Best AI Engineering Podcasts for 2026: 14 Compared

The AI podcast you need depends on the question you are trying to answer. A discussion of post-training can help you understand a model’s behavior. An interview about agent memory can change how you design an application. A daily news recap helps you decide which developments deserve a closer look.

For AI engineering, start with Latent Space, Practical AI, or TWIML. For technical research, try Machine Learning Street Talk or Interconnects. For agent context, infrastructure, and deployment conversations, include Chain of Thought. The table below compares fourteen shows and gives you a specific episode to try for each. Ten more are listed after the profiles with one line each.

Disclosure: I host Chain of Thought, which is included here. These are editorial recommendations, not independent rankings. My show is subject to the same requirement as the others: a named audience, a linked starting point, and a reason you might choose something else.

Compare the shows

Sources checked September 6, 2026. Dates and lengths describe the selected examples, not each show’s latest release, average duration, or publishing schedule. Two newsletter feeds did not supply duration for the selected audio entry. Download the source metadata, including feed URLs and the example episodes.

Podcast	Choose it for	Example published	Example length
Agentic Conversations (formerly MLOps.community)	Production architecture and operating costs	2026-09-04	39 min
AI Engineering Podcast	Infrastructure, observability, and governance	2026-02-15	51 min
Chain of Thought	Agent context, evaluation, infrastructure, and deployment	2026-09-02	55 min
The Cognitive Revolution	Long interviews connecting AI systems and strategy	2026-09-01	97 min
Dwarkesh Podcast	Research, frontier AI, and compute economics	2026-08-25	Not supplied
Gradient Dissent	ML infrastructure and the companies building it	2026-08-02	79 min
How I AI	AI workflows and product demonstrations	2026-09-03	32 min
Interconnects	Post-training and open models	2026-06-16	Not supplied
Latent Space	AI engineering, models, and infrastructure	2026-08-21	70 min
Machine Learning Street Talk	Technical research and conceptual debates	2026-09-02	100 min
No Priors	AI companies, investment, and hardware strategy	2026-09-03	37 min
Practical AI	Applied AI and architecture	2026-09-03	46 min
The AI Daily Brief	Following AI news and industry changes	2026-09-04	24 min
The TWIML AI Podcast	ML research and engineering	2026-09-01	66 min

Where to start with each podcast

Agentic Conversations (formerly MLOps.community)

Demetrios’s show now appears in its feed as Agentic Conversations, so search for that name if you knew it as the MLOps.community podcast. The current description puts AI agents at the center of the show. It is a useful candidate for engineers who own the systems around a model: infrastructure, scaling decisions, and the work of keeping an application running.

Start with: The Five-Layer Cake Approach to Scaling AI Without Wasting Money (2026-09-04; 39 min).

The scaling episode is an entry point for thinking about where AI spending goes. Choose an episode tied to your architecture; a broad agent discussion will not necessarily answer a narrow model-serving question.

AI Engineering Podcast

Tobias Macey’s interviews concentrate on the operational parts of AI. The OpenLit conversation is a useful starting point if you need to understand how an LLM application behaves after deployment. Another listed episode examines Kubernetes, compliance, and AI sovereignty.

Start with: From Blind Spots to Observability: Operationalizing LLM Apps with OpenLit (2026-02-15; 51 min).

Treat this as an archive recommendation for now: the newest episode in the feed checked on September 6 was dated February 25, 2026. That is a statement about this feed snapshot, not an announcement that the show has ended.

Chain of Thought

I host Chain of Thought. The show is built around conversations with the engineers, founders, and technical leaders putting AI into products. If you are working through agent memory, retrieval, permissions, or evaluation, choose the interview that addresses the decision in front of you. The linked Slack conversation examines how collaboration context connects to coding agents.

Start with: Slack Wants to Be the Context Harness for Code | CPO Jaime DeLanghe (2026-09-02; 55 min).

For a different starting point, Jerry Liu discusses context quality and AI frameworks. These are interviews, not step-by-step coding tutorials. The Slack episode page includes a transcript and chapters so you can inspect the conversation before committing to the audio.

The Cognitive Revolution

Nathan Labenz’s conversation with MongoDB’s Pete Johnson is a useful companion to a discussion of agent memory. It connects database architecture to retrieval and the problem of remembering, changing, and forgetting information. The broader show covers builders, model capabilities, and the consequences of deploying those capabilities.

Start with: Write, Change, Recall, Forget: MongoDB’s Pete Johnson on How Retrieval Drives Agent Performance (2026-09-01; 97 min).

Give yourself time for this one: the selected interview runs about 97 minutes. Pick by guest and problem rather than assuming every episode is an implementation discussion.

Dwarkesh Podcast

Dwarkesh Patel’s interviews suit listeners who want to examine the assumptions behind an argument about AI. For infrastructure readers, the Dylan Patel conversation is a relevant starting point: it considers how compute could concentrate around frontier labs. Treat the guest’s forecasts as arguments to assess, not settled facts about the future.

Start with: Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028 (2026-08-25; not supplied).

