Gonçalo Esteves / building capable systems Field Notes →

The Vault / People

Person

Astro Teller

co-founder of X, Alphabet's moonshot factory, where he leads it with the title "Captain of Moonshots". Guest of Moonshots (2026-10-05, no feed number) — Astro Teller: Inside Google X's 2,000-Idea Moonshot Factory & How Google Brain Made the T in ChatGPT, filmed at Moonshots Live

status · researched added 26 Oct 2026 updated 26 Oct 2026 from Moonshots #unnumbered personorganizationai see it in the graph →

What he is here for: co-founder of X, Alphabet's moonshot factory, where he leads it with the title "Captain of Moonshots". Guest of Moonshots #unnumbered, filmed at Moonshots Live 2026 and aired 5 Oct 2026, with Peter Diamandis, Dave Blundin, Alexander Wissner-Gross and Salim Ismail on stage. Peter introduces him on tape as his mentor for breakthrough projects.

Why it is in this archive

Promoted 2026-10-26. He is the guest of one held episode and named in Peter's Moonshots Live promotions on earlier ones. He is the operator behind the moonshot method the show is named for, and Peter attributes to him the 10% vs 10x formula and the "report to the CEO" lab structure (0:15:28), which he confirms on tape.

What he says on tape (5 Oct 2026)

  • A three-part test for a moonshot (0:01:08–0:03:04). There is a huge problem you can name; a product or service that sounds like science fiction; and a breakthrough technology that gives at least a small, testable chance. Then: find out as quickly and cheaply as possible whether it can work, so you can move on.
  • Killing ideas fast (0:03:16). X screens ideas on their techno-economics before building anything.
  • The pipeline (0:05:09). In his description, X opens 100 to 200 code-named projects a year, and a few graduate.
  • Google Brain started at X (0:07:10–0:08:14), about fifteen and a half years ago. He traces TPUs and the transformer, the "T" in ChatGPT, back to it.
  • People stay in the loop (0:28:19–0:28:41). Asked when AI spending overtakes personnel spending, he says he is not sure the work reaches 100% AI in the next decade, and that people are needed in the process for at least the next ten to twenty years.
  • "Monkey and pedestal" (0:31:04–0:31:46). If the goal is a monkey reading Shakespeare on a ten-foot pedestal, work on the monkey first. Do the hardest unknown first; the pedestal is the easy part.
  • Most projects are stopped on purpose (0:30:24). He gives the share X kills as about 98%. That is his own count, not an independent one.

Open

  • Cost of a moonshot. He says moonshots are cheaper than they were 16 years ago, giving both "about 3x" (0:20:35) and "10 to 20% a year" (0:21:25). Those two figures don't agree, so neither is repeated here as a number.
  • The transcript for this episode is machine-translated. Timestamps are right to about ten seconds. Check the audio before quoting him word for word.

Connections

Reading rule. Anything marked unverified or on-air is a claim made on the podcast and has not been checked against a primary source yet. It stays marked until it has. Sourced numbers link to where they came from.