简体中文
English
Website review
This review may include AI-generated content and is not individually verified. Domain, certificate and other facts come from the observed data on this page.
What is Chain of Thought Podcast?
Chain of Thought Podcast is a show for people building AI systems rather than just talking about them. Host Conor Bronsdon interviews engineers, founders, CTOs and product leads from companies such as NVIDIA, Google DeepMind and AMD, focusing on AI infrastructure, agents, evaluation and enterprise deployment. Its framing is practical: what is changing in the stack, and what does that change mean for teams shipping AI.
What you get
- Episodes — long-form conversations, often around an hour, with show notes and transcripts. Topics cluster around AI agents, evals and reliability, enterprise AI, RAG, hardware, observability, coding and agent memory.
- Explainers — short "plain answers" pieces aimed at questions builders actually ask, such as whether to use a single agent or a multi-agent system, how to give an agent identity and permissions, and how to choose between LLM-as-a-judge and human review.
- A newsletter — new episodes plus the host's essays connecting them, described as up to once a week.
- Multiple listening options — YouTube, Spotify, Apple Podcasts and RSS.
Who it suits
It is most useful if you are an engineer, founder or operator who needs to make decisions about architecture, evaluation or tooling, and you prefer hearing how named practitioners handled a problem in production over vendor marketing. If you want beginner-level AI explanations or quick news summaries, the episode length and depth will feel like more than you need. If you want academic research discussion, the guest mix is skewed toward industry roles.
One recent example is an episode with Kimberly Logan of Walrus Protocol on switching models without starting over — a concrete infrastructure concern for teams trying to avoid lock-in. Another features Sonar CTO Andrea Malagodi on why strong AI engineers can still be measured badly, which is a management and metrics question as much as a technical one.
How to start
Pick the topic list rather than the newest episode. If your immediate problem is evaluation, start with the evals or LLM-as-a-judge explainer, then listen to the related episode; if it is agent design, start with the single-agent versus multi-agent explainer. Episode pages include show notes and transcripts, so you can scan the argument before committing an hour. For a broader set of practitioner interviews, Latent Space covers similar ground from a builder angle.
Which episodes should I start with if I'm new to the show?
Start with the newest episode and one topic-specific episode that matches your work. The show’s own page offers a “New here? Start here” entry point, and the episode list is organized by topic, so you can pick a lane rather than sampling randomly.
Good starting points
- If you care about model portability: EP 77, “Switching Models Shouldn't Mean Starting Over,” with Kimberly Logan of Walrus Protocol. It is the latest episode, so it reflects the show’s current focus, and the subject suits anyone worried about vendor lock-in or rework when swapping models.
- If you work in a large organization: EP 76, “Your Best AI Engineer Might Have the Worst Metrics,” with Sonar CTO Andrea Malagodi. This is the enterprise AI angle, and the title signals a metrics and measurement discussion rather than a pure tooling demo.
- If you are building agents: EP 75, “Why Context Alone Isn't Enough for Enterprise AI Agents,” with WisdomAI CPO Kapil Chhabra. A useful counterweight if you have been told that better retrieval or a bigger context window solves everything.
- If you are a product engineer: EP 74, “AI Codes: Product Engineers Decide What to Build,” with Laurie Voss of Arize. This leans toward how teams decide what to build with AI, which is often the harder question.
How to choose
| Your situation | Start with |
|---|---|
| You want the current state of the show | EP 77 |
| You need buy-in inside a company | EP 76 |
| You are designing an agent system | EP 75 |
| You are deciding what to ship | EP 74 |
A practical next step: open the show’s explainer index first, find the question closest to your own, then listen to the episode linked from that explainer. The explainers are short and point to the guests who have handled the problem in production, which is a faster route than starting at episode one. If you prefer audio only, the show is distributed on YouTube, Spotify and Apple Podcasts, and there is an RSS feed for any podcast app.
How do I subscribe to the podcast and the newsletter?
Subscribe on the platform you already use, and sign up for the newsletter separately by email — they are two different lists.
- Podcast: Use the "Subscribe" link on Chain of Thought Podcast, or follow the show on YouTube, Spotify, Apple Podcasts, or its RSS feed. Subscribing gets you new episodes automatically; the site also offers direct episode audio downloads and show notes with transcripts.
- Newsletter: Enter your email address in the newsletter signup box and confirm. It is described as essays from the conversations plus new episodes, sent up to once a week, with unsubscribe available anytime.
A practical next step: if you mainly want episodes, pick one podcast app and subscribe there. If you want the host's written takeaways and explainers, add the newsletter as well. Expect roughly one email a week rather than a daily stream, so the newsletter is low-commitment alongside a podcast subscription.
Where can I listen to Chain of Thought Podcast on YouTube, Spotify, or Apple Podcasts?
You can listen on all three: YouTube, Spotify, and Apple Podcasts. The show's "Follow the show" section lists those three alongside an RSS feed, so any standard podcast app can also pick it up if you prefer a player that isn't one of the big platforms.
