Sarah Guo: "The Second Station"

Sarah Guo published her own essay and pitched it with one line: startups that "want AI-scale revenues must measure themselves on outcomes over seats and stations" @saranormous. She builds it around a factory example. A smart robot raises the slowest station from about 100 parts an hour to 500, but the next station tops out at 125. To get paid on the whole line, the vendor may have to help the plant manager fix that second station too @saranormous. She argues that companies will pay at scale "only for returns" @saranormous. Getting those returns means someone has to run the negotiation inside the customer. Her candidate is an "outcomes owner" @saranormous.

She admits outcomes pricing "stresses me out some" because it gets complicated. She still expects the market to move toward it, and she cites Sierra and Cognition's Devin guarantee as companies already there @saranormous. Her conclusion: "If the return you're promising depends on the second station, then make the second station your business, too" @saranormous. Article

Martin Casado: Tutor Intelligence's robotics report

Martin Casado called Tutor Intelligence's report on using frontier models for robotics training "a great technical report." He said it is "a compelling demonstration that this is the direction things are going" @martin_casado. Report

Ryan Hoover: "How Product Hunt Almost Died"

Ryan Hoover shared his own post on how Product Hunt nearly died in 2016 and what he learned from it: "Hope it's useful for other founders" @rrhoover. Post

The AI cost-curve posts

Two investors responded to a set of posts about the pace of AI progress.

• Casado quoted @ramez, who claims Anthropic is gaining "exponentially less capability per billion dollars." According to @ramez, it now costs 8x as much to raise AI capabilities as it did two years ago. He credits @alexolegimas, @sebkrier and @michael_wiebe for inspiration @ramez. Casado's take: that is what the original scaling laws predict, but those were power laws, not exponentials. "Good news, scaling laws hold. Bad news, fighting a power law is really hard. And requires tons of money" @martin_casado.
• Amjad Masad shared Seb Krier's updated chart post @amasad. He drew on his time at Facebook, where growth that looked exponential was really "a series of overlapping S-curves." He thinks AI is probably the same, which "doesn't mean progress will have to slow down" @amasad.

Shorter picks

• Bill Gates on Great Question: he recalled asking Dario Amodei what children should learn for the world AI is building. Amodei said "they should study the classics" CBS Evening News. Gates's own advice was to pursue what you're curious about and get ahead on using AI CBS Evening News. Video
• Amjad Masad pointed readers in Japan to Enterprise Vibe Coding, a book by SMFL's digital innovation team on how AI is transforming enterprise software development. It comes out November 27, and Masad appears on the cover @amasad.
• Jason Lemkin promoted a new SaaStr piece on revenue per employee. Its headline numbers: $237K is the old median and $655K the new top tier. Its takeaway: "Under $300K per employee will draw board questions now" @jasonlk. Article He also quoted a passage citing Wiz, which counted at least nine AI developer platforms that disclosed security incidents in October @jasonlk.
• Elad Gil called it a "Neat idea" that @Salim_AIi is running Opus 5.5 agents over roughly 12,000 pages of notebooks by Gauss, Riemann, Hilbert and Grothendieck. Gil noted that Feynman also left large stacks of unpublished work @eladgil. Salim's best find so far, by his account: unpublished Riemann sheets that state the Rankine–Hugoniot conditions a decade or two before Rankine and Hugoniot @Salim_AIi. Post
• Patrick Collison shared a San Francisco Chronicle article on a parks donation by Airbnb CEO Brian Chesky: "Cool! Kudos to @bchesky" @patrickc. Article
• Elon Musk called @BollFILMS's article "The Lonely Wolf Killer in Polen and the Status Quo of Reality" "Insightful" and gave no further reason @elonmusk. The piece argues that media and crime statistics misattribute violence, using the Krakow shooting as its example @BollFILMS. Article