Agentic AI for Engineers & How MATLAB & Simulink Workflows Are Changing - Seth DeLand | Podcast #171

How do you think engineering has shifted from like the traditional way we do engineering now with AI? To me, AI, like a decade ago, relative to what it is today, it really was quite niche. AI was a specialized tool. Only a certain number of engineers kind of knew how to work with it.

Maybe they were using it on applications like developing a virtual sensor. And so there were good applications of it, but it was not something that every engineer had to or could reach for every day. How are you going to evaluate the success of the use of the tool? You need to have a good benchmark or a good way of measuring what success is going to look like before you start an evaluation.

Today, we'll talk about the future of engineers 2026 and beyond, the role of, as we all know, agentic AI, whatever that means, workflow, and the role of MathWorks in the topic of agentic AI. But before we get started, Seth, can you give us maybe a little bit of an introduction? Who are you to begin with and what is your role at MathWorks? Yeah, so my name is Seth DeLand.

I'm a product manager at MathWorks. I focus specifically on generative AI. I've been at MathWorks for just over 15 years in a number of different roles, starting from tech support and on the front lines, helping customers with questions about MATLAB and Simulink to different roles in the statistics optimization, machine learning, and now generative AI space. My background is in mechanical engineering, though, where I worked on diesel after treatment combustion and or diesel combustion after treatment and reducing NOx emissions.

Super funny. I think we have a similar background because actually one of my internships in uni in Germany was exactly the same topic, which is funny. Now the question of the more important question, how do you go from combustion engines to Gen AI? How did that happen?

Yeah, so a lot of what we were trying to do was working specifically on the SCR, which is selective catalytic reduction. It's one of the components in the after treatment system. And the chemical kinetics that happen in that specific component are quite complicated. but from a mechanical engineering controls point of view, we're trying to figure out how much of the urea or the ad blue, or depending on what country you're in, it's called something different, but needs to be injected into the exhaust stream to essentially react away all of the knocks to try and clean up the emissions before they come out the tailpipe.

And so my role was to build a mathematical model of that whole system so that we could use that as a plant model for developing the control strategy. And so it was a lot of mathematical modeling, parameter estimation, optimization, using techniques from simple gradient descent through much more complicated global optimization techniques, genetic algorithms, and different optimization techniques to try and get a good model that we could use to develop the control strategy so you can start to see where it's starting to lead into some of the math statistics optimization space and that's where i really found my passion so it's where i've i've made my career at math works i see interesting and maybe a more general question to get started with the topic of ai maybe we don't have to specifically now cover ai but how do you think going back like maybe also looking at your colleagues back then how do you think engineering has shifted from like the traditional way we do engineering now with AI.

How do you think the role of an engineer has changed in the past, let's say decade? Yeah, I think it's a really great question. And to me, AI, like a decade ago, like I don't want to call it too niche, but relative to what it is today, it really was quite niche. You know, AI was a specialized tool, only a certain number of engineers kind of knew how to work with it.

You know, maybe they were using it on applications like developing a virtual sensor for estimating some, you know, some sense, some signal on a vehicle, you know, so they didn't have to use another physical sensor. And so there were good applications of it, but it was, it was not something that every engineer had to or could reach for every day. It was, again, it was more of a specialized tool. You know, been clearly a big push, I think, to try and use more of those like machine learning and deep learning capabilities as part of engineered systems.

But it wasn't an everyday thing. If I contrast that with where things are today and I think where we're heading, generative AI, agentic AI, these are horizontal tools. These are things that I think will have an impact similar to the use of statistics in engineering or the use of math and engineering. They fundamentally change a lot of the core workflows.

They're helpful for everything from thinking about requirements to writing code, to testing and verifying and validating code. And so I think it's shifted from kind of, there were pockets of people that knew how to use it and were using it effectively to now, Now, it's almost like a completely different thing. And it's really impacting, you know, things at a much more fundamental level. From an engineering point of view, because now AI has been commoditized to some extent, right, and democratized.

Do you think that's always a good thing or are there also some negative sides to it? Yeah, that's a good question. I mean, I think it's a good thing in general. I think, you know, as an engineer, I always want to have as many tools in my toolbox as I possibly can.

