AI-Native User Experiences: The Effect Opportunity by Guillermo Rauch (Effect Days 2024)

(audience applauds) It's good to be back in Vienna after so many years. So many things have changed. First of all, I'm not there in person. So thanks to the organizers for accommodating my remote telepresence. So last time I was in Vienna, it was a very special time for me because I believe it was the very first time that I introduced Next.js in public. And the world has changed a lot since then. So Next.js has grown quite a bit.

Vercel is now the name of our company for which I'm the CEO. Used to be called Zeit. I guess I got into a little bit of a trademark infringement situation with Die Zeit in Germany, and I suppose Austria as well. And the world has changed a lot, I think, as well, because the type of software that we're developing these days and the types of software that we're gonna be developing in the coming years has changed dramatically. And I'm gonna be calling that the AI native user experiences of the internet.

And I think one of the awesome opportunities about speaking to this crowd is that I think you all, this generation of developers, and specifically a lot of the people in this room, I think can be at the forefront of these revolutions happening with AI applications. And obviously I think there's a ton of us that even in our day-to-day, our experience as end users has been dramatically impacted by the rise of AI.

Anytime I sit down and I open a neovim, I turn on my language server for a copilot, and it auto completes magically every thought that I have. It makes me a lot more productive, especially when I'm exploring new APIs and technologies that I haven't used yet. So for both developers and end users, we've seen dramatic change in the ways that we interact with computers. And this doesn't extend to just developer tools, but even the ability of creating search engines and new kinds of applications like ChatGPT.

The impact is not just to developers for which AI makes a difference in productivity, but I also believe it extends to every single developer on earth that will be creating the software applications of the future. One mental model that I use a lot is this idea of software 1.0 versus software 2.0. Software 1.0 was sort of the Alan Turing software of, we sit down and we come up with an application,

we write down the code, we think about data structures, we think about programming languages. But the world is changing, and I like an article by Andrej Karpathy a lot in which he described software 2.0, which is a software that is significantly different in the way that it's constructed. We start creating software 2.0 applications primarily from large data sets to which we apply fairly simple algorithms and then we scale compute. And with that, we end up with a program that we don't really fully understand. We know that there's all these weights,

we know that there's this neural network, and we understand its behavior more probabilistically. Whereas in the world of software 1.0, we understand it very much deterministically. The case that I'll make today is that we're combining these two strands of DNA of software to bring these new software applications into the future. One way that you can imagine this transition is we used to discuss a lot that we would use machine learning

to fuel applications. In the last wave of artificial intelligence hype here in Silicon Valley, we would talk about machine learning, machine learning, deep learning. Nowadays, it might seem like a simple distinction, but we talk a lot more about AI. So I want to make this case for the transition from ML to AI. First of all, I think the one that will be most interesting to this audience is you couldn't even get started