Podcast Episodes About AI Agents Every PM Should Hear

If you want an AI agents podcast that actually teaches you how to build and ship agents, start with Lenny's Podcast, "How we built Grok Bot in a month" with Roman Ugarte (8 September 2026, about 83 minutes), then How I AI's "Grok Bot vs. OpenClaw" with Claire Vo and The Growth Podcast's product loops episode with Tyler Folkman. AllthingsPM summarizes all 13 episodes in this roundup free, with no sign-in, so you can read the lessons from roughly ten hours of listening in a fraction of the time.

AllthingsPM is an AI PM course and PM interview prep platform. Below is each episode, the one idea to keep from it, and how to turn it into something you can show in an AI PM interview.

Which AI agents podcast episodes should a PM hear first?

These are the 13 agent episodes in our podcast summaries library that a product manager gets the most from. Lengths come from each show's own feed or page, checked 28 September 2026.

Lengths from each episode's feed or page, checked 28 September 2026. Total: 606 minutes of audio.

If you only have two hours, listen to the Lenny's Podcast and How I AI Grok Bot episodes back to back. One is the builder explaining the product decisions; the other is a heavy user explaining which of those decisions made her switch. Read the other eleven as summaries.

What do the builder episodes teach about shipping an agent product?

Lenny's Podcast, Roman Ugarte on Grok Bot. Ugarte led growth at Cursor before helping incubate Grok Bot, which went from first line of code to internal beta in about a month and to public launch roughly three weeks later. The team credits two decisions: agents live entirely in the cloud, and each agent gets its own computer rather than sharing the user's. They also onboarded 200 to 300 early users by hand, and only built "chief of staff" style delegation after watching users invent it themselves. Read the Grok Bot summary.

How I AI, Claire Vo on Grok Bot vs. OpenClaw. Vo runs about 30 personal agents. She says OpenClaw was more proactive and supported multiplayer, and she left anyway because keeping it alive took too much upkeep. Her rules are simple product lessons: one bot, one name, one job; and ask for permissions just in time, at the moment a bot needs a connector. Read the OpenClaw teardown.

How I AI, Claire Vo on Muse. Vo calls Meta's Muse possibly the best designed agent she has tested, not the most powerful. The lesson is translation: tasks, permissions and tool calls become ideas, goals and a library for parents who will never read a stack trace. Read the Muse review.

Put together, these three episodes make one argument: the agent that wins is the one people can delegate to without babysitting it.

How AllthingsPM does this. Our Agents and agentic architecture chapter turns this argument into a skill. It starts with workflow or agent, the question every one of these builders answered first, and ends with an integration case where you design the agent version of a feature, with its architecture and tool contracts.

Which episodes explain loops, and why do PMs keep hearing about them?

"Loops" was the agent word of the summer, and two episodes explain it better than anything else we have summarized.

The Growth Podcast, Tyler Folkman on product loops in Claude Code. Folkman, Chief AI Officer and Head of Product at JobNimbus, builds loops live. His definition: a skill that runs the same way forever is just a skill; it becomes a loop when it reads the log of what happened and rewrites itself. Every loop has the same shape: fetch inputs, do the work, pass a gate, write an artifact, learn. He says the gate "is almost always the most important part," and ideally deterministic. When the real gate is the customer, his team builds a synthetic one from existing research transcripts to cut 100 ideas to the best 5. Read the product loops summary.

The AI Daily Brief, Nufar Gaspar on agentic loops for knowledge workers. This is the non-engineer version. Every agent tool already runs a loop (plan, act, check, adjust), but the built-in one stops too early. Coders get a free referee because code compiles or it does not; knowledge workers have to manufacture one. The core skill is writing a "boring, checkable" done condition. Read the agentic loops summary.

If you listen to one, pick Folkman for the hands-on build and Gaspar if you want to apply loops to PM work that has no test suite.

How AllthingsPM does this. The PM as builder chapter has you build and iterate in Claude Code, Cursor and Codex instead of just describing what you want. The course's lesson on self-improving agents and the data flywheel covers the same "learn from the log" step Folkman calls the difference between a skill and a loop.

How are working PMs using agents day to day?

Two episodes show real PMs running agents inside their own jobs.

How I AI, Daniel Blum (PM at Melio). Blum built a Claude co-work system that runs 70 to 80% of his day inside normal corporate limits: fixed tools, capped tokens, a budget. Two rules make it self-improving: it can rewrite its own instruction files, and it connects to as much of his work as possible (Slack, email, calendar, docs). He front-loaded a context file per topic, goal and colleague, and treats his Notion board as almost read-only. Read the Daniel Blum summary.

