The 11 Best AI Podcasts to Listen to in 2026
Most lists of the best AI podcasts have a small problem: they look as if someone copied the descriptions from Apple Podcasts, added a few famous names, and called it research.
That is not especially useful when every feed promises “insightful conversations with industry leaders.” It is even less useful when the show quietly stopped publishing.
So this list is deliberately short. Twenty-eight shows and 36 recent episodes were reviewed, and only 11 made the cut. Each winner does a different job: daily news, engineering, frontier research, practical workflows, startup building, policy, or a much-needed reality check on the hype.
One famous show did not make it because it had just ended its run. Another excellent podcast appears only as a bonus because, technically, it is still a product podcast. Curation means saying no… otherwise this would be a directory of 100 feeds you will never open.
Last checked: August 31, 2026 · 28 shows considered · 36 episodes reviewed
The best AI podcasts at a glance
| Podcast | Best for | Format | Cadence | Start here |
|---|---|---|---|---|
| Latent Space | AI engineers and builders | Technical interviews | Weekly / biweekly | Anima Anandkumar on foundation models for physics |
| Dwarkesh Podcast | Frontier-AI depth | Long-form interviews | 1-3 per month | Dylan Patel on compute and AI labs |
| How I AI | Workflows you can copy | Interview plus demo | Weekly | Daniel Blum’s self-improving PM assistant |
| The Cognitive Revolution | Capabilities, agents, and safety | Technical interviews and analysis | Several per week | Apollo Research on reward seeking |
| The TWIML AI Podcast | Machine-learning research | Researcher interviews | Roughly biweekly | Max Welling on AI and physics |
| No Priors | AI startups and company building | Founder interviews | Weekly | Eon on legacy data infrastructure |
| Practical AI | Applied AI without the paper prerequisites | Interviews and explainers | Weekly | Models, Harnesses, and Multi-Agent Systems |
| The AI Daily Brief | A compact daily update | News analysis | Daily | How to Start AI Coding If You Haven’t Yet |
| Interconnects | Open models and post-training | Conversations and audio essays | 1-3 per month | The open-model roundup |
| The AI Policy Podcast | Policy and regulation | Expert interviews | Weekly | Responding to AI Agent Containment Failures |
| Mystery AI Hype Theater 3000 | A critical counterweight | Evidence-led commentary | Monthly | Who’s Responsible for Responsible AI? |
These are not ranked from best to worst. A daily news show and a two-hour frontier-research interview are solving different problems. Choose the job you need done, not the logo you recognize.
If you only subscribe to three AI podcasts
For the typical DivByZero reader, I would start with this trio:
- Latent Space to understand what builders are actually shipping.
- Dwarkesh Podcast to think harder about the frontier, compute, science, and long-term consequences.
- How I AI to steal a workflow on Monday morning instead of merely collecting opinions.
That combination gives you one technical trade publication, one deep interview, and one practical demo. It is a much healthier information diet than subscribing to six daily-news feeds that repeat the same model launch.
If your job is different, use one of these stacks:
- AI engineer: Latent Space + TWIML + Interconnects.
- Founder or executive: No Priors + Dwarkesh + The AI Daily Brief.
- Product leader: How I AI + Practical AI + the Lenny’s Podcast bonus below.
- Policy or critical thinker: The Cognitive Revolution + The AI Policy Podcast + Mystery AI Hype Theater 3000.
How were these AI podcasts selected?
The first filter was simple: is the podcast still active? A show needed an episode in the 90 days before August 31, 2026. Seasonal shows could qualify, but a recent trailer or a final episode did not magically make a dead feed alive.
Then each finalist was judged on five things:
- Signal: Does an episode add reasoning, evidence, or a reusable workflow instead of paraphrasing announcements?
- Host craft: Is the host prepared enough to ask a useful follow-up or challenge a promotional answer?
- Consistency: Does the quality hold across at least three episodes?
- Distinct job: Does the show add something the other ten do not?
- Honest caveats: Are its sponsors, ownership, bias, length, and shelf life clear?
