AAH #815 – Automotive Software Is Dead and AI Killed It

0:01AutoLine After Hours is brought to you by Bridgestone Tires, solutions for your journey.

0:08Gary Vasilash: Hey everybody, thanks for joining us on AutoLine After Hours. We're busting with excitement here. We can't wait to get into this show. Gary? It's going to be a fascinating one. It's going to be a really interesting one. We've got Dave Robins, the CEO and the founder of Intrepid Controls, great AutoLine sponsor, by the way, too. Sean McElroy is joining us for the show again. So Dave, you're like bouncing off the wall talking about AI here. How do we get going on this conversation? Because we're going to cover all kinds of stuff. Well, you know, this year, it's like a golden age where AI actually can do just unbelievable, amazing things.

0:49And I turned in kind of, they call it AI vampire, where you just sit in front of your computer. I did it all summer. And it's like when I used to be addicted to video games, I would just be there and not get up. Actually, I put water bottles by my desk because I didn't want to move. But I really just poked at it. Every technology around what we've done at our company, electronics and hardware and stuff. And I'm just constantly blown away about what's possible. So your company, though, I mean, for those who are not familiar with Intrepid, give us a thumbnail so they know where you're coming from in terms of AI and its utilization.

1:30Yeah, Intrepid's been around for 30 plus years now. I started it out of my dorm room at college in Arbor. We work with almost every OEM and we help them build their electronic system. So all the computers in the car, all the, you know, the software and the testing around it, all the way from, you know, the design all the way to production. And then also in service, the service area, we kind of are there for like the real hard problems because our tools are like you go to a diagnostic tool. You know, you go to a shop and they plug something that's more of like for the technicians. But, you know, there's certain issues that are just real hard to solve and they'll use our tools.

2:13but the tools we do the whole thing or whole solution so we do it from the design of the circuit board something like this, we do the design of this this is a very simple one I'll talk about later to the actual hardware and then on the host PC or website we do the software too our company covers all dimensions of computer technology that's what we do

2:40Gary Vasilash: your tools have become the de facto standard in the industry by this point. Yes. We have competitors. We have good competitors in Germany, but we basically are relied, our tools are essential to get the vehicle out the door for many OEMs, yeah. Yeah, you talk about diagnostic tools, and these are obviously much more advanced than, say, like you said, the technician would have, where they can even detect faults that the system in the vehicle cannot detect, right? Right. Yes. Yeah, it's like at the engineer level. Right. Something's happening that they've never seen before, which happens because it's a complicated system and all these things.

3:27Then why is there so many software problems in the industry? It's hard work.

3:32Gary Vasilash: I mean, it used to be. We're going to talk about something else now, right? You mean it used to be until this year when AI came along? Yeah, AI changes everything upside down. So, I mean, it's like having a magic genie. You just have to want something, and then it gets produced. So it's almost not believable, the technology. and it's really it is it's like the biggest opportunity the automotive industry's had and I would say most companies have had to like compete in a different way. Do you think they're taking advantage of that? I think it takes time for the light to turn on and once you see the light then you're like oh my god you know like you know it's hard to understand what it means.

4:29And I don't think anyone knows where it's going exactly, but I think if you get behind, you should probably catch up. So, okay, in terms of automotive, when you're saying that they could take advantage of this and they should embrace this, I mean, what is it that it's going to do for them? Well, I can, like everyone says the car is a computer on wheels. And what does the computer use? Software. Software. So now you can just tell it what software you want and it gets created and it gets created um the frontier models are so good at it um they know the automotive protocols you know i don't know how but they do maybe because we talk what do you mean frontier models frontier models so so so this technology it's around like a like a low they call it llm or it's like a model of it just adjusts things and finds patterns and And then, so the frontier models are the companies that have the best ones and they're spending the most.

5:32So Anthropic and OpenAI are far ahead of everyone. Grok is probably third. And then, you know, the other companies I wouldn't bother with. Wow. So. So Anthropically or Open is what you say.

5:50Gary Vasilash: They're at the leading edge right now. Yes. Yes. And I would say, you know, Anthropic for most of the year has been ahead. And then what they've been doing is introducing new models and then kind of switch and check them. But I would say, you know, OpenAI Codex is their coding agent and Anthropic is Claude. I don't know if you heard that one. That's their coding agent. They've kind of been flip-flopping. Claude has generally been ahead. So, but, you know, when the Astro was the name of a model that came out that was mostly AGI, which is general intelligence, which is not, is not, is not. But, you know, everyone switched to that that week, you know.