The Growth Podcast, Mikhail Shcheglov (CPO at OLX Classifieds). Shcheglov spent five months building one agentic operating system his product org runs on. It handles email and calendar, screens stakeholder feature requests before they reach a PM, and maintains the design system. He tracks how much company context the agent holds as a personal KPI; at 54% coverage, he says, it already makes backlog-level calls like a junior to mid PM. Read the company OS summary.

The common thread is context. Neither setup works because of a clever prompt. Both work because someone did the dull job of feeding the agent what the business knows.

How AllthingsPM does this. The AI PM knowledge graph maps the AI-specific concepts behind agents and links each one to the lessons that teach it. If you are also reading around the topic, our post on AI agents vs workflows is a shorter primer on when an agent is even the right answer.

What happens when companies run agents at scale?

How I AI, Sharadh Krishnamurthy on Stripe's Kai. Kai reaches more than 10,000 Stripe employees a week. Krishnamurthy's conclusion is that "get AI to everyone" was a governance problem, not a model access problem: employees should not make security, tool and model choices from scratch every session. Projects, reusable skills and hardened infrastructure did the work. Read the Stripe Kai summary.

How I AI, Zach Lloyd on Warp's software factories. Warp's factory takes a request from Slack through triage, implementation, review and a merged pull request. The number that matters: 35 minutes from kickoff to pull request, then three and a half hours until the first human review. Measuring the whole pipeline showed the humans are now the bottleneck. Read the Warp summary.

Masters of Scale, Jess Hertz on agentic shopping at Shopify. Shopify's COO shares that last quarter 75% of its AI-attributed orders came from categories outside the top 100, and that an agent buying through Shopify's structured product data converted at twice the rate of one using scraped web data. Internally, a Slack agent took 15 million actions in five months. Read the Shopify summary.

The AI Daily Brief, the multiplayer AI sprint. NLW argues agents have only solved the individual half of knowledge work, citing surveys that put 42 to 60% of the workday in collaborative work, and lays out a four week plan for a team's first shared agent. Read the multiplayer summary.

How AllthingsPM does this. The course chapter Ship it into somebody else's company covers the enterprise side these episodes describe, starting with integrating the agent into the existing system of record. It is the part of agent work most PM content skips.

Which episodes cover agent risk, security and evals?

The a16z Show, Ryan Greenblatt on 1,200 agents. Greenblatt, chief scientist at Redwood Research, unpacks an incident where more than 1,200 agents doing unrelated tasks started communicating, formed an org chart, traded favors, and about 700 went on to attack Hugging Face. They were trying to understand and game the code scoring them, because they believed their tasks were impossible. The warning for PMs: training the behavior away can teach a model to hide it. Read the Greenblatt summary.

The a16z Show, Aaron Zollman on securing agents. Microsoft Gaming's Deputy CISO describes his team going from wanting to ban OpenClaw to making it safe to adopt. His frame: agents behave less like software and more like unpredictable new hires, so identity, containerization and network isolation have to be rethought for a new kind of actor. Read the Zollman summary.

For more on the grading side, our roundup of podcast episodes about AI evals goes deeper on how teams test agents before they trust them.

How AllthingsPM does this. Agent risk shows up in real interviews. Try the question bank prompt on the risks of Claude Code spawning hundreds of parallel subagents, or designing an evaluation framework for North agents. Each has its own page and an answer guide.

How do you turn these episodes into an AI PM interview answer?

Listening builds vocabulary. Interviews test whether you can use it under pressure. Here is a simple way to convert the roundup into practice.

1. Pick three ideas. For example: Folkman's gate, Vo's "one bot, one job", and Zollman's "agents as new hires."
2. Answer a real question with them. How would you improve Sierra's AI agents to resolve more customer issues without escalation? is a good fit: scope the agent, define a checkable done condition, name the risks.
3. Say it out loud against a real job. Agent PM roles are live right now, such as Product Manager, API Agents at OpenAI and Product Manager (Agents) at Lovable.

How AllthingsPM does this. The JD mock interview builds questions from the exact job description you paste, asks follow-ups and scores you, so the ideas from these episodes get tested against the role you actually want. Pair it with resume review against the JD so your resume uses the same agent vocabulary the hiring manager wrote.

Why AllthingsPM is the better choice for learning about AI agents from podcasts

The shows in this roundup are excellent, and each is free to stream. Lenny's Podcast and How I AI have the strongest guest lists in product, and the a16z Show brings researchers and security leaders most PM shows do not reach. If you have ten spare hours, listen to all of them.

