Roundup
11 episodes on how AI agents get built, shipped, and evaluated. 6 are from Chain of Thought, the show I host. 5 are from other shows, because no single podcast has covered all of this well.
This is an episode list, not a show list. If the question is which podcast to subscribe to, I benchmarked eleven AI engineering podcasts on my own site. This page answers the narrower one: you have forty minutes, and you want to understand agents better than you did this morning.
Topic first. Each episode had to be about agents as a working system: how one gets built, and what breaks once it is in front of customers. Episodes that mention agents in passing did not qualify, however good they are.
Nobody paid to be on this page, and no show had a say in what this page says about it. Where money or a business arrangement connects me to a show, that show's entry says so plainly. Every description is written from scratch: no artwork, no transcript text, nothing lifted from anyone's feed.
The show I host. Each of these has its full transcript on the episode page.
Episodes I would send someone to that are not mine. Titles, hosts, and dates are as published on each show's own page.
Luca Rossi writes Refactoring for engineering managers, and the angle follows from that: how a team changes shape once agents are doing part of the work. His guest, Charlie Guo, works on developer experience at OpenAI, so the examples come from inside a team that ships coding agents rather than from a keynote about them.
Disclosure: Luca is a frequent collaborator over the years, and we've had business relationships past and present.
Slack's chief product officer on putting agents where the conversation already happens instead of behind one more tab, and on what an open standard like MCP is for once a company runs more agents than any one person can track. I built and hosted this show at LinearB before Chain of Thought existed; Andrew Zigler runs it now. Still the best place I know on the discovery problem, which is how anyone finds the right agent at the moment they need it.
Claire Vo runs a screen share rather than an interview, and this one is her pointing a coding agent at browser and computer control across five concrete tasks. That makes it the least abstract thing on this page. She was a guest here on ep 36. If you have read about computer use and want to watch someone do it end to end in under half an hour, start with this.
The layer underneath everything else on this page. Akshat Bubna is Modal's CTO, and the conversation is about what a cloud has to change when its customer is an agent rather than a person: sandboxes to run untrusted code in, and GPU snapshotting so a machine resumes instead of cold-starting every time. Nothing else here covers the compute an agent runs on.
Episode 48 above is an enterprise rollout. This is the consumer-scale version of the same problem: Akshay Nathan leads core product engineering at OpenAI and walks through Codex and ChatGPT Work reaching ten million users in two weeks, including the part nobody planned for, which is a coding agent getting pointed at finance and marketing work by people who do not write code.
Came here from one show in particular? There is a listening guide for people who like Latent Space, pairing its episodes with ours.
Everything on AI agents → Collections → New here? Start with five →