Coding Agents Lunch & Learn Session 25: Verification and Recovery

Join us for Session 25 of the Coding Agents Lunch & Learn, our weekly community gathering for anyone interested in coding agents, agentic AI, AI-powered developer tools, and the rapidly evolving world of software development.

This week, we’re joined by Tanmay Sah and Dolly Sah for a special discussion on verification and recovery in coding agents.

As coding agents become increasingly capable of making changes autonomously, two important questions emerge: How do we verify that an agent’s changes are safe and correct? And what happens when verification fails and an agent has already made a harmful or incorrect change?

Tanmay and Dolly will bring together insights from two of their recent research papers:

1. The Verifier Tax - exploring the tradeoffs involved in verifying agent behavior and the relationship between verification, safety, and successful task completion.
2. EvoUndo - exploring how agents can recover from harmful or incorrect changes after things have already gone wrong.

Together, the papers tell a broader story about building coding agents that can not only act, but also check their work and recover when they fail.

This will be a loose, conversational session rather than a tightly structured presentation. We’ll hear from Tanmay and Dolly about their research, the practical lessons for developers and agent builders, and what these ideas mean for the future of reliable coding agents.

We’ll then open things up for questions, discussion, and community conversation around agent verification, safety, recovery, and real-world failure modes.

Whether you’re building coding agents, experimenting with AI developer tools, researching agentic systems, or simply curious about where the space is heading, come join us to learn from researchers, ask questions, and hang out with the community.

Bring your lunch, bring your questions, and come hang out with us!