Thomas Dohmke

And over the years with open source components and higher programming languages and things like that, we moved closer, but we never got really there until GPT-3 came.

And when OpenAI launched GPT-3 in 2020, they created a fine-tuned version of that called Codex, fine-tuned on an open source code that for the first time allowed developers to describe their ideas directly

in English and then later in German and almost any major human language.

And then the model generated the code for me.

And in 2020, that meant I could say, give me a method to determine prime numbers or do sorting algorithms.

And if the listeners have gone to university and did computer science, they know that you have to learn all these different sorting algorithms, quick sort and bubble sort and whatnot.

And then you never needed that again, because ever since you started your career in a company, you just use some library and do, you know, array.sort, right?

And so we've always moved up that ladder, but all of a sudden with AI, we were able to get so much closer to how we're actually thinking.

The language and the descriptions that we're thinking in.

So I think that's number one of why the adoption, the diffusion of these tools has been so fast in software development because it finally got us to the step where we always wanted to be.

I believe we're already past that point.

With or without AI, we passed that point a few years ago when it became clear that the majority of projects, commercial or not, are based on 90%-ish plus or minus of open source libraries, open source systems, a few

You look at what runs, you know, on a platform like Substack today, it's not only the product itself, it's all the open source libraries.

And keep in mind, those open source libraries are maintained by millions of developers around the world that are not part of, you know, the organization that sells the product, right?

That do not follow a work schedule, that do not the same trainings.

that might decide on a Sunday night to push an update to the library.

And so now all of a sudden, you effectively gave commit access to those people because they're part of your stack, all the way down to the operating system that almost nobody builds themselves.

And even if you look into the Linux, Windows, or other macOS operating systems,

They're incredibly complex.

It's very rare that a single person can navigate that whole code base, right?