#309 – John Carmack: Doom, Quake, VR, AGI, Programming, Video Games, and Rockets

Episode Summary

Executive Summary: John Carmack reflects on his path from early Apple II hacks to id, Oculus, and now AGI research, emphasizing that deep systems understanding, hard work, and user value drive real breakthroughs. He revisits the technical and organizational lessons behind Commander Keen, Wolfenstein 3D, Doom, Quake, VR, and large-company engineering, then lays out a pragmatic AGI view: likely small core insights, huge leverage, no fast takeoff, and major value from embodied or companion-like AI before full general intelligence.

Main Topics: Early programming, hacking, and the roots of Carmack’s style (Priority: 5/5): Carmack describes first learning to program on TRS-80 and Apple II systems, making games from BASIC, assembly, and clever hardware exploits. These early constraints shaped his lifelong preference for understanding systems deeply and finding faster, non-obvious implementations. Language philosophy: C, Python, Go, Rust, JavaScript (Priority: 5/5): He compares languages through the lens of simplicity, maintainability, performance, and team-scale development. He favors C for serious low-level work, appreciates Python for convenience, sees value in Go’s simplicity, is cautious about over-abstraction, and recognizes JavaScript’s enormous ecosystem power. Id Software, shareware, and the invention of modern game feel (Priority: 5/5): Carmack recounts the rise of id from Softdisk contract work to Commander Keen, Wolfenstein 3D, Doom, and Quake. He emphasizes deadlines, constraints, rapid iteration, and the hacker ethic as central to the company’s breakthroughs and shareware success. Technical breakthroughs in Wolfenstein, Doom, and Quake (Priority: 5/5): He explains scroll hacks, ray casting, compiled sprites/scalars, BSP trees, texture mapping, and the transition from 2.5D to more general 3D. He highlights how pragmatic compromises and optimized low-level code enabled immersion and user-startle effects. Work habits, discipline, and craftsmanship (Priority: 4/5): Carmack argues that long hours, consistency, and focused hard work matter for mastery, while also stressing sleep, exercise, and avoiding burnout. He sees work as a craft that benefits from prolonged dedication and deliberate optimization of one’s schedule. VR, the metaverse, and user value (Priority: 5/5): He frames VR as a route to experiences that are better inside the headset than outside, with strong near-term promise in meetings, fitness, and social presence. He argues the metaverse should emerge from compelling products people love, not capability-first abstractions. AGI, narrow AI, and the path forward (Priority: 5/5): Carmack says AGI may emerge from a small set of key insights, likely already foreshadowed in the literature, and he doubts fast takeoff. He believes real progress will come through scalable systems, long-term memory, learning loops, and practical deployments before full generality.

Key Arguments: Constraints and resource limits often force the most important innovations; many of Carmack’s breakthroughs came from asking how to make something 5-10x faster rather than simply adding more compute. Programming languages should be judged by how they help teams build maintainable, valuable systems over their full lifespan; C remains strong because it is simple, direct, and understandable, even if unsafe. Debuggers, static analyzers, asserts, and other guardrails are essential because large codebases inevitably accumulate mistakes; “write it and see” is inferior to interactive inspection. User value should be the top-level metric for engineering decisions; technical elegance matters only insofar as it improves real outcomes for users. Work longer and harder can increase both output and craftsmanship, especially for people who are deeply motivated, though it is not a universal prescription. VR’s best opportunities are experiences that exploit its strengths—presence, audio, short sessions, fitness, social proximity—rather than trying to mimic the entire physical world. The metaverse should be built bottom-up from beloved experiences with growing capabilities, not from an abstract platform spec. AGI is likely to come from a relatively small number of key insights, possibly on the order of less than six, and not from simply scaling current systems indefinitely. There will probably be no sudden “fast takeoff”; real AGI will likely be bounded by compute, data center deployment, interconnect limits, and real-world integration constraints. Embodiment is not necessary for AGI, though embodied or VR-style environments may help training and interaction; digital simulation can substitute for the physical world. A good AI future may first look like highly useful companions, collaborators, and narrow specialists before it looks like a fully general mind.

Data Points: Podcast duration: Over five hours - The conversation is described as the longest ever on the podcast Early work schedule: 60 hours/week - Carmack describes his long-term work pattern during his programming years Daily work schedule: 10-hour day, six days a week - His standard productivity cadence for decades Sleep target: 8 hours - He says he always tries to get eight hours of sleep Diet Coke intake: 8 or 9 a day - He says this is still part of his routine Running distance: About 4 miles a day - He describes daily exercise while living in Texas Commander Keen revenue: $30,000 a month - Early shareware success after launching the game Quake development count: 2 million+ lines of code - He refers to id’s codebase as a couple million lines during analysis-tool discussions Human genetic code: A couple billion base pairs - Used as a comparison to biological information density Brain encoding estimate: Maybe 50 megabytes - He speculates on brain-information scale in a discussion of biology and memory AGI code size estimate: Tens of thousands of lines - He argues the core AGI breakthrough may be much smaller than a full operating system or browser AGI insight count estimate: Less than 6 key insights - His estimate of the number of critical conceptual breakthroughs needed AGI timing estimate: 50-60% chance by 2030 of signs of life - He gives a probability estimate for credible AGI-like progress VR business resources: $10 billion/year - He cites Meta Reality Labs spending as a scale reference VR headset price/performance trajectory: 100 million headsets desired - He says ubiquity is needed for VR to replace tools like Zoom Mars bet: Less than 50% chance by 2030 - He says he bet against boots on Mars by 2030 Self-driving bet: Robo-taxis in major cities by 2030 - He bets in favor of autonomous driving reaching practical deployment View latency threshold: 50 milliseconds feels amazing - He contrasts user experience response times

Pivotal Quotes: "Focused, hard work is the real key to success. Keep your eyes on the goal, and just keep taking the next step towards completing it." — John Carmack: Final closing advice to listeners "The thing that will define the future still requires you to operate at the limits of the current system." — Lex Friedman / Carmack discussion: A central idea about innovation, VR, and systems engineering "If you've got thousands of GPUs necessary to run these things, it's going to be kind of expensive... and then you're kind of like, well, I should have an entourage of AIs that are following me around." — John Carmack: His vision of how AGI may first be deployed and then become affordable

Implications: For builders, the message is to optimize for real user value, understand the full stack, and use constraints as a creative force. For AI, expect incremental but powerful progress, likely from small core breakthroughs, with companionship, collaboration, and narrow automation arriving before full AGI.