The show also ranges beyond engineering into economics and other subjects. Choose it when you want to understand a research or industry question at length; a product walkthrough is a different need.

Gradient Dissent

Lukas Biewald’s show offers a view into AI through conversations with builders and company leaders. The selected interview with Fireworks CEO Lin Qiao is a relevant starting point for readers interested in inference infrastructure. The title’s token-volume figure is the publisher’s claim; this guide has not audited it.

Start with: 40 Trillion Tokens a Day (Yes, More Than OpenAI) | Lin Qiao, CEO of Fireworks (2026-08-02; 79 min).

Weights & Biases publishes the show, which is useful context when evaluating tooling discussions. The feed links to the publisher’s podcast index rather than a unique episode page; find the Lin Qiao interview by name there.

How I AI

Claire Vo’s show is organized around specific ways people use AI in their work. Its stated format includes screen sharing and workflows readers can try. That makes it a useful complement to interviews about model training or company strategy: the question is what someone actually does with the tools.

Start with: GPT-6 Astra is a banger - here’s everything I’ve built (2026-09-03; 32 min).

The selected episode walks through things built with GPT-6 Astra. Watch the video version when the demonstration depends on what is happening on screen; audio alone can leave out the useful part. Product reactions are time-sensitive, so check the episode date.

Interconnects

Nathan Lambert’s Interconnects is a strong candidate when your questions concern how models are trained and improved. The conversation with Finbarr Timbers focuses on frontier post-training recipes. It gives the engineering discussion a different starting point from a show about building applications on top of model APIs.

Start with: Frontier post-training recipe review with Finbarr Timbers (2026-06-16; not supplied).

Interconnects mixes written work and audio. The source used here is a newsletter feed containing audio entries, not a guarantee of a complete podcast catalog. Pick the linked technical conversation without inferring a publishing schedule from this sample.

Latent Space

Swyx and Alessio Fanelli cover AI engineering across models, agents, infrastructure, and applications. The selected Joon Sung Park conversation is an entry point into simulation and the questions it raises for building AI systems. It is useful when you want exposure to ideas beyond the tools already in your stack.

Start with: Simulation: the new Scaling Law — Joon Sung Park, Simile AI (2026-08-21; 70 min).

The subject range is broad. If your immediate problem is retrieval or inference cost, search the archive for that topic rather than treating the newest episode as required listening. The publisher links show notes from the feed.

Machine Learning Street Talk

Tim Scarfe runs MLST, with regular appearances from Keith Duggar. The show covers machine learning alongside cognitive science and philosophy of mind. The selected Tom McGrath interview gives you a recent, substantial conversation to sample before deciding whether that blend works for you.

Start with: Designing How AI Grows — Tom McGrath (2026-09-02; 100 min).

This is a good candidate for the listener asking for more technical depth. It is a less direct choice for someone seeking a quick framework tutorial. Allow about 100 minutes for the selected episode, or use its chapters where available.

No Priors

Sarah Guo and Elad Gil’s show approaches AI through conversations with founders and industry leaders. The Rene Haas interview is a relevant choice for engineering leaders who need the strategic context around chip architecture. Pair it with a systems-focused show if your work also requires implementation detail.

Start with: Redefining Chip Architecture with Arm CEO Rene Haas (2026-09-03; 37 min).

Use the Apple listing or the feed linked in the source file. The old nopriors.com link was parked when checked for this update. The selected feed item has no episode-page URL, so search the show listing for Rene Haas.

Practical AI

Daniel Whitenack and Chris Benson focus on applying AI in working systems. The selected architecture episode is a useful first stop for a developer moving beyond model selection and into deployment decisions. Their conversation with Angie Jones about the Agentic AI Foundation is another way into the interoperability side of agent development.

Start with: Less about Models; More about Architecture (2026-09-03; 46 min).

Start here if you want an applied discussion with a manageable first-episode commitment: the architecture example is about 46 minutes. The show covers a wide range of experience levels, so choose the topic that matches your existing knowledge.

The AI Daily Brief

Nathaniel Whittemore, also known as NLW, publishes analysis of AI news and its implications. The summer recap is a useful starting point for someone trying to regain context after stepping away from the news. A recap serves a different purpose from a deep interview about one deployment.

Start with: How AI Changed This Summer (2026-09-04; 24 min).

Use it to identify developments you want to investigate further. For an architecture choice or a product claim, follow through to the underlying source. The show name is easily confused with similarly named feeds; look for Nathaniel Whittemore.

The TWIML AI Podcast

Sam Charrington’s interviews connect machine-learning research with applications. The selected Justin Johnson conversation on world models and spatial AI is a useful research-oriented starting point. The archive also includes discussions of inference systems, agent security, and evaluation failures, making topic search especially useful.

Start with: World Models and the Future of Spatial AI with Justin Johnson - #775 (2026-09-01; 66 min).

Pick it when you want a sustained conversation about a specific technical subject. The breadth of the archive means you should check the date and assumptions of an older episode before applying it to today’s tools.

