Weak day for launches, and what’s left has the same structure underneath: four ways to charge rent on an asset that already exists. Nintendo set November 5 for the remake of a game from 1998. Cacau Show put Mickey in a chocolate box with 25 windows and a R$159.99 tag. Three brothers bought legitimate news sites and turned them into a content farm — what they bought wasn’t a newsroom, it was the residue of authority the domain still carries. And the frontier labs propose that every new model pass through an independent evaluator before shipping, which is a safety thesis and also a price list.
What ties it together is the Nintendo biography published today: the asset that the Switch 2 and Cacau Show are mining took 92 years to exist and was built with the cheapest part on the shelf. Gunpei Yokoi called the method “lateral thinking with withered technology” — the 1989 Game Boy was monochrome and beat color screens because it ran more than fifteen hours on batteries. That’s one theory of the moat: do something with common components that nobody thought of. The other theory is in the Fast Company piece, where the cost of entry is the barrier itself, and where Anthropic shows up with an estimated $517bn in compute leasing commitments signed in eleven months. The cheap moat and the expensive moat, on the same day.
product and growth
90 minutes of unfiltered product advice — Peter Sellis was Snapchat’s first PM, stayed seven years, became Head of Product at Discord, left, and now talks without a comms team. Haven’t listened yet; from the summary, what matters is the thesis that growth almost always comes from the core product, not from a growth team off to the side, and the post-mortem on why Snap’s ads business never got where it could have. Worth the structural note: the product interview only gets useful when the person talking has no job left to protect, which explains why nearly every product interview is useless.
brand and ip
Ocarina of Time Remake arrives November 5 — announced at the 40 years of Zelda Direct, Switch 2 exclusive, with the first gameplay footage. The original game is from 1998. The decision isn’t technical: it’s using the franchise anniversary as a window and tying the classic to the new console, so the old catalog becomes a reason to buy the hardware. Exclusivity is the conversion mechanism, not a distribution detail.
Cacau Show’s Mickey & Friends Advent Calendar — R$159.99, or R$149.99 for Cacau Lovers members, with 25 windows and three figurines drawn at random from five characters. The line goes down to R$54.99 for the mini panettone (R$49.99 for members). Two decisions inside one: the random collectible gives you a reason to buy the second box, and the member price called out on every SKU turns seasonality into signup capture. The licensed IP is the trust shortcut that lets you ask R$160 for chocolate.
media and attention
10 Sunday Reads — Ritholtz opens the list with Futurism’s report on Brown Brothers Media: legitimate news sites bought and converted into AI-generated content, with fake writers and plagiarism, and 50 million pageviews a month claimed by the company itself. I haven’t read the whole investigation, but the model in the blurb is clear: the asset bought is the aged domain and the revenue is the distribution inherited with it. On the same list, a piece about Athene putting half its surplus into exposure to SoftBank’s stake in OpenAI — worth it as a map of where this cycle’s leverage is living.
market and capital
The big AI labs’ safety push could come with a competitive advantage — the thesis comes with a mechanism, which is what makes it publishable. If the market converges on independent evaluators (METR, Redwood, Apollo) as a condition of launch, the seal becomes a fixed cost of entry; analyst Harrison Rolfes, of PitchBook, compares the arrangement to the Big Four of auditing, and notes that agent evals get more expensive as the model gets more capable. The detail that binds it is Anthropic having written, in a 2024 post, that the cost of evaluating large models is high and rising and that, without tight scope, this creates a risk of regulatory capture — the company named its own incentive before the critics did. David Sacks calls it a deliberate strategy; the Gartner analyst quoted in the piece gives the most useful read, that sincere concern and competitive advantage don’t exclude each other. It’s precisely because they don’t exclude each other that the structure works.
reading
The biography of Nintendo, the company that didn’t invent much — Brazil Journal publishes an excerpt from Super Nintendo, by Keza McDonald (288 pages, translated by Fernanda Abreu, Portfolio-Penguin). The part worth reading for anyone who designs product is the eight-by-eight pixel palette: Mario’s mustache exists so Miyamoto wouldn’t spend pixels drawing a mouth, the cap so he wouldn’t draw hair, and the contrasting colors so the eye would register the running animation. A constraint from 1981 produced the most recognizable face in the world. And the description of level design — the first seconds teach the mechanic in a place where nothing can go wrong, and then they let you go — is the best definition of onboarding I’ve read this month, in a text that never mentions onboarding. “Once the person understands what they need to do, the game becomes theirs,” Miyamoto says.
The Biggest Risk to the Economy — Ben Carlson uses 1873, Liaquat Ahamed’s new book, to argue that the bigger risk isn’t the bust, it’s the policy mistake after it. The numbers: from 1800 to 1940 American inflation ran at 0.2% a year; the peak-to-trough drop in US industrial production after 1873 was around 6%, an ordinary recession, and it turned into two decades of deflation because the great powers reordered the monetary system at the same time, with prices falling 20% to 25% worldwide in six years and 35% in American wholesale. The railroad-to-AI-buildout analogy is already a cliché; the part that teaches is the other one. Prolonged deflation transfers income from borrowers to lenders, for decades. Anyone signing a long compute contract is on one of the two sides of that ledger.
who wrote
TBM 440: The Problem With Putting People in Boxes — Cutler against personality tests at work, with the right argument: the stable label answers the wrong question. What matters isn’t what type someone is, it’s under which conditions they switch modes — how reversible the decision is, what evidence they require, who has legitimate authority to decide. The sharpest part is the asymmetry he names: the company norm is sediment from somebody’s behavior, but it gets treated as landscape (“that’s how things work here”), while adapting to that landscape is billed as an individual competency of whoever arrived later. It talks straight to the Sellis episode: designing a team is designing conditions, not collecting types.
stalled sources
First Round Review has gone 328 days without publishing, Calculated Risk 252, Gurwinder 266 and the Y Combinator blog 96. Four archives that served as the default reference for people building product and now exist only as a collection.