A podcast on what engineers verify when agents write code
In this October 21 episode, Jessica Kerr and Avdi Grimm discuss how developers can define correctness and give coding agents feedback loops. They describe specialized reviewer agents, repeated audits and using evaluations to compare approaches. Their shared point is that software checks can measure some outcomes, but humans still need to decide what to build and whether it serves the wider product.
The Ruby on Rails Podcast · Oct 21, 2026
Why it matters
The conversation connects day-to-day agent review practices with the product judgment that tests cannot supply on their own.
More context
The 46-minute episode includes practitioner accounts of testing a demo Kubernetes upgrade, reviewing changes with separate agents and revising codebase guidance when old patterns slow agents down. Its transcript makes it possible to read those sections without listening to the full interview.
Keep in mind
These are the speakers' experiences and opinions, not a controlled comparison of agent workflows. The transcript is machine-generated and has transcription errors; interpret individual wording with care.
Transcript read