6:31So I would guess maybe that Anthropic, you know, third quarter numbers maybe were down because of that competition. But that's a big event that's happening this month, you know. So if you guys want to take your retirement funds and put it all in Anthropic. you know there's a big tech sell-off going on right now as we speak so right now the buy on the dip i don't know it's well i think that you know it's like the dot com right if it does bust the website the web is still here right so like the technology is is is traded and it you know it's amazing but back to this computer on wheels so okay so i'm an engineer and i need a new adass system for my vehicle and you guys are involved in that i know yeah um so i just tell claude don't mean adas and i give it a little scary right because it's you know it's there's a lot of um you know uh the standards around you know like safety you know that you have to follow uh but even at that level uh you can organize and plan something you know like well i i need to use this standard and this is my plan this is what i'm going to use and then you kind of work together through the design uh and uh you still have to have your you know engineer brain for that you know you and you have to be able to check its work because you're as an engineer at that level you're responsible you're accountable for whatever the ai does um you know like maybe for body electronics it's looking to be a little bit more you know cavalier about what would that be body electronics would be like uh you know displays and radio and stuff like that where you can you know seat massagers i mean there's like uh you know hundreds in these high-end high content there's hundreds of of ecus um and that's what you know like you said software problems um imagine trying to get 150 pieces of software to work together from different vendors that maybe are not even in Michigan.

8:32They're in Germany or all over the world. It's a global industry.

8:36Gary Vasilash: So it's real hard. But that's with domain computing, right? They're doing all this stuff to kind of make it simpler, you know, like switching to Ethernet. Zonal centralized computer. Yeah, zonal software-driven vehicle and stuff like that. But Dave, if I understand this correctly, you know, Gary's talking about using the AI to, like, help build the systems itself. But if I understand, it goes beyond that, too. Like, say I'm an engineer and I do the radar part of the ADAS system. And I'm probably taking in crap loads of data every day. You can use the AI to select the particular data that you want too, right?

9:13I mean, it's like asking a genie. You can look at data. Honestly, the radar is probably a great example because it's, you know, the radar is, it's like, it's a real strange sensor, you know, like the data coming back from it for interpretation. And AI is like really great at pattern recognition. So it might give you insight into actually how to do a better radar. And I think that's when you look at like where AI is going to make our lives better, like in medicine and stuff like that. I think that kind of stuff is going to be like great. It's going to change, you know. I ran into. But you should also, you know, it's a trick to use AI to do that.

9:52But like there's better ways of working at AI, you know, like maybe let's, you know, like we talked a little bit about you making like some. some sort of summary of an event like you might be able to like john how many pieces of software

10:05Gary Vasilash: have you written in your life none actually yeah i'm not a coder right but you you probably didn't even think that you could be correct but like you could go um you know after this because you're now you're excited after this i am so after this you can go in and say i want to write an app that that will summarize an event and it's like you use ai to make the app and then you don't need to use AI, you know, unless you want to make changes. And that's just, you know, you start to, you start to learn things after you use AI, like don't use AI to look at data, write an app that looks at data, and then maybe use AI a little bit on the side.

10:43So one of the things that we're going to do for our customers is, so we have the frontier models, and then you have a customer and OEM, and then the frontier models and a lot of other software companies around ai they're all they all work on tokens which is how you pay for it and someone has to be in the middle to like conserve on tokens and because they're getting really expensive i mean they're spending a fortune on tokens now what's scary about tokens is that the cost is not really known i mean i mean the anthropic might know but like it's kind of like you know they if you were totally dependent on them.

11:22They had this drug dealer model going where you get your company where it just works around AI and then they got this dial like, how profitable do we want to be? We've all been beyond that before. So I think one of the big things we're going to do for customers is we're going to be like, well, we're going to figure out how to minimize your token usage so someone to have that incentive because incentives drive the outcome. So, like, you know, those AI companies are going to want to make you use as many tokens as you want, and they want to lock you in and stuff like that. So who within an auto company would you be working with to accomplish, you know, to pull this off of?

12:04Well, we would work with our existing customers. But I mean, what are their functions within an organization?

