The Best AI Podcasts in 2026: 9 Shows Worth Your Hours

Sometime in 2023, everyone in tech started an AI podcast. Most of them are the same show: two hosts reading the week’s announcements to each other and saying “this changes everything” in slightly different orders.

Underneath that noise, a small set of shows has become genuinely essential. These are the ones that survived three years of my actual weekly rotation, not a list assembled from other lists. For each one: what it is, who it’s for, and the episode to start with.

They sort naturally into tiers, so that’s how I’ll give them to you.

The essential tier

1. Dwarkesh Podcast

The deepest interviews in technology right now, full stop. Dwarkesh Patel prepares like nobody else in the medium; guests regularly say some version of “nobody has ever asked me that” on air. The AI coverage isn’t news, it’s the people building the frontier being pushed on what they actually believe: lab CEOs, researchers, economists thinking about what comes after.

Start with: the Demis Hassabis episode, or the Satya Nadella one. Either will show you what preparation sounds like.

For: everyone. If you listen to one show on this list, it’s this one.

2. Latent Space

The AI engineer’s podcast. swyx and Alessio Fanelli talk to the people actually shipping: infra founders, model trainers, the engineers behind the tools you use daily. It assumes you build things, and rewards that assumption with the kind of detail that never makes it into keynotes.

Start with: any of their recap or “state of AI engineering” episodes, then work backward through the guests you recognize.

For: people who write code and want the practitioner’s view, not the commentator’s.

3. No Priors

Sarah Guo and Elad Gil, an investor and an operator who both actually understand the technology, interviewing founders and researchers at the top of the market. The questions are commercial without being shallow: what’s working, what’s pricing like, where’s the moat.

Start with: the Jensen Huang episode.

For: founders and investors who want the business layer on top of the research layer.

The technical tier

4. Machine Learning Street Talk

The most intellectually rigorous show in the space, and the least interested in being convenient about it. Episodes run long and go deep into whether these systems actually reason, what intelligence even is, and why the benchmarks might be lying to you. Guests are researchers, and the hosts push back like researchers.

Start with: any episode with François Chollet. The abstraction-and-reasoning debates are the show at its best.

For: people who want arguments, not takes. Bring your full attention.

5. The Cognitive Revolution

Nathan Labenz approaches AI with a scout’s mindset: he tests everything himself, updates in public, and interviews people across the whole spectrum from accelerationist to deeply worried without strawmanning either end. Consistently the most honest “what can these systems actually do right now” coverage.

Start with: one of his hands-on capability deep-dives, then an episode on a debate you already have opinions about. Notice how your opinions fare.

For: people who want calibration, not cheerleading.

The news tier

6. The AI Daily Brief

Nathaniel Whittemore, roughly twenty minutes, every day, ruthlessly signal-first. He reads everything so you don’t have to, and he’s unusually good at saying “this story doesn’t matter” out loud. This is the show that replaces doomscrolling AI Twitter.

Start with: today’s episode. That’s the point.

For: staying current in one commute. Fine at 2x.

7. Hard Fork

Kevin Roose and Casey Newton, from the New York Times. The most mainstream show on this list, and the funniest. It covers AI as culture and power, not just as technology, which is exactly the layer the builder-focused shows miss.

Start with: whichever recent episode covers a story you already know. You’ll hear the difference between reporting and reacting.

For: the one to recommend to friends outside tech. Also the one that will occasionally tell you something every other show missed.

The wildcard tier

8. Lex Fridman Podcast

The honest review: episode quality tracks guest quality almost perfectly, the interviews run three hours plus, and Lex’s style is to give guests room rather than push. With the wrong guest that means meandering. With the right guest it produces conversations that exist nowhere else, because nobody else gets these people to relax for three hours.

Start with: the Andrej Karpathy episodes. All of them.

For: long drives, and guests you’d listen to read a phone book.

9. Training Data

Sequoia’s AI show. Yes, it’s produced by a venture firm, and yes, you should keep that in mind. It earns its slot anyway: the access is real, the founders of the top labs and fastest-growing AI companies show up, and the hosts ask operator questions instead of press-tour questions.

Start with: an episode featuring a company you use.

For: a founder’s-eye view of where the money and conviction are actually flowing.

How to actually keep up with 9 shows (you can’t, and shouldn’t)

Nobody listens to all of these weekly. That’s 15 plus hours. The rotation that works: one daily news show, two deep shows, and let your group chat filter the rest. When three people you trust say the new Dwarkesh is essential, it enters the queue. The list above is a menu, not an assignment.

One more thing, because it’s the entire reason I’m qualified to write this list: AI podcasts are the densest note-taking material in the medium. Every episode is full of claims with expiration dates: model names, benchmark numbers, predictions with implicit timelines. The single most fun thing you can do with these shows is mark the predictions when you hear them, with a timestamp, and check back in six months. Half the value of the genre is watching who was right.

That’s what I built Margin for: press and hold while you listen, say “he just predicted agents replace SaaS by 2027,” and the note pins itself to that second of the episode. Six months later, you have receipts.

If you want the rest of my rotation beyond AI, it’s here: the 8 podcasts I actually listen to every week. And if you’re building a listening habit from scratch, start with the underrated tech podcasts nobody puts on these lists.

Selinay