TMTB Podcast Intel is produced by Ethmos, research agents that listen to every episode in TMTB’s coverage universe and write the briefs below.
Get briefs like this for the podcasts, companies, and themes you cover.
Today’s Essential Episodes:
Ben Thompson on Agents, Amazon’s Logistics Moat and Who Owns the Customer When the Agent Buys
TBPN with John Coogan and Jordi Hays, and Ben Thompson, founder of Stratechery (the Thompson segment)
Why he says agents win by killing tedious chores (returns, mail, subscriptions) rather than making anyone more productive, the home-inventory bot he built, why Amazon’s delivery certainty and returns beat Shopify’s asset-light model once an agent does the buying, Walmart as the overlooked rival, and the auto-cancel agents already generating support tickets at Stratechery.OpenAI’s Houda Nait El Barj on the Personal-Agent Landscape and How Agent Mediation Reshapes Interactions and Science
MTS, and Houda Nait El Barj of OpenAI
Why she says capability is solved and interface is the bottleneck (Instinct and Muse living inside iMessage and WhatsApp), the alignment line on automating meaningful human acts, agents mediating friendships, and how cheap hypothesis de-risking could make rare-disease drug discovery economically viable.YouTube CEO Neal Mohan on AI Across YouTube, the Slop Question and Why Human Storytelling Beats the Algorithm
Big Technology with Alex Kantrowitz, and Neal Mohan, CEO of YouTube
AI now sits under creation, discovery and search (drafts, thumbnails, A/B tests, a Gemini-powered Ask); why he says AI-made content gets no penalty and no favor, only the audience decides; the DeepMind partnership; Netflix’s podcast push; Sunday Ticket; and the view-count change some creators saw lift counts 30-40%.Cerebras EVP of Go-to-Market Alex Varel on Why He Joined, Selling Inference Compute Against NVIDIA and Building the GTM Machine
Hunters and Unicorns with Ollie Kuehne and Simon Kouttis, and Alex Varel, EVP of go-to-market at Cerebras
Why he left software (MongoDB, Zscaler) for a chip company he barely understood, the bet that inference, not training, is the growth market, CFO-ready value selling against NVIDIA in a supply-constrained market, a 200-megawatt European buildout across Norway, Finland and France, and a lean, execution-first sales org.Google DeepMind CEO Demis Hassabis on AGI Around 2030, the Foothills of the Singularity and the Race Between Labs
If/Then, Stanford GSB’s podcast, with Stanford president Jonathan Levin, and Demis Hassabis, co-founder and CEO of Google DeepMind
AGI at roughly 2030 plus or minus a year, the coming decade as ten times the Industrial Revolution compressed, why AlphaFold’s 200 million structures were given away, the prisoner’s-dilemma race between labs and nations, and his call for fast-adapting regulation informed by the leading labs.Reid Hoffman and Aria Finger on When AI’s Builders Sound the Alarm: Hubinger’s 10%, Amodei’s Pacing Call and the Case for Steering
Possible with Reid Hoffman and Aria Finger
Anthropic researcher Evan Hubinger’s above-10% extinction estimate, Amodei’s embedded-evaluator and inter-lab proposal, Sacks’s cartel charge, why Hoffman ranks bio, cyber and jobs as the real risks, China 6-12 months behind, and his dual-mandate case for AI leadership plus prosperity for the struggling 60%.Azure CTO Mark Russinovich on Fool’s Gold: Defending Open-Weight Models by Feeding Attackers Confidently Wrong Answers
Scott & Mark Learn To…, Microsoft’s podcast with Scott Hanselman and Mark Russinovich, CTO of Microsoft Azure
How refusals get stripped from frontier-class open models (Kimi K2, GLM, Qwen 3) with one prompt or a direction edit, and his answer: train the model to give plausible but fatally flawed recipes once the guardrails are gone, with no regression on normal use and generalization beyond the training set.
Read the Brief on Ethmos | Listen
