John Carmack defends Microsoft's generative Quake II demo

When an AI tries to generate a legendary 1997 shooter frame by frame, is it a technical triumph or an insult to the people who build games for a living?

Last Friday, Microsoft researchers unveiled WHAMM (World and Human Action MaskGIT Model), an experimental playable tech demo that ditches traditional rendering pipelines to simulate Quake II in real time using a generative world model. Microsoft was quick to manage expectations, explicitly noting that the demo was never intended to fully replicate the experience of playing the original title.

Still, the gaming community pushed back hard. An X user named Quake Dad slammed the demo as "disgusting," arguing that generative games spit on developer labor and risk erasing jobs across an industry already bruised by widespread layoffs.

Carmack and Sweeney step into the arena

On Monday, id Software co-founder and Quake co-creator John Carmack stepped in to defend Microsoft's demo, calling it an "impressive research work."

Responding directly to the backlash, Carmack argued that critics misunderstand what the demo actually represents. He pushed back on fears that AI is cheapening the hard-earned skills of coders and artists, framing generative models as the next logical chapter in computing history.

Carmack pointed out that when he built his earliest games, he had to hand-assemble machine code and manually translate graph-paper drawings into hexadecimal digits. Software progress eventually made those painstaking tasks as obsolete as repairing chariot wheels. In his view, building power tools has always been the entire point of computer science.

Epic Games CEO Tim Sweeney echoed that stance shortly after. Sweeney noted that AI will ultimately become just another staple in every creator's toolkit, comparing its trajectory to the arrival of high-level programming languages, early paint software, and visual scripting.

Why it matters

We think Carmack and Sweeney are diagnosing the technology accurately, but the clash reveals a massive perception gap between computer science pioneers and working developers today.

For creators and indie studios, WHAMM is far from a functional production engine. It lacks deterministic physics and fine-grained logic, meaning nobody is building a shippable game with it anytime soon. You do not need to drop traditional game design skills to chase raw world models today.

What this research does change is the long-term horizon for real-time rendering. Replacing complex graphics pipelines, physics engines, and asset streaming with a unified predictive model could eventually upend how interactive worlds are rendered. It threatens the traditional asset pipeline far more than it threatens high-level creative vision.

That said, Carmack's chariot-wheel analogy glosses over a raw reality. When high-level languages emerged, the software industry was rapidly expanding to absorb displaced coders. Today, generative demos land in the middle of severe studio contractions and hiring freezes. While tech leaders see an inevitable power tool, rank-and-file developers see an existential threat to an already precarious livelihood.