I made this selection from public episodes, transcripts, and show notes. I did not personally listen to every episode from beginning to end. Pretending otherwise would make this just another piece of AI-generated internet soup… and we have enough of that already.
1. Latent Space: the best AI podcast for engineers and builders
Latent Space, hosted by swyx and Alessio Fanelli, is closer to a trade publication for AI engineers than a generic technology podcast. It covers models, inference, agents, evals, infrastructure, data, and the awkward jump from a research result to a product people can actually use.
What makes it stand out is the connection between technical mechanisms and product decisions. In the Anima Anandkumar episode, the conversation moves beyond language models into weather, fusion, fluid dynamics, and why scaling an LLM is not the same as modeling a continuous physical system. The Joon Sung Park episode uses generative agents to distinguish imitation, digital twins, and causal simulation. A Chai Discovery episode asks whether an AI drug-discovery company should sell tools or own a therapeutic pipeline.
That is useful range. The show can talk architecture without forgetting the business built around it.
Start with: We have foundation models for language, not for physics with Anima Anandkumar, then Simulation: the new Scaling Law with Joon Sung Park.
The caveat: episodes are long and often assume that terms like eval harness, inference stack, or post-training do not need an introduction. If you are new to AI, start with Practical AI and come back.
2. Dwarkesh Podcast: the best for frontier-AI depth
Dwarkesh Patel gets famous guests, but access is not the reason his podcast belongs here. Preparation is.
He arrives with a model of the subject and keeps testing it with numbers, counterexamples, and follow-up questions. That produces unusually substantial conversations about frontier labs, compute, science, economics, and AI risk.
The Dylan Patel episode gets concrete about gigawatts, capital expenditure, supply chains, lab margins, China, and why Anthropic and OpenAI could control so much compute by 2028. Ryan Greenblatt tackles the messier question of what happens when AI can automate AI research, including recursive improvement and the difficulty of verifying machine-generated R&D. Grant Sanderson’s conversation moves into mathematical proof, conceptual discovery, and what students should learn when models can increasingly manipulate formal systems.
Start with: Dylan Patel for the industrial economics of AI; Ryan Greenblatt for capabilities and risk.
The caveat: many episodes run well beyond 90 minutes, and the feed also covers history, biology, and other intellectual rabbit holes. Subscribe if you like Dwarkesh’s method. Otherwise, choose by guest and topic.
3. How I AI: the best AI podcast for practical workflows
Most AI podcasts talk about what might be possible. How I AI, hosted by Claire Vo, asks guests to show what they actually do.
The screen-sharing and demo format changes the quality of the answer. Daniel Blum does not merely say that Claude helps with product management; he shows a Notion board managed by an agent, a morning brief, a company glossary that updates itself, and a weekly loop that learns from corrections. Ryan Carson explains how he spent $20,000 on Devin in a month, ran 15 agents in parallel, reviewed their pull requests, and compared Devin with Codex and Claude. Yana Welinder walks through using Codex and ChatGPT to research vendors, generate design specifications, work with CAD, and launch a fashion brand without an engineering team.
That level of detail makes the show immediately useful. You leave with a mechanism to test, not a vague instruction to “use AI strategically.”
Start with: Daniel Blum for a self-improving PM system; Ryan Carson for the cost and coordination problems of agentic development.
The caveat: tool-driven episodes age quickly. Treat them as working examples, not timeless best practices, and favor the most recent feed entries.
4. The Cognitive Revolution: the best for AI capabilities and safety
The Cognitive Revolution is for readers who want to understand what frontier systems can do, how we measure it, and where the safety story breaks.
Nathan Labenz tends to push conversations toward mechanisms and observable evidence. An episode with Apollo Research covers grader awareness, reward seeking, deception, and the limitations of monitoring chain-of-thought. Another asks whether recursive self-improvement is being rushed through poorly built evaluation and reinforcement-learning environments. The weekly highlights format connects cyber, bio, open weights, independent evals, governance, and military use without turning into a list of press releases.
