weekly-videos-2026-07

Agent “skills” — reusable folders of instructions, scripts, and assets that a model loads on demand — have quietly become the packaging unit of the agent ecosystem. Philipp Schmid opens this AI Engineer talk with a brutal statistic from Skills Bench v1.1: of 50,000+ published skills, almost none have evals. Most were AI‑written and never tested. And because agents are non‑deterministic, without evals you have no way to tell whether a failing task is your skill’s fault, the model’s fault, or just noise....

Jonathan Blow gave this hour-long talk at DevGAMM in 2019, and in the seven years since it has quietly become one of the most-cited critiques of modern software engineering. Blow — the game designer behind Braid and The Witness, and the creator of the Jai programming language — makes a claim most programmers instinctively resist: technological knowledge is not on a monotonic climb. Civilizations lose capabilities all the time. Ours is probably losing them right now, and software is one of the leading indicators....

Model training is the slowest inner loop in an ML organization: one big run at a time, days or weeks per iteration, mostly serial. Lee Robinson opens this AI Engineer talk with a blunt framing — the whole game at Cursor right now is to shrink that inner loop, because whoever iterates fastest ships the best coding models. Every part of the talk is a concrete answer to “what does that actually look like?...

This week’s Classic of the Week is Rich Hickey’s 2010 Clojure Conj talk “Hammock Driven Development.” It’s one of the most-cited talks in the Clojure community and — read charitably — one of the least dated pieces of software-engineering advice from that era. Sixteen years on it reads almost eerily well as a critique of “just tell the agent to build it” culture: Hickey’s whole thesis is that the important work happens away from the keyboard, and we’ve built an industry that pretends otherwise....

Geoffrey Litt — design engineer at Notion, ex-Ink & Switch — opened the AI Engineer Design Engineering track with what he called a hot take: in 2026, it is still important for humans to understand how their code works. The framing sounds obvious until he lands the actual claim: agents are now writing 50,000-line PRs, and the practices that let a person stay a genuine participant in that project are not the practices that used to work....

Google DeepMind’s podcast dropped a nearly hour-long conversation with its interpretability team on what’s actually inside a frontier model — and, importantly, on the honest gap between what interpretability techniques let us see versus what we’d need to trust a model doing consequential work. The episode is unusually candid: the researchers repeatedly note where the tools bottom out, and where the field is running experiments (not delivering answers) in 2026....