And so I like obtaining or learning how to use new tools, right? And so, like, in my toolbox, I kind of mean my hypothetical toolbox of, like, you know, do I reach for an optimization algorithm for this problem? Or do I reach for some signal processing technique? Or, you know, do I need, you know, I'm going to prototype something.

Do I need to reach for an Arduino board to do that? Like all these different technologies and tools that I like to be aware of so that if I am working on a new problem, I can, you know, quickly assess what tools to pull in. And I think that democratization really is helpful for, again, adding this as a tool to, you know, for other engineers. It's another thing they can learn.

It's another thing that they can reach for when they encounter a problem. So I view it much more as a positive. Before we move forward in terms of the topic of agentic AI, let us first define what is agentic AI to begin with and why is it so important, especially like in 2026 and beyond, especially for the engineers maybe listening to this podcast? Yeah, so I think maybe before I define agentic AI, I'll start with large language models.

And so large language models, these are the AI models behind popular tools like ChatGPT. They're trained on incredibly large corpuses of text from the internet and other sources. And those models themselves are incredibly powerful. they can you know respond almost as if you're talking to a human they can generate text they can generate code some of them can even you know work with images and so those large language models themselves are really powerful but there was there's kind of this limitation where if you're if you're having a conversation with with just a large language model it can reply with text but it can't actually do anything on your behalf.

And so when we talk about agentic AI now, agentic AI, the real difference is this notion of tools. And you can think of these as like APIs that the AI can actually call to accomplish a task. So rather than it just generating a bunch of text, it's actually going to generate calls to APIs. And those APIs could be something simple like create a file, you know, edit a file, something like that, to something much more complicated, like, you know, refactoring code or even, you know, sending an email or maybe even booking a flight for me for my next trip.

Like, and so once you start to give these AI capabilities, these tools that allow them to evaluate things against APIs, now they're not limited just to generate text as an output, but they can actually start to work on the same artifacts that, you know, we humans are working on when we're, when we're working on engineering problems. How would you maybe set a question for you that I have, which is, how would you see the difference between the LLM based agentic AI that you just talked about and defined versus the classical AI that we've mentioned previously, right?

The classical gradient descent approach. When we all got started with machine learning, I mean, a couple of years back, we all started with kind of gradient descent, classical machine learning models, right? Optimization models, et cetera, et cetera. Where do you see the benefit and maybe the disadvantages compared to LLMs and the kind of this classical, quote-unquote, ML models that we know?

Yeah, so I think the classical techniques, really the focus there was always on training models because most of the models that we were using back then, machine learning, deep learning, the classical techniques, They are custom trained for a specific problem. So if I'm an engineer and I'm using one of those technologies, my workflow is going to look like I need to focus on preparing a data set. I need to clean that data set and get it ready. I need to start training some different AI models.

Maybe I need to do things like feature extraction to improve the quality of those models. I need to validate the model and understand how well it's going to work, you know, both on the training data. but then maybe I need to simulate it as part of my larger system to understand how it behaves. And so the workflow that I as an engineer go through was very much a machine learning data science type of workflow.

Now, if I contrast that with generative AI, large language models, agentic AI, like I'm kind of using all of those interchangeably right now, the workflow is very different. these models have already been trained for me. The tools are already there. I'm most likely just going to be using some off-the-shelf large language model that somebody else has already trained for me.

And so rather than me focusing on how do I bring this model and use it as part of my system it more how can this AI help me as I developing that system So it it really almost at a different um level or paradigm where it you know it i not going to have to worry about training it i not gonna have to worry about how it's created i'm just asking it for assistance along the way and it's it's able to do that based on you know and it's doing that based on my knowledge of the problem and the instructions that i give it yeah i feel like 2026 is definitely an inflection point especially for engineers.

I mean, look at OpenClaw, CloudBot, and everything that's been published in the last couple of weeks and months. And the question really is like, how will the job landscape change for engineers? Like, do we need to become more of a generalist, maybe risking that we lose, quote unquote, kind of this first principle of thinking? Or what is your opinion on that, Seth?

Yeah, so I think, you know, one of the things I think about related to this is like the design processes that we go through, the design loops that we go through as engineers, like I don't think those fundamentally are going to change much. So you're still going to have this process of how do I define the problem that I'm working on? What are the requirements? You know, working down to implementation and then working back up through, you know, verifying and validating that the final solution meets the requirements and solves my problem.

