Andy Halliday

So you can, in effect, create a project within AI Studio that builds an application.

You built a gem, which is a different application.

platform, basically.

But you could build an application that does something very similar to what Brian just did using the same prompt.

But you would have an application that could actually be mounted and posted to the web, basically published to the web in AI Studio.

I think that's the main distinction that I would make.

is that you have projects and applications that are going to be in a library in AI Studio.

Its focus being to build applications, whereas in Gemini.Google, you have Canvas, which will present something very sophisticated, including very great visualizations, because Gemini 3 is probably the leading multimodal model out there.

Right.

Yeah.

Not just best reasoning, but also native audio, native video.

It does all those things and it can generate infographics right in the canvas or give you an HTML version of it.

But if you wanted to make that part of an application, you'd probably shift over from Gemini dot Google dot com over to Google AI studio.

I'm going to spin off the reference to Prometheus and just explain a little bit what that is about in the context of what world models are as a major...

development direction in AI.

Project Prometheus, which Jeff Bezos is now kind of coming out of quasi multi-billionaire retirement to be the CEO of, it's already instantiated.

It has 100 employees that they've kind of poached away from multiple AI labs.

And it's focused on physical AI.

Now you can imagine that Amazon would benefit largely from this kind of development.

When we talk about world models and the physics component of AI, we're moving beyond LLMs and the transformer models and the way that they deal with text and language, multiple languages and fluency across all of them.