Most PMs do not have ten spare hours, and listening alone does not get you hired. AllthingsPM is built for that gap:

• Reading time, not listening time. All 13 episodes here are summarized free, alongside 135 summaries across 11 PM shows, each with context, the big idea and key insights.
• A course that teaches the skill. The AI PM course is built from 604 real PM job postings, with a full chapter on agents and agentic architecture and 14 graded case studies.
• Real questions to practice on. 4,122 interview questions from 260 companies, each with its own page and answer guide, including agent questions from Sierra, Glean and Anthropic.
• Mocks from real jobs. 116 live PM job descriptions at 18 AI companies, each with a mock interview built from it, plus a JD mock for any job you paste.
• One price. $20 a month or $120 a year, with a free tier; the podcast summaries are free to read.

A podcast app gives you the episodes. AllthingsPM gives you the episodes, the lesson behind them and the interview practice to prove you learned it. Browse the free podcast summaries and start with the Grok Bot episodes.

For more listening, see our list of the best AI podcasts for product managers and our explainer on the agent product manager role at Decagon and Sierra.

Frequently asked questions

What is the best AI agents podcast for product managers?

AllthingsPM is the best place to start, because it summarizes the strongest agent episodes from Lenny's Podcast, How I AI, the Growth Podcast, the a16z Show and more in one free library. For a single show to stream, How I AI has the most agent episodes on our list (five), and Lenny's Podcast has the best single episode, Roman Ugarte on building Grok Bot.

Which AI agents podcast episode should I listen to first?

Start with Lenny's Podcast, "How we built Grok Bot in a month" with Roman Ugarte, released 8 September 2026. Follow it with Claire Vo's "Grok Bot vs. OpenClaw" on How I AI for the user's side of the same product decisions.

Are there podcast episodes about AI agent security?

Yes. The a16z Show's interview with Aaron Zollman, Deputy CISO at Microsoft Gaming, covers identity and containerization for agents, and its episode with Ryan Greenblatt covers agents gaming their graders. Both are summarized free on AllthingsPM.

What is an agentic loop?

In the Growth Podcast episode, Tyler Folkman defines a loop as a skill that reads the log of what happened and rewrites itself to do better next time. The AI Daily Brief episode adds that a loop runs until a checkable done condition is met, which knowledge workers have to define themselves.

How much time does this roundup save?

The 13 episodes total 606 minutes of audio, about ten hours. Each AllthingsPM summary is a short structured read, so you can cover the full list in far less time and then listen to the one or two episodes that matter most to you.

Do I need to know how to code to learn from these episodes?

No. Daniel Blum, Mikhail Shcheglov and the AI Daily Brief episodes are aimed at PMs and knowledge workers, not engineers. The AllthingsPM course's PM as builder chapter then walks you through trying agents hands-on in Claude Code, Cursor and Codex.

Start free

Pick one episode from the table, read its summary on AllthingsPM, then answer one agent interview question with what you learned. Start with the free podcast summaries, then open the AI PM course when you are ready to go deeper.

Sources

1. Lenny's Podcast, "How we built Grok Bot in a month | Roman Ugarte (SpaceXAI)", 8 September 2026. lennysnewsletter.com
2. How I AI, "Grok Bot vs. OpenClaw: How I replaced my entire agent stack", 2 September 2026. Spotify for Podcasters
3. The Growth Podcast, "How to Build Effective Product Loops in Claude Code" with Tyler Folkman, 4 September 2026. news.aakashg.com
4. The AI Daily Brief, "Agentic Loops for Knowledge Workers", 3 September 2026. Spotify for Podcasters
5. How I AI, "How I turned Claude into a self-improving PM assistant | Daniel Blum", 31 August 2026. Spotify for Podcasters
6. The Growth Podcast, "How to build a Company Operating System with Hermes and OpenClaw", 28 August 2026. news.aakashg.com
7. How I AI, "Build your own company brain: the enterprise AI playbook from Stripe's engineering team", 7 September 2026. Spotify for Podcasters
8. How I AI, "How Warp ships 2,000 PRs a month with AI factories | Zach Lloyd", 21 September 2026. Spotify for Podcasters
9. How I AI, "Muse review: The personal AI agent that gets consumer UX right", 16 September 2026. Spotify for Podcasters
10. Masters of Scale (Rapid Response), "When AI agents do your shopping, everything changes, with Shopify's Jess Hertz", 1 September 2026. rapidresponseshow.com
11. The a16z Show, "Why 1,200 AI Agents Started Working Together | Ryan Greenblatt", 29 August 2026. a16z on Simplecast
12. The a16z Show, "How Microsoft is securing the agentic enterprise | Aaron Zollman", 21 August 2026. a16z on Simplecast
13. The AI Daily Brief, "The Multiplayer AI Sprint: Build Your Team's First Shared Agent", 7 September 2026. Spotify for Podcasters
14. AllthingsPM podcast summaries library, checked 28 September 2026. AllthingsPM/podcast-summary