Carmack, John
▶ Watch on YouTubeCoverage note: full (over 5 hours — the longest episode in the podcast's run at the time)
Core ideas
Carmack traces his path from a self-taught childhood obsessive fascination with computers (devouring library books, teaching himself scrolling hacks on the Apple II by exploiting the overlap between text and graphics memory) through contract game programming in assembly and Pascal, to landing at Softdisk in Shreveport, Louisiana, where meeting fellow programmers John Romero and Lane Roath — the first people he'd encountered who "knew more cool stuff" than he did — catalyzed the founding of id Software. He describes the company's formative Gamers Edge period as a rapid-iteration "pressure cooker" that built the team's core skills through relentless, deadline-driven monthly game production, likening it to the Beatles' unglamorous years playing clubs before fame. He walks through his key technical inventions in granular detail: "adaptive tile refresh" and a wraparound EGA memory-addressing trick that enabled smooth scrolling (leading to the all-night Dangerous Dave demo that convinced the team to found id Software and pursue shareware distribution via Scott Miller at Apogee); Wolfenstein 3D's shift to a small, tightly hand-optimized assembly raycasting engine after an earlier, buggier polygon-based approach in Catacombs 3D proved unreliable; Doom's introduction of binary space partitioning (BSP) trees to solve rendering-order and the "Epsilon problem" of numerical precision with angled geometry; and Quake's leap to a true six-degree-of-freedom 3D engine with its own scripting language (Quake C), client-server networking, and — in collaboration with hired assembly expert Michael Abrash — heavily hand-optimized rasterization. Reflecting on Quake, he says his biggest process regret was not splitting its ambitions (new rendering tech, scripting, and networking) across two separate year-long releases rather than one grueling combined effort.
On programming philosophy, Carmack argues that value to the end user, not technical elegance or personal craftsmanship pride, should be the top-level metric guiding engineering decisions, while cautioning that this shouldn't collapse into blind metric-chasing over real human judgment. He argues working long hours (he sustained roughly 60-hour weeks for decades, always preserving 8 hours of sleep) doesn't just produce more output but genuinely tempers skill over time, while acknowledging this doesn't generalize to everyone and that some people burn out where he never has. He's a strong advocate for debuggers and static analysis tools as essential companions to serious programming (recounting how running a full suite of static analyzers over id's "reliable" codebase was humbling), prefers asserts and fixed-size arrays with deliberate limits over unconstrained "just make it generic" designs, and values C++ for encouraging both creative freedom and engineering discipline, while respecting Pascal, C, and even Go and JavaScript for their own tradeoffs — repeatedly returning to the idea that resource constraints (assembly-era memory limits, shareware deadlines) tend to produce better craftsmanship than unlimited resources.
On the hacker ethic, Carmack credits Steven Levy's book Hackers as formative, describing his choice to release id's full source code for its major titles as an attempt to embody a non-zero-sum, credit-agnostic culture of sharing — contrasting this with what he sees as a more possessive, artist-like mindset common among game modders — and reflects candidly on pushing John Romero out of id Software as a young, immature decision rooted in a flawed equal-partnership corporate structure, while affirming deep respect for Romero's design talent and expressing warmth toward former colleagues like Tom Hall.
On virtual reality, Carmack (CTO of Oculus VR from 2013, following its 2014 acquisition by Facebook/Meta) describes his guiding philosophy as pursuing "value gradient" improvements — making concrete, immediately valuable experiences better step by step — rather than chasing an all-at-once "Ready Player One" vision, citing Beat Saber's unpredicted success as a case study in exploiting VR's actual strengths (natural hand-holding, immunity to motion discomfort, short session lengths, built-in fitness value) rather than its assumed strengths (he'd expected first-person shooters, like his own Doom 3 VR demo, to dominate). He argues VR experiences should aim to be better than the outside world for the wearer rather than being dismissed as escapist, and sees near-term, achievable value in small-group virtual meetings (arguing this is more tractable than a full open metaverse), while criticizing large tech companies — including his own — for using vastly more resources (citing Meta's ~$10 billion/year Reality Labs spending) than the value produced would suggest is efficient.
Discussing his friendship with Elon Musk (formed during Carmack's Armadillo Aerospace years), he describes shared technical sensibilities and defends Musk against the "just a business guy" characterization, citing detailed personal engagement with propellant chemistry and welds; he separately describes a $100,000 bet that humans will not land on Mars by 2030 (distinguishing it from a bet on self-driving cars reaching Level 5 by 2030, which he thinks is more likely, since it depends on an entire industry rather than one person's focus).
On AGI — his current primary focus, backed by a newly signed investment term sheet after a period of self-funded "gentleman scientist" research — Carmack argues the core remaining insights are likely simple enough to fit in tens of thousands of lines of code (not millions), estimates a 50-60% chance of "signs of life" of AGI by 2030, and explicitly rejects the idea that embodiment in the physical world is necessary, arguing his own background in simulated virtual worlds supports building and training AGI in real-time simulated/VR environments instead. He predicts no "fast takeoff" scenario, reasoning that any credible AGI system will require large, physically real data centers (not a self-propagating "escape onto the internet") and will be extremely expensive to run at first (potentially $1,000/hour), becoming affordable only after further hardware cost reductions. He describes consciousness as likely just "the way the brain feels when it's operating" — a emergent, associative, stochastic process rather than something mysterious or symbolic — and doesn't believe mortality or fear of death is a necessary ingredient for machine intelligence, though he considers the broader question of what "reward" structure would substitute for it an open one.
He closes with reflections on martial arts (Judo and Jiu-Jitsu) as a source of humbling physical grounding, career advice emphasizing "weaponized curiosity" and deep, cross-layer understanding of systems as more valuable than a fixed goal-driven path, and — echoing his answer on AGI motivation — states he doesn't dwell on mortality or search for grand meaning, describing his life's throughline as a kind of personal "gradient descent": following local, concrete opportunities for impact rather than being driven by any overarching vision.
Excerpts
On the moment that convinced the team 3D games would be transformative:
“He had drawn these characters and he was slowly moving around... picked up a key, go to a wall, the wall disappears in a little animation and there's a monster right there, and he practically fell out of his chair... something in the back of your brain, some reptile brain thing, is just going 'oh, something just happened' — that was one of those early points where it's like, yeah, this is going to make a difference.”
On why he believes AGI doesn't require physical embodiment:
“Anybody that is all-in on the embodied aspect of it, they are tying a huge weight to their ankles... reality is, in so many ways, everything fails, Murphy's Law, even if you've done it right before, your fifth one might come out differently... I think you can make a simulated environment that provides much of the value of what the real environment does.”
On his personal philosophy of meaning, in contrast to grand-vision thinking:
“It's been my experience that people that don't focus on the here and now right in front of them tend to be less effective... it's the magic of gradient descent — people just don't believe that just looking locally gets you to all of these spectacular things.”