Start with: the Apollo Research episode RL’s a Hell of a Drug for a technical case study; use the weekly highlights to see whether the broader format works for you.
The caveat: the feed is busy, the episodes are long, and sponsor blocks are noticeable. Its frontier-risk lens is valuable, but it should not be your only perspective. Pair it with The AI Policy Podcast or Mystery AI Hype Theater 3000.
5. The TWIML AI Podcast: the best for machine-learning research
The TWIML AI Podcast has survived multiple AI hype cycles because Sam Charrington keeps doing the same useful job: take a specific piece of machine-learning research and make it understandable without flattening it into a paper summary.
The archive reaches well beyond chatbots. Max Welling discusses foundation models for chemistry and physics, materials, thermodynamics, and architectures that may matter after today’s scaling playbook. Fatih Porikli explains why image generation needs better planning, rendering, control, and artifact removal rather than only larger models. Damian Borth explores the strange idea that trained models themselves can become the next dataset, with knowledge transferred through weight-space learning.
Start with: Max Welling for a broad view of AI beyond language; Fatih Porikli for a concrete generative-AI problem.
The caveat: this is not a news show and it will not hand you a workflow for tomorrow morning. Some episodes expect a technical foundation. That is also why the archive remains useful months later.
6. No Priors: the best AI podcast for startups
No Priors, hosted by investors Sarah Guo and Elad Gil, looks at where AI companies can create durable value: infrastructure, proprietary data, new product categories, distribution, and organization design.
The best episodes avoid the predictable founder-origin-story template. Eon’s founders explain why legacy enterprise data, access control, and non-human identities become more important when agents can touch production systems. Max Hodak connects brain-computer interfaces, retinal implants, and vision to the limits of comparing brains with artificial systems. Chess.com founder Erik Allebest offers a less obvious case: what happens to cheating, product design, and human meaning after machines become superhuman at the core activity?
Start with: the Eon episode for enterprise infrastructure; Chess.com for a product problem that is not another coding-agent demo.
The caveat: No Priors is produced by Conviction. The hosts’ investor access is an advantage, but it also creates incentives around markets, portfolio companies, and the venture-scale view of success. Keep that disclosure switched on in your head.
7. Practical AI: the best AI podcast for beginners who still want substance
Practical AI sits in a difficult middle and handles it well. It explains applied AI to developers, managers, and technical leaders without requiring them to read the original paper… but it does not reduce everything to five prompts for ChatGPT.
Recent episodes cover MCP and A2A, neutral standards, the line between human work and agent delegation, and organizational adoption. A conversation with Mike Lewis introduces an L0-L3 proficiency model and asks how non-technical builders turn tacit knowledge into measurable systems. Models, Harnesses, and Multi-Agent Systems separates the model from the feature, agent, harness, and multi-agent system, which is a distinction many “AI strategy” decks still fail to make.
Start with: Models, Harnesses, and Multi-Agent Systems as a primer; then Angie Jones on the foundations of the agentic era.
The caveat: experienced AI engineers may find some discussions introductory, and sponsor segments can interrupt the flow. For deeper implementation detail, graduate to Latent Space or TWIML.
8. The AI Daily Brief: the best daily AI news podcast
The AI Daily Brief is the feed for people who want to know what matters today without adding another hour-long interview to the queue.
Nathaniel Whittemore usually takes one central story and surrounds it with a compact set of headlines. Its value comes from framing rather than exclusive access. How to Start AI Coding If You Haven’t Yet distinguishes problems that automate, upgrade, or invent a workflow and proposes a realistic first project. Another episode reviews useful browser, temporary-chat, Gmail, voice, and video features. A third looks at real containment failures and the safeguards they exposed as missing.
Start with: the newest news episode, because freshness is the product. How to Start AI Coding If You Haven’t Yet is a good test of the practical format.
The caveat: there are plenty of ads, and daily-news episodes have a short shelf life. Use this podcast to notice a story. Use a primary source or one of the deeper shows here to understand it.
9. Interconnects: the best for open models and post-training
Interconnects is Nathan Lambert’s high-density guide to open-weight models, post-training, RLHF, distillation, and the competition between American and Chinese AI ecosystems.
