Nov 5, 2026 · 26 min · 9 segments
Jacob Beningo explores the shift from using AI as a coding assistant to agentic firmware development, where AI agents autonomously build, debug, test, and verify embedded systems. He shares a real…
Jacob BeningoHost


Instead, what I'm finding is that they are using AI as coding assistance, which is great.

It's great when you're writing code, if AI can suggest a couple of lines of code for you to put in there.

Or if you have a problem, you can go over to AI and say, hey, can you help me fix this device tree? Or can you help me solve this strange issue that I'm having? That's great.


Not where it's just generating code for you for a small part of your system, but where it's actually building, debugging, testing, and verifying the firmware for you.

I'm talking about you as a developer moving out of the role of firmware engineer and into the role of an embedded systems engineer, where you're not just focusing on a small piece of the system, but moving out into the larger realm of systems engineering.

And when we do that, what you're going to find is that there's a lot of not just cool things that you can do, But there are pieces of the development cycle that need to be automated.


It's really when you put together a good model and you start to use it over and over to work on similar types of activities on a daily basis and build out your workflow that AI is really going to become a game changer for you and for your team.

Now, over the last several months, I've been working on all kinds of interesting ways of working with AI and applying them to developing embedded systems.

And oftentimes, when I do talk to people, they'll tell me, oh, well, AI doesn't know enough about the hardware to be useful.

Or it hasn't been trained on firmware, so it doesn't really know what it's doing.

And what I'm here to tell you is that that's not the experience that I've had, okay? What I have found is that when you connect an AI agent to a good workflow, you can not just get some code written.

You can actually have it iterate through, write good code, develop the tests for it, move through an entire validation process that starts with requirements and ends with code that's actually production ready.

Now, I know that, again, For many of you who are just using AI as a coding assistant, this sounds like one of those pie-in-the-sky type of ideas, right? And I'm not advocating that you get a list of requirements, push a magic button, and on the other side out comes a widget, okay? That's not the type of agentic firmware development we're talking about here.

What we're talking about is developing a robust workflow that allows you to provide AI with work that it needs to do that maybe goes into a work queue, that is something that you do periodically over and over again, that allows to actually go out and do the work for you, maybe while you're at lunch, or maybe when you leave your desk and you go home for the evening to come home and discover that some research activity you needed done is completed, or that some module that you were maybe struggling with or just had to get done and that you were under the gun for is suddenly in a state where you can actually use it and deliver it, okay? So this is the type of agentic firmware that we are talking about here.



Instead, what I'm finding is that they are using AI as coding assistance, which is great.

It's great when you're writing code, if AI can suggest a couple of lines of code for you to put in there.

Or if you have a problem, you can go over to AI and say, hey, can you help me fix this device tree? Or can you help me solve this strange issue that I'm having? That's great.


Not where it's just generating code for you for a small part of your system, but where it's actually building, debugging, testing, and verifying the firmware for you.

I'm talking about you as a developer moving out of the role of firmware engineer and into the role of an embedded systems engineer, where you're not just focusing on a small piece of the system, but moving out into the larger realm of systems engineering.

And when we do that, what you're going to find is that there's a lot of not just cool things that you can do, But there are pieces of the development cycle that need to be automated.


It's really when you put together a good model and you start to use it over and over to work on similar types of activities on a daily basis and build out your workflow that AI is really going to become a game changer for you and for your team.

Now, over the last several months, I've been working on all kinds of interesting ways of working with AI and applying them to developing embedded systems.

And oftentimes, when I do talk to people, they'll tell me, oh, well, AI doesn't know enough about the hardware to be useful.

Or it hasn't been trained on firmware, so it doesn't really know what it's doing.

And what I'm here to tell you is that that's not the experience that I've had, okay? What I have found is that when you connect an AI agent to a good workflow, you can not just get some code written.

You can actually have it iterate through, write good code, develop the tests for it, move through an entire validation process that starts with requirements and ends with code that's actually production ready.

Now, I know that, again, For many of you who are just using AI as a coding assistant, this sounds like one of those pie-in-the-sky type of ideas, right? And I'm not advocating that you get a list of requirements, push a magic button, and on the other side out comes a widget, okay? That's not the type of agentic firmware development we're talking about here.

What we're talking about is developing a robust workflow that allows you to provide AI with work that it needs to do that maybe goes into a work queue, that is something that you do periodically over and over again, that allows to actually go out and do the work for you, maybe while you're at lunch, or maybe when you leave your desk and you go home for the evening to come home and discover that some research activity you needed done is completed, or that some module that you were maybe struggling with or just had to get done and that you were under the gun for is suddenly in a state where you can actually use it and deliver it, okay? So this is the type of agentic firmware that we are talking about here.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.