Also worth knowing

Ten additional shows not compared above. Each entry uses the publisher’s feed description and one dated episode example, with no starting-episode pick. Treat them as leads for further evaluation rather than recommendations. Feeds checked September 11, 2026; download the source metadata.

• Lex Fridman Podcast. Long conversations about AI and other technical and cultural subjects. The checked DHH episode runs 322 minutes. Example published 2026-08-26; 322 min.
• Hard Fork. Weekly New York Times show in which Kevin Roose and Casey Newton make sense of the week in tech and AI. Example published 2026-09-04; 79 min.
• NVIDIA AI Podcast. NVIDIA-produced series on the technologies shaping industries. The checked episode looks at Instacart’s AI-powered shopping cart. Example published 2026-06-24; 40 min.
• Last Week in AI. Skynet Today’s weekly summary of the AI news that matters. Example published 2026-09-08; 74 min.
• Training Data. Sequoia Capital partners interview AI builders and researchers about what is being built and why. Example published 2026-09-01; 52 min.
• AI + a16z. a16z talks with people building AI about its effects on enterprise IT and other sectors. Example published 2026-08-07; 24 min.
• Unsupervised Learning. Redpoint Ventures’ Jacob Effron asks AI guests what is real today and what may come next. Example published 2026-09-03; 58 min.
• Eye On A.I.. Biweekly interviews hosted by longtime New York Times correspondent Craig S. Smith with people working in AI. Example published 2026-09-10; 55 min.
• The Pragmatic Engineer. Gergely Orosz’s deep dives with experienced engineers at Big Tech and startups. The checked episode covers the building of Codex. Example published 2026-09-09; 73 min.
• Google DeepMind: The Podcast. Hannah Fry goes behind the scenes at DeepMind’s research lab to examine how AI is being used. Example published 2026-09-09; 44 min.

Build a listening queue around your work

If you are building AI agents

Pick an interview about the failure you need to understand. For context and memory, compare the Chain of Thought Jerry Liu conversation with The Cognitive Revolution’s Pete Johnson interview. For architecture, start with Practical AI. For operating costs, try Agentic Conversations. You do not need to subscribe to all four to get something useful from one episode.

Write down the decision the conversation might change: what to retrieve, what to remember, what to measure, or what authority to give an agent. If the guest never gets specific enough to help with that decision, move on.

If you work on inference or infrastructure

Combine an operational discussion with an industry one. AI Engineering Podcast’s OpenLit interview concerns observability; Gradient Dissent’s Lin Qiao conversation brings in an inference-provider perspective. Dwarkesh’s Dylan Patel interview is relevant to the bigger compute picture. Those conversations answer different questions, even though all three belong under AI infrastructure.

If you want technical research

Try MLST, Interconnects, and TWIML. Read the episode description first. A discussion of cognition, an account of post-training, and an interview about spatial AI require different background knowledge. Technical depth is useful only if it connects to what you want to learn.

If you lead a team or build products

How I AI is a starting point for workflows you can inspect. No Priors supplies a company and industry perspective. Chain of Thought and Practical AI offer conversations to bring into architecture and deployment discussions. Choose a concrete example your team can evaluate; an interview is a source of questions, not proof that a tool will work in your environment.

If you mostly need to keep up

Use The AI Daily Brief for a news overview and choose one longer conversation when a subject matters to your work. A daily feed can become another inbox. Skip episodes without losing the sense that you are allowed to follow the subject at your own pace.

Transcripts, activity, and what this guide can establish

A transcript is useful when you want to search for a concept, inspect a guest’s exact explanation, or share a passage with a colleague. Check the episode website as well as your podcast app. An absent RSS transcript tag does not mean a transcript is unavailable elsewhere. The linked Chain of Thought Slack page has a transcript and chapters; that individual check is not a comparative transcript-coverage score for the fourteen shows.

This edition uses public feed descriptions and episode metadata to identify shows and starting points. The recommendations are editorial judgments about fit. It does not claim a fresh listening review of every episode, a complete inventory of every AI podcast, or measured superiority on interview quality. Publication dates establish when the selected items appeared in the fetched feeds; they do not certify that advice is still current.

For the same reason, there are no download rankings here. Feed size, release frequency, and RSS metadata cannot establish audience size or whether you will find a conversation useful. The separate podcast-feed report addresses a different measurement question and has its own source data.

Corrections and version history

September 11, 2026: retitled to name AI engineering podcasts explicitly. This update also added the FAQ and a ten-show, feed-checked list. The fourteen compared entries did not change.

The July 5 version compared eleven shows and included numerical benchmark claims. A September 5 review could not locate the original archived dataset, so those historical figures could not be independently re-verified. This September 6 edition removes those figures from the recommendations, expands the guide to fourteen shows, and supplies a new dated source file for its episode examples. It does not reconstruct or validate the July benchmark.

The older Medium republication has not been updated to this edition and should not be used as its evidence. This page is the current version.

Choose one episode that addresses a problem you have this week. Read its notes, listen to the relevant section, and test whether the explanation changes what you would do. That is a more useful outcome than collecting another dozen subscriptions.

Get new Chain of Thought conversations and essays in your inbox.