“Astro Teller’s AMA: Allocating Resources for Audacious Ideas” — Moonshots EP #24

Episode summary

Live audience Q&A follow-up to Teller’s EP #25 talk. Covers practical implementation of moonshot culture: the premortem technique (before Wing’s first public launch, Teller had the team write down “why this was a gut-wrenching failure” — then spent 2 months fixing those problems before launch), X’s ~2,000 killed projects (maintained in a graveyard that projects can be resurrected from), deliberate under-resourcing of teams to force creative “cheat codes” instead of brute-force solutions, why OKRs are “horrendous” for moonshots (70% achievement shouldn’t be celebrated when working on radical ideas), why equity shouldn’t be granted during the exploration phase (it creates incentives to protect darlings), and the Google Glass post-mortem (X pretended it was a product when it was still a learning platform). Audience questions from entrepreneurs and VCs on capital allocation, global talent, and collaboration.

Key arguments / segments

• [00:01:00] Premortem technique: “It’s 3 months from now, it was a gut-wrenching failure — write down why you knew”
• [00:03:00] 2,000 killed projects: maintained in a graveyard, some resurrect (Foghorn -> green hydrogen breakthrough)
• [00:07:00] OKRs vs. moonshots: OKRs are “horrendous” for moonshots — either non-negotiable deliverables or audacious experiments
• [00:10:00] No equity during exploration: stock options create incentives that kill intellectual honesty
• [00:14:00] “If it doesn’t sound crazy, we’re not interested”: obvious ideas mean everyone else will do them too
• [00:17:00] Tight learning loops: X is obsessed with evidence (not progress) within months, which makes healthcare hard
• [00:20:00] Deliberate under-resourcing: forces teams to find “cheats in the video game of life” instead of brute-forcing
• [00:25:00] Tel Aviv satellite office experiment: success metric is “how exy are they,” not what they produce
• [00:28:00] Google Glass post-mortem: “we pretended we were done when we weren’t” — the real failure was misrepresenting a learning platform as a product

Notable claims

• X has killed ~2,000 projects and maintains a graveyard; some have been resurrected as enabling technology matured
• Foghorn (seawater-to-methanol) was killed at $15/gallon equivalent; green hydrogen breakthroughs may revive it
• OKRs with 70% achievement targets actively harm moonshot work — “how can you set key results when you don’t even know if you should go over there?”
• No equity during exploration phase; equity only once >50% of exploration is complete and risk is reduced
• Teams are deliberately under-resourced to prevent brute-force approaches and force creative solutions

Bias / sponsor flags

• Teller promoting X/Alphabet’s innovation model — polished institutional narrative
• No discussion of X’s actual financial returns to Alphabet or whether the model is economically justified
• Google Glass framed as “still alive in B2B” rather than a consumer product failure
• Audience questions are friendly and reinforcing, no tough pushback

RDCO relevance

Moderate-high relevance. Several directly applicable operational principles: (1) The premortem technique is immediately usable for any RDCO initiative — force the team to write down why it failed before launch. (2) Deliberate under-resourcing to force creative solutions over brute-force approaches aligns with lean operations philosophy. (3) The “evidence not progress” distinction for tight learning loops is the right evaluation framework for AI agent experiments. (4) No equity/incentives during exploration phase to preserve intellectual honesty maps to how to structure AI evaluation — don’t reward outputs, reward learning quality. (5) The OKR critique (either non-negotiable or audacious experiment, nothing in between) is a useful framing for task prioritization. File as organizational design / innovation methodology reference.