. We thank Packt Publishing for their ongoing support and technical resources, and JetBrains for their generous sponsorship over the past year which helped us keep the community thriving.

Closing Thoughts

It was a treat to see so many of you in person. The intersection of AI and low-level C++ debugging is a fascinating frontier, and Greg’s talk gave us a lot to think about regarding the future of our workflow.

Don’t forget to register on Luma to stay in the loop for our April meeting.

See you at #85!

Kobi