However, what I do think is changing is the role of the human as part of that process. I think as these AI capabilities become more and more powerful, we will be able to delegate more and more of that lower level work to the AI. And so as a human, I think, you know, you said generalist. I do think that that's a legitimate way to describe it.

It's more that we are working at a higher level now. We are thinking much more in terms of what are the problems? What are the constraints? How do I shape the solution space so that I can turn my AI loose in that solution space and have it go work, have it go create something, and then come back to me?

So I do think that in the long run, this will force engineers to really hone and improve their higher level problem definition skills, requirement skills, defining those types of things. And I think maybe not for all engineers, but certainly for some engineers, the lowest level of implementation is perhaps not going to be as important anymore. They will be able to delegate that to the AI in most cases. I'm thinking back in 2022, when I believe 2022, it was right when they released ChatGPT in November.

I'm thinking back then, everybody talked about prompts. And I think with Agentic AI, it's more like not that we talk so much about prompting, that prompt engineering and context engineering. It's more like, okay, I have an abundance of tools. How do I combine the tools in the most intelligent way so that I get to my outcome in the fastest way possible and the most efficient way possible?

Do you think this is a good way to think about this? Or maybe there's a more intelligent way to think about agentic AI and tools? Yeah, I think that's certainly a big part of it. To me, prompting is still really important because you do need to be able to describe the problem that you're working on.

And so I, you know, I prompting is just, I think one of those fundamental skills for, for working with anything that involves a large language model. And I don't think that that's going away anytime soon. But yeah, you're right. There are other important things that, that come into the picture with agentic AI.

And certainly one of them is, is tools. What tools do I need to give, you know, give this agentic system access to so that it can, you know do the work and so that it has the best tools available i think there's also a um you know with some of these tools you have this notions of like skills files and actually knowledge that they gain over time from working with the problem so giving giving the agent the proper skills you know maybe giving it coding best practices or helping it learn more about the you know the development pipeline that you're working on all of these are different things that you you can supply to the agent that helps improve its knowledge of the system and the workflow.

And so there's certainly a lot to learn there, but there's clearly also a lot of potential and excitement. I'm wondering, when you approach a new problem, Seth, how do you break down the problem? Like, how does your divide and conquer approach, so to speak, look like as an engineer, especially now when we talk about like the traditional engineering versus today, the agentic AI age? Yeah, so I think, you know, traditionally, if I was if I was encountering a new problem, you know, it was a lot more deeper thought on my part, you know, a lot more time before I would start implementing anything.

It was it was much more, you know, really completely wrapping my head around every detail. I would want to have a lot of it figured out before I went to implementing anything. That was just, you know, and that's maybe a little bit more of a personal preference. But and then, you know, I think once once those things like once I had my head wrapped around the problem, then I start to do a little bit more of the implementation.

But, you know, the implementation was was very manual work. It was writing lines of code. And so, you know, the before I you know, it was it was a lot a much longer time before I could get to anything that was workable or that I could start to try out. Now I feel like with with agentic AI, one, I can I can reach for it very early in the design process.

Maybe I have a couple ideas for different approaches that I might like take I might take to solve the problem. Well, why not have it implement four prototypes of all of those different options and see which one I like the best? You know that even historically that might have just been too much work or added too much time to the timeline. Now it's very much an option.

And so it really, it allows me to shorten that cycle to get from kind of an idea to some type of proof of concept. And I think, you know, that's super impactful in just in terms of how it changes the design process. I can get some really quick feedback and then I can, you know, spend more time maybe optimizing the specific final design. This is how you approach Agentic AI, which is interesting.

How does MathWorks think about this whole topic of Agentic AI? Does the organization, you work with so many customers, do you guys see a pattern or maybe signals of where we can actually see that fundamental shift going from traditional engineering to Agentic AI? Yeah, so it's certainly something that we're seeing a lot of interest in. So I would say a few things on that.

So one, we've been working on generative AI in our products for years now. We have a Matlab co-pilot that's been out on the market. We have a Simulink co-pilot that's on the way. Another thing that we released in November last year was a MATLAB MCP server.

