So getting rid of pointer errors in C or C++ in favor of managed memory in Java or Kotlin, for example.

Yeah, for example, I think the easier example will be... type systems where, you know, you can say like, oh, but I know that here, this object is actually my specific class and I want to call this method on it, but the compiler doesn't know that.

And you have to like swear on your life and do an explicit cast and just make it go away.

And this like in Kotlin, for example, we deliberately made it verbose to do a type cast because you shouldn't do that.

Because generally, it makes your program less safe and less predictable.

Although you may really know very well that in this particular case, it's absolutely safe and guaranteed and so on and so forth.

So generally, we've been adding those constraints that people can rely on.

So

So I would say let's start with setting the foundation here and what is even a spec? And if you have an example, that would be great.

Yeah, so just a little bit of context of why exactly we're talking about specs at all.

So CodeSpeak is one of the tools in the very vague space of quote-unquote spec-driven development, which basically means we're trying to use human language, natural language to make programs.

And that's... you know, the space is pretty wide.

A lot of people are using the same terms very loosely.

So if we talk about specs here, we'll be probably finding associations from very different spaces.

And like when a lot of people hear the word spec and hear that, you know, we are now using specs instead of code to describe our programs, their first association is spec is something very serious.

And how is that problem amplified and what exactly we are solving here? Yeah, so in terms of how many models are involved, I

Plus, a one-line observation about the next five years that genuinely stopped me cold.

Actually, when we were starting, I wrote down this formula that CodeSpeak equals software engineering minus writing code.

So we want to keep all the engineering aspects of it, but, you know, we, we of course see that humans shouldn't be writing code manually anymore.

So [clears throat] this idea with intent recovery is pretty fundamental because right now everybody who prompts agents to get working code, they're doing work that is being, like, partly accepted and translated into the code, but the rest of it is being discarded.

And there is this kind of an unfair situation where, you know, you're talking to your agent in English or, like, in a natural language anyway, right? And then you get code, and you check this code into a repo, and if you're working in a team, other people check your code out, but not the human language, the code, right? So you're talking to a machine in a human language, but talking to your colleagues on the team in machine language [chuckles] That makes not very much sense.

So it's very obvious that there has to be a next level where we all talk in, like, reasonably high-level language, which is close to human language at least.

And here, like, the simple observation behind what we're doing right now at CodeSpeak is that you already wrote these words down.

How can that be addressed?