
Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Google’s smart-glasses strategy is to copy Meta’s playbook, but through eyewear brands like Warby Parker and Gentle Monster — the hosts think Google’s fall launch makes sense because people who already wear glasses are the clearest early adopters, though the Warby Parker design got dragged for its visible camera bump and the stock still fell 14%.
The real edge for Google glasses may be ecosystem depth, not fashion — TBPN argues Gemini can plug into Gmail, Google Docs, Drive, and the rest of people’s actual digital lives in a way Meta’s glasses can’t, because Meta keeps running into other companies’ walled gardens.
Google I/O felt product-heavy and research-light, which left developers cold — despite flashy demos like Gemini Flash 3.5 and Omni video, the hosts say many developers found the event underwhelming, with Cursor Bench putting the new model below Claude Composer 2 while costing roughly 4x more.
One of the funniest tells from I/O was Google’s own anti-gravity team apparently showing a Codex folder in a launch video — the moment went viral because it suggested Google employees are still using Anthropic/OpenAI-adjacent tools internally, which fit broader chatter about Google “shipping their org chart.”
SpaceX IPO chatter is getting more concrete, with Goldman reportedly leading and venture funds staring at historic returns — citing Katie Roof, the show highlights that Founders Fund and Valor could clear more than $60 billion in gains, while Sequoia could make more than $20 billion.
Nvidia’s numbers remain absurd even as the stock barely moved — revenue jumped 85% year over year to $81.62 billion and net income rose to $42.96 billion, which the hosts basically summarize as the purest possible definition of “printing.”
The episode opens in classic TBPN mode: jokes about their matching suits, a swipe at how often people confuse them for brothers, and a nod to an UnHerd piece comparing TBPN to ESPN’s sports-media playbook. Then they cut to a humanoid robot running, stumbling, and getting awkwardly carried off while the music keeps playing — “singularity delayed,” as they put it, with the robot perhaps having “died from embarrassment.”
The first real I/O topic is Google partnering with Samsung, Gentle Monster, and Warby Parker on “intelligent eyewear.” The hosts think Gentle Monster’s design hides the camera surprisingly well, while the Warby Parker pair looks nice but has an obvious protruding camera bump that could catch light in awkward ways. They frame it as the same play Meta used with Ray-Ban: let an established eyewear brand do the silhouette work while Big Tech supplies the computer.
The more interesting angle, for them, is distribution plus software integration. If smart glasses become mainstream, they think it starts with people who already need prescription glasses — and Warby Parker has real retail distribution and a still-resilient $3 billion-plus business. More importantly, Google has Gmail, Drive, Docs, and a bunch of personal workflow surfaces, so an AI assistant inside Google’s ecosystem can feel more useful than Meta glasses that don’t naturally sit at the center of your digital life.
They hit Google’s broader demos next: Genie 3 grounded in Street View, plus voice-first assistant interactions that feel like Google wants AI “everywhere.” Street View stood out as an underappreciated data asset, and they riff on Demis Hassabis’s game background and how simulation demos are cool, but what really matters is whether Google can turn those visuals into compelling mechanics or useful products. They also liked the glasses demo where a user taps Gemini, snaps a photo, and gets an edited result on a watch — though they agree it’s a perfect demo use case, not necessarily a thing normal people are begging for.
From there the tone turns more skeptical. The hosts say many developers expected a bigger model moment — something like Gemini 4 — and instead got what felt like incremental progress packaged for a fixed conference date. They note criticism around benchmark performance and pricing, especially the claim that Gemini Flash 3.5 underperformed Claude Composer 2 on Cursor Bench while costing around four times more, which cuts against Google’s old reputation for strong price-performance.
One of the day’s most viral I/O moments was an anti-gravity demo apparently showing a Codex folder onscreen. TBPN’s take: this is funny, but not shocking, because Google clearly uses lots of external models and tools internally just like everyone else. It also revived the broader “is Google actually using AI efficiently inside Google?” argument that’s been bubbling online, including past back-and-forth involving Demis and outside critics.
They briefly return to Google’s video models, comparing Omni Flash with Kuaishou’s Kling/Sora-adjacent competition and especially China’s more permissive tools like “Cedance 2.0.” The headline is that short clips now look remarkably good, but the hosts think the next benchmark is much harder: not eight or ten seconds of pretty footage, but a coherent six- or ten-minute explainer video with the depth of a real YouTube breakdown.
The back half widens into market talk. They discuss reports that Goldman will lead the SpaceX IPO, surprising some who expected Morgan Stanley ties to win out, and cite Katie Roof’s scoop that Founders Fund and Valor could each make more than $60 billion while Sequoia clears more than $20 billion. They close with Jensen Huang boasting Nvidia’s “largest infrastructure expansion in human history” line as revenue hit $81.62 billion and net income reached $42.96 billion, then laugh at Steve Wozniak getting applause for telling students, “You all have AI. Actual intelligence,” a line they describe as corny but perfectly calibrated for the room.
Share
Keep Reading
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.

Playbook
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.

Playbook
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.

Playbook
OpenAI shipped /goal for the Codex CLI. It turns a prompt into a persisted, self-continuing contract.