So MCP is Model Context Protocol. It's kind of the interface or standard in between the tools and the AI agents that lets you plug and play different tools with different agents. And that's received a lot of positive feedback. So we're certainly hearing a lot of interest from our users and our customers.

they certainly see the potential here. And so it's really been up to us to deliver things like that MCP core server so that we can let them explore these new workflows and understand what the impact is going to be. I think in the engineering space, the way I would characterize it is a lot of people are evaluating what's possible right now. With agentic AI, and you mentioned OpenClaw earlier, like there are legitimate security vulnerabilities and other things that need to be worked through if you're actually going to adopt these tools especially in a commercial environment where you you have intellectual property that's going to these systems but but more and more we're seeing people doing these evaluations staying up to date on what these tools are certainly with an eye towards the future and adopting these these tools as part of their their production workflows yeah Yeah, super interesting.

I have one maybe more critical question, Seth, which is, I mean, we at Apex, I have my own company, Apex, that helps businesses implement AI, right? And I'm wondering if you see maybe the same thing that people see AI more as, well, of course, it's a commodity, but also like a nice to have. And some of or most of them actually never see the true value of AI because maybe they don't have a concrete problem or they don't know how to use the tools effectively. So you now mentioning AI co-pilots a couple of minutes ago, like how do you combine agentic AI with something very concrete, which involves security concerns and safety critical systems and multimodal models and maybe also physical systems?

So how do you combine all of that and find the perfect use case, so to speak, for engineering applications where you can then use agentic AI to reap the benefits of AI? If you get what I'm saying? Yeah, I do. And I think there's actually some really interesting challenges that still need to be worked through.

Because essentially, or especially when you bring up the notion of, you know, these are, in many cases, safety critical systems. This is, you know, it's very important that these things be correct. You still need a human to review the output. But I think we still live in a world where ultimately the human beings are the ones that are responsible for the output of these systems.

And so I think that one of the challenges is just if you have these large language models that are incredibly powerful, but they're untrusted, how do you help the humans quickly validate or verify and validate the output that's coming through those? And for that, I think there's a lot of opportunity in how we process the results coming out of these systems. Is it in a programming language or like in Simulink, in a model or a block diagram type of format that's easy for humans to read and process and understand?

So that level of abstraction that it comes out in is actually super important I think also you know more and more integration with existing tool chains that are being used for software testing for verification for validation That's another way that we can build confidence in these outputs. So, you know, I think, you know, when I think of kind of what the ideal kind of future solution looks like, it's a world where the engineering tools that, you know, we've spent decades building over the last years are still the important part, the foundation that is being used as part of these systems.

because those tools, you know, they define the workflows. They also, from a generative AI point of view, they kind of implement guardrails, right? They keep the generative AI from doing anything that's, you know, too crazy. It's like, no, these tools are the ones implementing the best practices.

They're providing traceability from requirements to models to, you know, final code generation. And so I think, you know, we're working towards that future where we're using these tools to keep the agentic AI on track and at the same time helping the human engineer process what the results are that are coming out of the agent so that we can be confident in the work. I see. Yeah, that makes sense.

I'm wondering, do we maybe have a concrete workflow for the audience that we can work for maybe from the automotive aerospace sector that we can talk about something public yeah yeah i mean i think um i can talk in general terms for sure so so like you know one of the examples that we're seeing you know let's take a an automotive example maybe working on some type of um you know uh algorithm that's going to be implemented on on the vehicle um you know we're exploring i would say so many different ways right now that that generative ai and agentic AI could help with that.

Everything from the requirements and, you know, are the requirements written correctly? Are they, you know, could the requirements be improved? Because AI models are great at working with text and requirements are often text. You know, defining system architecture and getting an idea of what the architecture of the system could look like, getting assistance with actually implementing the algorithm up through, you know, as you start to, you know, generate code and do things like verification and validation.

Can you use the AI system to help write tests based on the initial requirements that you have? Can you use it to, yeah, really improve things, you know, on that validation side as well? So I think, like, Like as you think about that workflow or that life cycle, there's a ton of potential throughout it. Yeah.

In general, what is your, and maybe your opinion differs from the one of MathWorks, what is your or maybe MathWorks philosophy in terms of like genetic AI? Will it substitute engineers? Will it amplify them? Or what is the actual long-term goal?

