7 AI podcasts worth listening to in 2026 (skip the hype)
AI podcasting has a hype problem. The category is full of shows that read a press release, invite the founder on, and describe whatever they're demoing as the beginning of the end of human labour. That is not a report; that is content marketing with a microphone.
But there is also a smaller set of shows in this category that do actual journalism, actual technical explanation, and actual argument about what these tools mean. If you listen to two or three of them regularly, you end up better informed than if you doom-scroll a hundred X threads a day.
This is a curated shortlist of seven such shows. They are chosen for signal over volume — the ability to distinguish "this thing shipped and I've used it" from "this thing was announced." All seven are on Podtastic, so if you want to try one, tap and go.
1. Hard Fork — The New York Times
Kevin Roose and Casey Newton, weekly, roughly 90 minutes. This is the show most sensible people should start with. It is not the deepest technical show and it is not the most contrarian, but it is unusually good at three things: getting the newsmakers on the record (they had Satya Nadella on and pressed him), covering the incident of the week in enough depth that you leave the episode understanding what happened, and admitting when they have changed their minds.
Roose is a reporter first and it shows. He asks the annoying follow-up. Newton is more of an industry insider — he ran Platformer for years — and brings the "here is who is quietly furious about this at the company" layer. Together they cover consumer AI, the labour and copyright fights, and the safety-vs-shipping culture wars.
Listen when: an AI incident hits the front page and you want the reported version by Friday, not the hot take on Wednesday.
2. Intelligent Machines — TWiT
Jeff Jarvis and Paris Martineau, weekly, roughly 90 minutes. This is the show I keep coming back to when a big AI incident happens because they take the time to get a specialist on. When the Hugging Face breach story broke, they had Nate Jones on to dedicate most of the arc of one episode to it — "Hugging Face Breach and Autonomous Agents", "Analyzing the Hugging Face Security Incident", "Guardrails and Unintended Model Behaviors". That's the level.
Jarvis is a media professor and his framing is often about journalism and the open web, which pairs well with tech coverage. Martineau brings the reporter's instinct for what publishers and platforms are actually doing to each other. Guests tend to be researchers and specialists rather than founders on a press tour.
Listen when: you want a technical incident actually explained by someone who's read the paper.
3. Practical AI — Chris Benson and Daniel Whitenack
Weekly, roughly 45–60 minutes. This is the show for people who work with AI, not just people who read about it. Benson and Whitenack are engineers and their guests are engineers, so the discussion tends to be about what actually breaks: infrastructure, evaluation, agent architecture, edge deployment, the economics of running a model at scale.
If you have ever had to explain to a non-technical friend why "just fine-tune it" is not actually a plan, this is the show that would help them understand. It is also unusually good at not confusing "we announced" with "we shipped."
Listen when: you are trying to build with AI and you want peers on the microphone, not evangelists.
4. The Cognitive Revolution — Nathan Labenz
Weekly, often 2+ hours. Labenz is a red-team practitioner who spent time testing GPT-4 pre-release, and it shows in the questions he asks. This is deep-in-the-weeds AI safety, alignment, and policy — with guests who are actually building or evaluating frontier systems.
The episodes are long. That is by design. If a topic is complicated (recursive self-improvement, agent hierarchies, export controls, evaluation methodology), Labenz will spend the time on it. This is not a show to put on while cooking; it is one to sit with. Bring the summary from Podtastic and let it map the arc for you before you commit two hours.
Listen when: you want to understand a frontier problem to the point where you could argue about it, not just describe it.
5. Dwarkesh Podcast — Dwarkesh Patel
Long-form interviews, no fixed cadence. Patel does an unreasonable amount of pre-reading before each conversation, and his guests notice. When he talks to a physicist he speaks physics; when he talks to an ML researcher he speaks ML; when he talks to a historian he speaks history. This is closer to Tyler Cowen's Conversations with Tyler than to a tech podcast — the AI episodes are excellent but the show is really about how people who are the best in the world at something think.
The AI episodes worth starting with are the ones with lab researchers and semiconductor supply-chain analysts. The former will teach you why models scale the way they do; the latter will teach you why they might stop scaling for boring physical reasons.
Listen when: you want to leave a conversation genuinely smarter, and you can give it your full attention.
6. The Vergecast — The Verge
Nilay Patel and David Pierce, daily as of 2026 (Monday to Thursday hosted by Pierce with rotating expert staffers, Fridays with Nilay + Pierce). This is not strictly an AI podcast, but AI is now such a large fraction of consumer tech coverage that The Vergecast reliably covers it several times a week — often ahead of the specialist AI shows, because they have the reporting muscle.
Patel is a lawyer by training and treats AI companies with the appropriate skepticism about their platform-power ambitions. Pierce is more product-focused and often the one asking "yes but does anyone actually use this every day?" Their AI coverage sits inside a broader beat, which means you never lose the industry context.
Listen when: you want the AI story of the week in the context of the rest of the tech week.
7. Lex Fridman Podcast
Long-form conversations, weekly-ish, often 3+ hours. Fridman gets a specific kind of guest others don't — Elon Musk regulars, Meta researchers, Anthropic executives, foreign heads of state. That access is a feature, but it comes with a style you have to know going in: he does not push back hard, and the tone is contemplative rather than adversarial.
Read him as a place to hear people speak at length with minimal interruption, not as a place to hear them cross-examined. For frontier-lab founders in particular, this is often the longest-form version of what they think out loud. Pair with Hard Fork for the adversarial version of the same story.
Listen when: you want length and access, and you're willing to bring your own skepticism.
A working rotation
You do not need to listen to all seven. If you listen to two of them consistently, you will be ahead of most people who claim to follow AI. A working default rotation:
- One weekly reported show. Hard Fork or Intelligent Machines. Both are excellent; Hard Fork is broader, Intelligent Machines goes deeper on the incident of the week. Pick one, add the other later if you have appetite.
- One weekly practitioner or specialist show. Practical AI if you work with AI. The Cognitive Revolution if you want to understand safety and alignment.
- One long-form show when time permits. Dwarkesh Podcast or Lex Fridman Podcast — as an occasional deep-dive, not a daily driver.
- The Vergecast as your ambient tech feed. It will surface AI stories anyway, plus the rest of the industry.
Filtering the rest
If you are evaluating an AI show that is not on this list, three questions I use to filter:
- Do they have researchers on, or just founders? Founders are not neutral — they are selling. Researchers can be wrong but they are usually wrong for interesting reasons.
- When they cover an incident, do they explain what actually happened? A show that describes an outage or a paper without walking through the mechanism is a headline-reading show.
- Do the hosts admit when they were wrong last week? The AI space is moving fast enough that everyone gets calls wrong. A host who never mentions their misses is a host who is not tracking their own accuracy.
Most shows in this category fail at least one of those. The seven above pass all three, at least most weeks.
Try one right now
All seven of these are on Podtastic, so subscribing is one tap:
- Hard Fork
- Intelligent Machines
- Practical AI
- The Cognitive Revolution
- Dwarkesh Podcast
- The Vergecast
- Lex Fridman Podcast
If you're new to Podtastic, it's an independent podcast app designed for listeners, not for advertising companies. Smart Skip skips past interruptions other listeners skipped so you don't have to; Smart Summaries and Smart Topics let you triage a 90-minute AI podcast to decide whether it's the one for your commute; Skip Silence and Enhance Voices make voice-heavy content easier on the ears. Download now to try it.