The feed is part interview show, part roundtable, and part audio newsletter. One episode asks how long it will take AI to write a technical textbook better than its human author, exposing the difference between fluent sections and coherent long-form knowledge. The open-model roundups compare Kimi, Qwen, frontier gaps, distillation, safety, and geopolitics. A deeper Kimi episode examines performance per dollar and what an escalating open-weight ecosystem means for closed labs.
Start with: the open-model roundup for the conversational format; I wrote an AI textbook for Lambert’s personal, analytical mode.
The caveat: publishing is irregular, and some entries are essentially essays read aloud. If you want host-guest chemistry, choose another feed. If you want an informed map of open models, that quirk is easy to forgive.
10. The AI Policy Podcast: the best for AI regulation and governance
The AI Policy Podcast, from the Wadhwani AI Center at CSIS, connects model capabilities and technical incidents to the less glamorous machinery of regulation.
The show notes help. They include source documents, timestamps, and further reading, which makes it possible to check the argument instead of accepting a policy vibe. Responding to AI Agent Containment Failures covers incident reporting, lab incentives, technical talent in government, liability, and safe harbors. Other episodes connect cyber capability, watermarking, the EU AI Act, labor, wealth concentration, and planning under uncertainty.
Start with: the containment-failures episode for the bridge between engineering and policy; Adrian Brown on the AI-powered economy for the social and macroeconomic questions.
The caveat: this is an institutional, Washington-centered perspective. When the issue is European or global, use the documents it surfaces and add a source closer to the relevant jurisdiction.
11. Mystery AI Hype Theater 3000: the best critical AI podcast
Mystery AI Hype Theater 3000 is the corrective lens this list needs. Linguist Emily M. Bender and sociologist Alex Hanna examine AI as a socio-technical system involving language, labor, surveillance, education, healthcare, infrastructure, and power.
The tone is openly skeptical and often funny, but the arguments are built from papers, reporting, and expert guests. Who’s Responsible for Responsible AI? questions the presumed inevitability of automation in education and examines surveillance and harm reduction. State of Emergency looks at 911 triage, clinical systems, medical evidence, and accountability. Another episode dissects singularity narratives, consciousness claims, cyber incidents, and the media’s habit of anthropomorphizing software.
Start with: Who’s Responsible for Responsible AI? for the method; State of Emergency for a concrete high-stakes case.
The caveat: the perspective is deliberately polemical. Do not use it as your only source for judging technical capabilities, just as you should not use a lab or VC podcast as your only source for judging social effects.
Bonus: Lenny’s Podcast for product leaders in the AI era
Lenny’s Podcast stays in the bonus slot. It remains a podcast about product, growth, careers, and company building. But its 2026 feed has become too useful for AI product leaders to ignore.
Tara Seshan, product lead for Codex and ChatGPT Work, explains the move from rowing to steering, persistent AI coworkers, and product strategy built around capabilities that may still be months from release. OpenAI design leader Ian Silber discusses the convergence of roles, speed versus craft, and what remains distinctly human in design judgment. Anthropic’s Dianne Penn gets concrete about eval-driven development, the incubation of Claude Code and MCP, and why a product requirements document can turn into an eval set.
Start with: Tara Seshan for agents and product strategy; Dianne Penn for the technical-PM view inside a frontier lab.
The reason it stays a bonus is simple: a recent episode was entirely about enterprise sales, and future episodes will continue to cover hiring, growth, and classic product craft. For AI-native product work, though, this is probably the most valuable non-AI podcast in the queue.
For more shows about building companies rather than AI specifically, see my guide to the best SaaS podcasts.
Why Hard Fork is not on this list
Hard Fork published an episode on August 28, 2026. A mechanical activity check would call that active.
It would also be wrong.
Kevin Roose had already announced that the August episode would close the show’s run and that he and Casey Newton planned to launch a new independent podcast. At the time of this review, that successor did not yet have a verified name or feed. Roose’s announcement is exactly why checking a date is not enough.