Because obviously, I mean, we cannot walk or talk around it, is there will be a post-labor economics to some extent that some engineers definitely will lose. their jobs. And this is exactly going back to the initial question that I had is how will, first of all, well, will it amplify it or will it substitute engineers? And second of all, how will the job profile of an engineer change?

Yeah. So, so I think one of the things I think about is, you know, first of all, who are the engineers, maybe as a counter example, who are the engineers that are going to be the most successful in the future? And the ones that going to be the most successful are the ones that you know i think are are really um you know good at thinking about things at a high level they're motivated by and they like to focus on just building things you know they're you know they're like the you know these these builders if you will that they're more focused on the output of their work like could i build a new tool to automate some part of my workflow or you know what would my product look like if we added these three features and the people that like to focus on that building aspect of it, they now have this great tool that can help them evaluate probably all these different ideas they had that in the past they've not been able to try.

On the other end of the spectrum, though, I think the engineers that will struggle the most are the ones that get hyper focused on implementation details, maybe not in all applications, But specifically in certain applications where perhaps that's not always as critical, you know, there's there's just certain foundational things where, you know, I think there's certain foundational things where that type of an engineer will still be valued. Maybe they're working on very low level things like firmware, things like that, where there's just not a lot of great training data for the agentic AI.

So it's, you know, there's still a lot of value for the human to provide there. But say in applications where maybe every microsecond of performance isn't critical or there's already well-established design patterns and those well-established design patterns are kind of well-defined and known by the agent, I think it's going to be harder and harder for human beings to compete with the LLM for a lot of kind of that core implementation type of work. Yeah, absolutely. I'm wondering, you probably have more data points than I have when it comes to the adoption of AI in general.

We don't have to talk specifically about agendic AI. There is like the technical side of things, maybe cultural pushback, organizational structure, maybe even legislation that would allow organizations to not maybe adopt AI as fast as they want to. So what do you think is the most critical parameter in that equation, which maybe prevents people from adopting AI quicker? Or is it maybe even the people or organizations that don't even know what to do with AI to begin with?

So maybe it's like a two-folded story here. Yeah, I do think it's twofold. I think on one hand, you certainly have people in organizations that are new to this space and that need some guidance on how the technology can help them, what types of applications are realistic. You know, we certainly, especially back in like 2023, I remember getting into a lot of conversations with people where once they started saying a little bit more about what they were hoping to get out of the system, it was clear that, you know, maybe they didn't quite understand what the technology was capable of and what it, what its limitations were.

I think that's gone away a little bit just because these tools are, you know, becoming more and more ubiquitous. And so people do have a little bit better fundamental understanding of them now. The other thing I would mention, though, is certainly data security, regulatory types of issues. You know, if you want to use state-of-the-art large language models today, you still need very expensive GPUs, very expensive hardware.

And that hardware is really only economical at scale. And so a lot of the systems today are hosted on the cloud so that that can be shared. And, you know, still today, I think one of the biggest adoptions is a lot of commercial organizations are just not comfortable with the idea of their IP being sent to remote servers. And so, you know, or maybe they're, you know, they're regulated by depending on the industry that they're in.

And so that's certainly a, you know, still a barrier to adoption. Yeah. But then when we talk about use cases, I think still the engineers, maybe even decision makers listening to this podcast, maybe they're not even sure where to use AI to begin with, So how do we pick a good use case? Like what does a good use case, what parameters make a good use case?

Yeah. So I think perhaps one of the most important things in this, you know, this kind of brings me back to my days working more on traditional machine learning, deep learning types of problems as well is like, how are you going to evaluate the success of the use of the tool? Like you need to have a good benchmark or a good way of measuring what success is going to look like before you start an evaluation. I do think that a lot of organizations have just kind of dove in to this technology and they're just trying it out.

And they're, you know, many times just relying on qualitative feedback from people trying it out on whether or not they think it's going to be beneficial. But, you know, if you have existing, like existing code repositories or existing models or things like that, where you know what, where you have a well-established benchmark, you know, to me, that's a great opportunity to run a fairly well-controlled experiment where you're actually going to learn a lot more about the benefits of the technology, the limitations of the technology than, say, if you just have people go off and try it on new problems.

So I do think there is a fair amount of science that can be brought in there to actually assessing the tools and evaluating them. And we're seeing more and more, there's like some software benchmarks and things like that that are out there online for evaluating these models. But those benchmarks are well known. They've been in the public for a while.