Choosing between them
| Platform | Best for | Trade-off |
|---|---|---|
| YouTube | Seeing guests and any on-screen material; the show reports 3,050+ subscribers there | Video eats more bandwidth and battery than audio-only |
| Spotify | One app for music and podcasts, with playback speed and offline downloads | Requires a Spotify account; some catalog features vary by region |
| Apple Podcasts | iPhone, iPad, Mac and CarPlay users who want a built-in app | Apple-only ecosystem; less convenient on Android or Windows |
A practical next step
If you are new to the show, start with the episode page rather than the platform feed: each entry has show notes and a transcript, so you can skim the argument before committing to a 55–85 minute listen. Episodes like EP 77 with Kimberly Logan of Walrus Protocol on switching models, or EP 76 with Sonar CTO Andrea Malagodi on AI engineering metrics, are good tests of whether the interview style suits you. If it does, subscribe on whichever platform you already use daily — the show publishes roughly weekly, and the newsletter covers the same themes in written form for people who prefer reading.
One caveat worth knowing: episode lengths run long, so if you commute in short blocks, download in advance and use a faster playback speed rather than trying to finish in one sitting.
What are the key takeaways from the podcast's explainers on enterprise AI failures and AI agent guardrails?
The explainers do not offer one universal fix. They pair each question with guests who have hit the problem in production, so the takeaways read as practical decision rules rather than research summaries.
Enterprise AI failures and ROI
The recurring theme is that failure is usually organizational and measurement-related, not a shortage of model capability. Projects stall when no one defines what success looks like before launch, when the metric chosen rewards activity rather than business outcome, and when the pilot never gets a path into daily operations. The Sonar episode with CTO Andrea Malagodi makes the counterintuitive point in its title: your best AI engineer might have the worst metrics. A team can ship impressive demos and still look unsuccessful if it is measured on the wrong thing.
For a reader deciding whether to fund a project, the practical test is to write down the business decision the system will change and the number that should move. If neither exists, the project is not ready, regardless of how good the model is.
AI agent guardrails
Guardrails are treated as a design layer, not a final filter. The explainer connects them to permissions, identity and evaluation: an agent needs to know who it acts for, what it may touch, and how its actions are checked. Context alone is not enough — the WisdomAI episode with CPO Kapil Chhabra argues that giving an agent more information does not give it judgment about when to stop or escalate.
How the two topics connect
| Question | Where the risk sits | Useful first move |
|---|---|---|
| Why do enterprise AI projects fail to show ROI? | Measurement and ownership | Define the decision and the metric before building |
| How do you set agent guardrails? | Scope of action and oversight | Decide what the agent may do, for whom, and when a human steps in |
An evaluation lead at a mid-size company can use the first explainer to reset executive expectations, then the second to scope an agent's permissions before it touches customer data. The two are meant to be read together: ROI depends on the agent behaving predictably, and guardrails are only meaningful if someone owns the outcome.
A useful next step is to start with the explainer that matches your current blocker, then follow the linked guest episode for the production detail. Browsing by topic such as AI Evaluation & Reliability or AI Agents helps if you are unsure which applies. The show is also listed on YouTube, Spotify and Apple Podcasts if you prefer audio over the written explainers.
Who are the typical guests on Chain of Thought Podcast and what topics do they cover?
Chain of Thought Podcast features practitioners who build and operate AI systems rather than commentators or futurists. The show's own framing is "conversations with the people building AI agents and infrastructure," and its guest roster includes people from NVIDIA, Google DeepMind, and AMD, alongside startup founders, CTOs, and product leaders.
Who tends to appear
- Engineers and technical leads working on AI infrastructure, model serving, and hardware.
- Founders and executives at AI startups — for example, the CPO of WisdomAI and the CTO of Sonar have appeared in recent episodes.
- Product and platform leaders at larger AI and developer-tooling companies, including Arize.
- Researchers and specialists from major labs, given the NVIDIA, Google DeepMind, and AMD connections.
The common thread is that guests have shipped something and can speak from production experience, not just theory.
What they talk about
Episodes cluster around a set of recurring themes, which the site groups by topic:
- AI agents — architecture, memory, and guardrails.
- AI evaluation and reliability — evals, LLM-as-a-judge versus human review, and what leaders should know.
- Enterprise AI — why projects fail to show ROI and how teams measure value.
- Open source AI, RAG and retrieval, AI hardware, and observability.
- AI coding — how product engineers decide what to build.
Recent episode titles illustrate the range: switching models without starting over, why context alone isn't enough for enterprise agents, and why a strong AI engineer might have misleading metrics.
How to use it
If you are choosing episodes, start with the topic index rather than the newest release. The site's explainers map common builder questions — agent identity and permissions, single versus multi-agent systems, eval strategy — to the guests who have handled those problems in production. That makes it easy to pick an episode that matches a decision you are actually facing, whether you are an engineer implementing agents or a leader trying to justify AI spend.
For broader context, other shows cover adjacent ground: Latent Space leans toward engineering deep dives, and Software Engineering Daily covers infrastructure more generally.
Cloudflare
User reviews (0)