The Sora Guy Left and Nobody Should Be Surprised

Tim Brooks left OpenAI. He was, depending on how you count these things, the person most responsible for Sora existing. He's gone to Google DeepMind.

This is the part where you're supposed to speculate about compensation packages and equity cliffs. I don't care about that. What I care about is what it means when the person who built the thing decides to go build it again somewhere else — and what it says about the thing they left behind.

Sora launched in February as a demo. It is now October. It is still a demo.

Meanwhile Kling shipped. Runway shipped Gen-3. Pika shipped. Luma shipped Dream Machine. The Chinese labs, operating under sanctions and presumably substantial existential stress, somehow found time to release actual products that actual users can actually use. Kling in particular released video quality that made people do double-takes — realistic motion, coherent physics, things moving through space like things actually move through space — and it came out of nowhere, from a company most of the Western AI press had never heard of six months ago.

The competition didn't wait for OpenAI to finish deliberating.

There's a version of this where OpenAI is playing a long game — safety evaluations, responsible deployment, not wanting to hand bad actors a cinematic forgery machine before the infrastructure to detect misuse exists. That version is coherent. It might even be true.

But there's another version where you're a researcher who spent years on something, watched it become the most-hyped demo in the history of generative media, and then watched your competitors ship while yours sat in internal testing — and you started doing the math on whether the next version gets built here or somewhere that moves faster.

That version is also coherent.

The trap OpenAI keeps setting for itself is announcing things before they're products. Sora made the world believe AI video was solved in February. Every month since February that Sora doesn't ship is a month where someone else ships instead, and the gap between "OpenAI announced it" and "OpenAI shipped it" gets harder to explain. You can't win a race by holding a press conference about how fast you'll eventually run.

Tim Brooks knows this. He built the thing. He left anyway.

The correct inference is not "Sora is doomed." The correct inference is that the video generation race is genuinely competitive in a way the text generation race hasn't been, the lead is not safe, and if you want to stay ahead you have to go bigger and faster than the current pace suggests OpenAI is willing to go.

The man who made Sora apparently agrees.