Hard Fork produced excellent work. It is simply not a sensible new subscription today. The successor will be reconsidered when there is something real to evaluate.
Other AI podcasts worth sampling
Eleven slots create painful exclusions. These five are still worth choosing for a specific episode or need:
| Podcast | Try it when… | Why it missed the 11 |
|---|---|---|
| 80,000 Hours Podcast | You want extremely deep AI-safety interviews | Very long, not exclusively AI, and overlaps the capabilities-policy stack |
| Google DeepMind: The Podcast | You want direct access to lab researchers | Excellent but vendor-owned, with research territory already covered by TWIML |
| Machine Learning Street Talk | You want maximum technical depth and debate | Narrow audience and an uneven mix of rigor and speculation |
| Lex Fridman Podcast | A particular AI guest is uniquely strong | Active but generalist; only a minority of recent episodes are AI-centered |
| Better Offline | You want a forceful critical-tech perspective | Broader than AI and more polemical; MAIHT3 brings a tighter academic evidence base |
If books fit your learning style better, start with the best AI books. If you prefer watching people build, I also maintain a list of the best YouTube channels for startup founders.
How to build a three-podcast listening stack
Do not subscribe to all 11 because a list told you to. That is how podcast apps become guilt-management software.
Pick three layers:
- A radar: The AI Daily Brief for fast updates, or Practical AI for a slower applied overview.
- A depth source: Dwarkesh, Latent Space, TWIML, or The Cognitive Revolution, depending on your work.
- A corrective: The AI Policy Podcast or Mystery AI Hype Theater 3000 to expose incentives and consequences that builder feeds naturally underweight.
Then add a fourth show only when it has a distinct job. Product leaders can add Lenny. Open-model specialists can add Interconnects. Founders can add No Priors.
A good stack helps you notice, understand, and use what matters without consuming even more AI content. Three feeds, chosen deliberately, beat 30 episodes sitting unplayed.
Frequently asked questions about AI podcasts
What is the best AI podcast in 2026?
Latent Space is the best overall AI podcast for builders, thanks to its combination of technical depth, product context, and active coverage. Dwarkesh is better for long-form frontier thinking, while How I AI is the strongest choice for practical workflows.
What is the best AI podcast for beginners?
Practical AI is the best starting point for beginners who want substance. It explains models, agents, standards, and adoption without assuming research-paper knowledge. The AI Daily Brief is easier for news, but it should be paired with a deeper source.
Which AI podcasts are best for developers and engineers?
Start with Latent Space for AI engineering, TWIML for machine-learning research, and Interconnects for open models and post-training. The Cognitive Revolution is a strong addition for agents, evals, and safety.
What is the best daily AI news podcast?
The AI Daily Brief is the best compact daily update in this selection. It frames one main story and summarizes the rest. Because daily episodes age quickly, follow important claims back to primary sources.
Which podcasts cover generative AI and AI agents?
Latent Space, How I AI, The Cognitive Revolution, and Practical AI all cover generative AI and agents from different angles: engineering, real workflows, frontier capabilities, and applied adoption.
Is Lenny’s Podcast an AI podcast?
Not formally. Lenny’s Podcast still covers product, growth, careers, and company building. Its recent feed contains unusually strong AI-product interviews, so it is included here as a bonus rather than one of the 11 AI-specific shows.
Are Lex Fridman and Hard Fork AI podcasts?
Lex Fridman’s podcast is active but generalist, so it is better chosen by guest than treated as an AI subscription. Hard Fork ended its current run in August 2026; its announced independent successor had no verified feed when this list was checked.
How do you check whether an AI podcast is still active?
Look beyond the date of the latest item. Check the official RSS feed or site, confirm that the entry is a real episode rather than a trailer or cross-promotion, and look for closure or rebrand announcements. This list uses a 90-day activity window and will be checked quarterly.
This guide was last checked on August 31, 2026. Feed activity, recommended starting episodes, rebrands, and closures will be reviewed quarterly rather than cosmetically changing the date.
