Short answer: Ryan Aytay says AI-generated code is mostly correct but must be perfect, covering every requirement, security scenario and hardware assumption, then physically tested. Code Metal translates, verifies and optimizes, using test generation and fuzzing to prove it matches the original intent.
Watch this answer from Episode 48 (starts at 7:21) · Episode 48 notes and transcript
From Episode 48: The Code Running Your Plane Was Written Decades Ago. This Startup Proves It Still Works., starting at 7:21.
Ryan Aytay [7:20]: Well, the reason why sometimes it will fail or why it will fail in general is there's a gap. And that gap is really what we try to solve. But, the gap exists because, AI generated code is mostly correct. Right? We know this, and you can actually go in and ask, said AI platform, like, can you do it in an accurate way? Well, pretty much. But we need it must be correct. Right? It has to be perfect. It has to consider every requirement. It has to make sure that it tests every possible issue or security scenario or hardware assumption, and then it has to be physically tested. Right? And so this is a lot of, like, what we do at CodeMetal. And what I like to think about it is kinda like a promise. It's like, okay. So we don't just do code generation. That in some ways is getting easier and easier. We are all about, we translate, we verify, and we also optimize to make sure that we're doing it in the most efficient way because we know that in this world, there are certain systems that just can't afford to fail. It's kinda like if I'm trying to, change the electrical in a city, like, can't just do it I don't want the lights to go out, but I want it to change. Like, these are the scenarios you have to think about.
Ryan Aytay [8:29]: And so, just because it's always fun to show a little bit of, like, what does this look like? This is a little picture into something that we actually haven't even fully brought out yet, but we will, which is, bit of our own platform, which is, code in, code out. This is a bit of our test generation suite of how things look. And if I think about, like, where do these things go from here, a lot of the things that we talk about at CodeMetal are formal methods, and within that, there's different types of testing. There's, can call it testing and fuzzing and, go down the list of things that we believe are very important to make sure that we're able to look back at what was the intent and the behavior and the origin of the code and how do we use that as a way to kind of go through and do a proof to say like, is this right? Does this work? Will it run on said hardware? So we have this belief, right, where, generating code actually, the price is going down. It doesn't really cost anything, but if you get it wrong, it's pretty substantial. And so that's a bit of where we are.
Ryan Aytay [9:30]: Sorry. That's a very long answer about like what is CodeMetal, but really I just it's a it's it's a very nuanced thing to talk about, but very important and most people don't realize, especially I didn't realize before I joined here is how critical this really is. And I think AI helps us, but it doesn't solve every single problem.
Jordan Metzner [9:51]: Okay. Awesome intro. I've been building software and hardware for years and have tried to stay away from mission critical softwares. And that's always been, hey, guys, nothing we're doing here. We're not flying planes, you know. It's always been kind of a safety net for building, web SaaS and mobile apps and things like that. But, when you have to pay the brass tax of, human life is on the line, obviously, the requirements and the seriousness of the product and software and hardware go to another level, which, historically in my career, I've tried to stay away from because obviously this is a highly risky area and, it takes a lot of time to make sure it's it's gonna go right. And I think on top of that, you've got a lot of historical systems written in legacy code bases that, many engineers don't even know anymore. Right? And they have to kinda like go back to learn. And probably they'd never, but without AI, they probably would never wouldn't, have the ability to be rewritten and, more efficiently and run on, lower cost hardware and things like that. So tell us a little bit, obviously, we see like, huge major, defense contractors on your website and obviously, we don't wanna discuss anything confidential. But, tell us, like, a good use case of, where a company came in and, they applied CodeMetal and they were able to kind of, take old legacy systems and implement, them in a new way.
