Q&A with former Oculus CTO John Carmack about his AGI startup Keen, how to “solve” AGI, how AGI could impact the economy, AI companies' “groupthink”, and more
Dallas Innovates :
Context & Ripple Effects
Carmack's interview sits at the start of what has become a multi-year argument over what AGI even means. When he laid out Keen's plan to "solve" AGI in early 2023, he was positioning a small lab against the frontier companies he accuses of groupthink — the same consensus mindset Demis Hassabis would later push back on by refusing to call 2025 the year of AGI.
The arc since then shows the definition fracturing rather than converging: Jensen Huang now says "we've achieved AGI" (on Lex Fridman) while Anthropic's leadership treats AGI's economic arrival as gradual and uneven, with Jack Clark arguing whole sectors will be affected last. Carmack's contrarian framing was an early data point in exactly this split.
First-order effects
- Keen enters a field where the incumbents' public claims are already inconsistent — a small lab can differentiate by defining AGI narrowly and technically rather than marketing it.
Second-order effects
- The definitional fight becomes competitive strategy: Huang's "we've achieved AGI" claim and Hassabis's skepticism are rival framings that shape customer expectations and capital allocation toward whichever lab controls the term.
Third-order effects
- If the pattern holds, AGI stops being a single milestone and becomes an accounting problem — labs staff economists (as DeepMind has done) and publish sector-by-sector impact analyses to argue about distribution of gains rather than whether the threshold was crossed.
The trend: AGI is shifting from a shared engineering target into a contested benchmark whose definition each major lab now frames to its own advantage.