And so I think, you know, trying it out on a problem that's unique to your organization, I think, is really key in assessing it. On the other hand, I think there's even a saying in architecture, which is never plastered over a crack, I believe it is. I'm not sure if that's the correct English saying. So the question is, is there a risk that organizations or individuals will actually superficially use AI, not knowing how to properly implement it and maybe rethinking workflows from the ground up?

Do you think there will be a problem eventually? Or do you really have to start from the ground up and maybe rethink systems completely before you even start thinking about AI to begin with? Yeah, that's a good question. I don't think that it needs to be approached from a re-implementing things from the ground up.

I think one of the impressive things about a lot of these AI models is that if you can teach it about the system that it's working on. And from an agentic AI, that point of view, that could be helping it understand the code base that you're working on, giving it an overview of how the code is architected, what the system is. These agentic systems are also going to be reading the code. So they're going to be knowing about the functions and things like that that you have.

They can build on top of existing things that you have. I think in a way that's much different from previous technology changes where it required more of a fundamental shift and a paradigm shift So I would not look at it from that perspective of throwing away everything and starting from scratch I think there a lot of opportunities there to continue building on existing things that are in place today and essentially teaching the AI how to operate in that environment that was built with humans in mind.

But it's suitable for an AI as well. Mm-hmm. Fast forward maybe to, let's say 2030. I'm not sure if that's too far into the future, because I mean, I couldn't even imagine that something like OpenClaw would exist like one month ago.

So I'm not sure what's going to happen in 2030. But how do you think the landscape of engineering will change moving forward? What? Yeah, so I think, I think to me, you know, engineers fundamentally will, will have a much better understanding of what generative AI is, where it can help, you know.

And so I think that will be, you know, that in and of it alone, in and of itself, you will have more engineers coming out of school with knowledge of these tools, you know, we're seeing a lot of interest in adoption of these tools in academia. Industry wants students to come out of school prepared to use these tools because industry sees the productivity benefits and that means that schools, engineering schools, want to be teaching them to students and so you will have just a higher baseline knowledge of what these tools are, how to work with them, how to use them.

I think we will certainly continue to see more and more integration of these tools. And I think, you know, I, one of the things that, that I'm hopeful for as well is just that this bridges the gaps between a lot of different engineering domains and engineering tools. So I mentioned, you know, model context protocol MCP earlier and and one of the really exciting things to me about mcp is you know historically if i want to um you know do work in three or four different engineering tools it's often a lot of work to connect those tools together you know to map inputs and outputs between them to get them to to run run alongside each other um but if you you know if if organizations embrace this mcp concept or embrace this notion of building tools for agentic AIs and now the agents have, you know, the ability to work across these different tools.

I think that's going to open up a lot of really interesting, you know, cross domain types of workflows that, you know, historically would have been very manual or very tedious for engineers to implement. So that's certainly, I think, one of the things that excites me the most. And then, you know, So I think lastly, of course, there's just going to be more and more of the tools that we work with every day are going to have some option to use agentic AI. I think initially there was kind of this initial wave of organizations feeling like they just had to have something and so they built something.

But I think over time, over the next five years, those will get much more refined. They will become much more targeted at the workflows where clearly agentic AI is adding value, not just kind of like a shiny new feature, but something where it can really provide a lot of value to the end user. One of my favorite things about these tools is just getting explanations of code. you know if i go look at a new you know code repository historically you know reading code from a new code repository is like a really tedious task it takes a long time i i can now just ask an ai for an explanation of this and get high level overviews i can you know get you know i can i can wrap my head around things so much quicker and so i you know workflows like that are you know both both not a a threat to me in terms of like job security but also just super valuable as an engineer because it's going to allow me to work with so many more tools that I would not have been able to do in the past.

What do you think most engineers get wrong still about AI? I mean, AI is not really a young technology. It's relatively old if you look at the timeline. What do you think engineers get a lot of times maybe wrong about AI?

What do you think, Seth? Yeah, I think, especially in today's world, I think there are certainly people who think about this more as like replacing jobs or, you know, that are concerned that AI is going to replace their jobs. I think a lot of that comes from not having the background, the decades of seeing how this technology has evolved from machine learning to deep learning and, you know, not being able to view these as, you know, text generators or pattern generators, kind of similar to, you know, how people who have trained models from scratch view these things.

