When AI Understands Hardware & Raises the Stakes for Cybersecurity

I know that you've got a really cool series of interviews on this episode.

Can you give us a little preview before folks get into it?

Yeah, one of our interviews that we actually spoke about on the podcast back in March after Embedded World when we were recapping everything and kind of just talking about some of the favorite things that we saw.

And one of them that I mentioned was Embedder and how great my meeting went.

Kind of just speaking in general to younger engineers who are, they have a really cool solution that, that I spoke to them about at embedded world and finally got them on the podcast.

And I spoke with their CEO, Ethan Gibbs and yeah, just their AI firmware engineering platform.

And it's, it's called Embedder.

It's also the name of the company, but it's a hardware aware coding agent, you know, designed to, I guess, cut the time it takes for engineers, you know, who have to spend time reading data sheets and writing code.

And I feel like we talk so much about, you know, AI, the negatives and the positives, but I thought this was pretty, a pretty positive solution.

And I think you felt the same.

Yeah.

Yeah, absolutely.

I mean, it was, um, nominated for the Embedded Award at Embedded World this past March in Nuremberg.

And the thing I found most interesting about it and was really impressed by was it was a really good example of what I think is a really good trend of eschewing the large data lake and the large model for specifically trained tools.

So in this case, Embedder is trained on the specific documentation, data sheets, handbooks, everything for whatever your platform is that you're building on, your hardware, your software, whatever.

And then you use that model that is only trained on the thing that you're working on to help assist you in design and code.

And because it's so specifically trained, it's purpose-built.

You avoid things.

You avoid a huge percentage of errors.

Not that you don't still have to double check, obviously.

You avoid, you drastically decrease the chance of hallucination.

You just allow yourself to work more efficiently and more accurately.

Speed's time to market.

And it's really smart design.

And that's what I'm seeing now at shows that I've been going to ever since Embedded World, where the AI tools, especially edge AI tools, but not just there, are being used sort of in a way that I call practical AI, where instead of being worried about it being able to do everything poorly, they're using AI tools to do one specific thing or set of things really, really well.