Talks, interviews, and podcasts worth the time — what was actually said, what holds up, and what it changes for the people building.
16 articles
CNBC's Memphis report is not really about pollution. It is about what happens to an AI buildout when the binding constraint stops being chips and becomes power, permits, and the patience of the people living next to it.
Most of Datadog's two-hour keynote is a product reel. One slide is not: the vendor selling you AI query optimization discloses that 90% of its model's suggestions failed validation.
Ng's fireside chat at LangChain's Interrupt has one piece of advice with a number attached, and one example that explains why most enterprise AI projects produce a rounding error instead of growth.
Andrei Alexandrescu's ACCU keynote argues abstraction survives AI-generated code for an unfashionable reason: not because humans need it, but because throwing it away is inefficient. He has an experiment to show it.
Microsoft re-ran its founding slogan for the AI era and pushed inference to the edge. The per-token meter does come off — and reappears as hardware you buy up front.
Thirty years after co-writing the book that named software architecture, Mary Shaw walks through the field's AI anxieties and shows most of them are re-runs. Then she names the one that isn't.
CrowdStrike's CEO says the industry drew the wrong lesson from July's autonomous-agent intrusion. His reinterpretation is more alarming than the original reading, and it is checkable.
Astra ships with cyber capability gated behind access tiers rather than available to everyone who pays. That is a structural change in how frontier models are released, and Altman says the quiet part on camera.
Dylan Field predicted AI would lower the floor and raise the ceiling on design work. On his tenth Config stage he marked his own homework, and only half of it passed.
Gergely Orosz told a Budapest audience the Meta AI account-takeover was caused by AI-written, AI-reviewed code. We checked which parts of that are on the public record.
Howard Marks came to Wharton with three memos and a lot of specific figures. We checked the ones that can be checked, and one of them is far more severe than the record supports.
IBM's four-way taxonomy for feeding an agent knowledge is the clearest one going. Its own worked example needs all four at once, which changes the question you should be asking.
The AMD CEO's address is built on one piece of advice she got at 25. We checked what the bet she used it on actually returned, and what that does to the advice.
The Alphabet CEO's commencement advice is that most early decisions are noise. Labor economics has measured exactly that question, and the first job is the one exception.
Chairman Warsh says pulling forward guidance let markets do the repricing. We pulled 20 years of Treasury data to check the claim, and the answer splits by tenor.
A computer science professor argues our field systematically values hard over useful, and that this is why LLMs got pointed where they did. We reproduced her numeral test to see if it still holds.
One email every Friday with the tools worth your time. No spam, unsubscribe anytime.