I wrote for Naavik about Project Genie and what I see as the stock market’s category error in reacting to it. Read the full article here.

Why I wrote about this

When Google previewed Project Genie, Unity’s stock dropped more than 20% and several game companies followed. The reaction implied that a new AI system had just made game engines obsolete. That didn’t match what I saw in the demo, so I wanted to unpack what Genie actually is and what it isn’t.

The gist of it

Project Genie is an experimental “world model” that generates short, navigable 3D scenes from prompts or images. It feels magical at first glance: type something in, and you can move around a generated space. But under the hood, it’s closer to a probabilistic video stream than a game engine. Worlds only last about a minute, run at modest fidelity, and break down under closer inspection. Objects shift, geometry drifts, and the system “forgets” prior states.

That distinction matters. Games are not defined by having a 3D world; they’re defined by rules, determinism, and consistency. Players care that actions have predictable outcomes and that progress sticks. World models, by contrast, generate approximations of what should happen next. There’s no obvious path from that to a stable, monetizable, live-operated game.

Game engines and world models solve different problems. Engines provide deterministic simulation, performance, tooling, and production pipelines. World models offer fast, approximate generation. The more realistic near-term outcome is complementarity: world models helping with prototyping, mood setting, or rough layout generation, while engines remain the execution layer.

The broader lesson is familiar. Since ChatGPT’s launch, the biggest impact of AI in games hasn’t been end-to-end automation. It’s been thousands of small productivity gains across coding, art, design, and marketing. The tools that stick tend to be boring and practical. Genie, at least for now, is an impressive technical demo, not a replacement for Unreal or Unity.

Key takeaways

  • The market reaction to Genie treated a world model as if it were a game engine, although they are fundamentally different categories.
  • Games depend on deterministic rules and consistency; world models generate probabilistic approximations.
  • Unity and Unreal are far more likely to integrate world-model-style AI than be replaced by it.
  • Near-term value from world models is additive: prototyping, ideation, and creator tools.
  • The most valuable use cases for world models are probably found outside gaming.