And so I do think that there's certainly that challenge there of if folks refuse to evolve how they work, you know, then, you know, some of the things we talked about earlier where, you know, they're not going to be able to make that leap to working at higher levels of abstraction, thinking about things. And, you know, like in pretty much every industry, at least every industry that I've worked with, you know, there's this trend that the engineered systems, they continue to get more and more complex.

Timelines get shorter and like maybe humans are all getting smarter. But I do think it's also a lot about the processes, the workflows and the tools evolving to support the needs of engineering organizations as this complexity increases. And so I do think that, you know, if you're going to be successful to, you know, in the future as an engineering organization, you will have had to have adopted key or generative AI into key workflows. You'll constantly have to be on the lookout for new opportunities where generative AI could impact the business.

And I think that's really, you know, what's going to determine who's going to thrive in the future. Absolutely. Maybe one of the last questions I have for you is how can people already start using AI today in MATLAB and Simulink? I mean, you talked a little bit about future vision that Agentic AI will come for Simulink.

But what are some concrete use cases and what can people actually fiddle with today in MATLAB and Simulink when it comes to Agentic AI? Yeah, so there's a whole lot of different things to talk about here. Back in 2023, we built what we call the AI chat playground, which was just an app on our website that was actually a great way for people to understand what it's like having a chat bot helping with MATLAB. That's still there.

It's probably the easiest thing to access. We've built MATLAB Copilot now. So MATLAB Copilot's available in MATLAB. It's a great way to start understanding how generative AI can help with either giving you advice about how to tackle problems in MATLAB, giving you assistance with writing code in MATLAB in the editor.

It can do things like adding comments to code, getting explanations of code. So there's a lot of different workflows in MATLAB where we've added these, you know, in some cases, like single button clicks or single, you know, single prompt that I write to get, you know, to ask MATLAB Copilot to help me with my problem. So that's, you know, that's certainly a great way to get more familiar with this. I mentioned earlier the MATLAB MCP core server.

this is um if if folks are using agentic ai tools today and want to integrate matlab as part of this i encourage you to check out that that mcp server that we've we've put up on github and then simulink copilot um if you go to our website you'll see that's currently in beta um so you know if folks are interested in that we'd love to to have you um uh try it out i think you know there's a ton of potential here with simulink we really see this as a you know a tool that can provide guidance on on simulink workflows that can explain how simulink models work that can really be that that assistant that's there to help you um as you as you work through um the block diagram models in simulink so um you know kind of regardless of if you're if you're using you know some mathworks tools there's probably ways that you could be using generative ai today um and And we've really been trying to really just kind of meet our users wherever they are on their AI journey.

Maybe we can also leave some links in the description of this video so people can check out some other resources. Last question for you, Seth. How do you keep yourself updated with the latest trends in AI? I myself, I'm not sure if you do the same thing.

I go a lot on Twitter or X, how they call it nowadays. It's like super useful when it comes to like AI and agentic AI, etc. where you see all the updates there. And all the famous people who work in the Gentic AI are also on Twitter.

So I'm not sure what you do, but maybe you can let us know. Yeah, it's a whole bunch of different sources. So it's, yeah, certainly Twitter or X. There's some blogs that I like to read.

Simon Willison's a great one. He's very thoughtful and has a bunch of great things about AI. Reading Y Combinator News. There's a couple YouTubers that I subscribe to, you know, Andres Karpathy.

You know, there's no shortage of information out there. And yeah, it's certainly a fire hose. I try to drink from it every day. But, you know, it's always a challenge to keep up with.

And so I'm also fortunate to have a bunch of great colleagues who will share stuff with me when interesting stuff happens. That's great. Yes, it certainly feels like that. Like if you miss 24 hours in the world of AI, it feels like you're 12 weeks behind somehow.

Yep. All right, Seth, thank you very much for taking the time to be on this podcast today. As I said, all the relevant resources will be linked in the description. And then if people have any more questions, they can reach out to you, I guess, on LinkedIn.

If they have any more specific questions around MATLAB or Simulink and the topic of Agendic AI. So thanks again and hope to see you in a second episode in the future. Yeah, thanks again, Yusuf. It was great being here.

I really appreciate the opportunity.