12:12Well, they are, you know, a particular OEM has 4 ,000 engineers. and whatever they work on, it's like the electronics are always involved. So our customers, basically they don't know what the car does without one of our tools. That's why I have a pretty good business.

12:35So we would work with them to... I feel like we have a software business and we make hardware too. But you could certainly argue that software is done, it's over, in terms of businesses. Because you can just tell the model to make your software how you want. I don't like Microsoft Windows, I use it, so I had it rewrite all the apps I use. And they work exactly how I want them to work. and, you know, like movie player, web browser. And I'm not using any skills that I had as an engineer, really. I'm just aware that that's possible, you know. So I think that for us, our business is we, you know, We provide the software that an engineer at GM, whatever, that would use.

13:48So when you build software, you kind of build it from pieces. And a lot of the software you'll see that people make on Twitter or X, it's quickly done, but it borrows from a lot of libraries. It's just software that's already done and tested, hand-coded, been around forever. But the end result's amazing. So like our model is basically that same thing. We have these libraries that are for vehicle network technology that are proven, handwritten. They're looked over by experts have been doing it for dozens of years. And so I feel like there's no more software application software that goes to our customers.

14:28so like if you're um say you were in charge of like the new escalade seat and it's got like massage and you know it talks to your ear about how great you are and you know they just go extreme right so uh you know that thing it has to interface with the rest of the vehicle and these data from the rest of the vehicle uh it has to be able to diagnose itself you know all that stuff but if you were the engineer working on that um before before or using any software you've ever used like microsoft card there's like a million menus like why not just have software that is exactly what you use and nothing else um and that's totally possible now like in our software now you can ask it to just i want to see this data and nothing else and it just writes itself inside the software uh and uh you know if you start to go you know what can you do in an automotive of like how about we just ask the car to fix itself wow like why do you need diagnostic tools you don't need diagnostic tools that should know like and like maybe it just you know you know i mean just fix yourself you know i think that's on result like i imagine like so like say we go into a customer and they have like this is us but we go into a customer and our normally our sales engineer very technical guy he would meet a customer and say well we need this you know we're working on this Rivian program, blah, blah, blah.

15:53And we need this software. I mean, we don't have it, but we're like 90 % there because we have hardware and stuff. So like, you know, our sales engineer does the presentation and then he just opens up his laptop and then types in what he wants, what the customer wants. They go to lunch and then they come back and it's done. And before, like, it might take three months or six months. You know, so when you look at, like, maybe we can finish a program a year early in automotive now if we if we we embrace this

16:26Gary Vasilash: technology i did that's just what i was going to ask you about where does this leave the auto industry because they're working to slash the time it takes to get to market to be able to compete with the chinese but also i mean we're you got you can do it now yeah this is this is like we're chinese yeah we are chinese now with this technology you know but but the other thing too is, you know, in most automotive programs, at least at the peak, you know, there's a from the start to begin, there's at least a thousand engineers involved. It doesn't sound like you're going to need that many people going forward.

17:01My opinion on it, like for us, right, because I have a tech company and obviously software, it's like an extinction level thing. You know, if you could go there like nightmare, like, but I think what we can do is we can we can do a whole lot more for our customers and we can do a whole lot of better stuff, you know, better stuff that we couldn't do before. Cause it just, it wasn't possible to work on, you know, like you know, stuff like making the software fix itself and like providing solutions that, you know, a customer needed in a, you know, a week instead of a day. And now like, you know, I'm just want, I just want to be optimistic about that.

17:41Like, you know, obviously management can make other decisions, But I think the smart, you know, enlightened thing is like, well, the smarter enlightened thing is now you've freed up your people to go work on high value added stuff. I mean, just let's, let's just get it, get to see how far we can put, take it. Like, can we be Chinese? You know, can we like, you know, I was in China, you know, for 20 years, our office there, and I'm just sitting there eating ice cream cone at, you know, midnight. And I'm seeing these Chinese guys working on it. I'm just like, what are we going to do? They're just so hungry to, you know, work.

18:09but you know um maybe you know ford created the 55 dollar workday yeah maybe ford could create a four dollar four day work week you know if ai can let you now we're talking now we're getting someplace i mean if they like you can do 30 days of work in one day should you work 30 days or maybe 29 i mean everyone's scale is going to be different i mean i don't know i mean like it's not that Like that's just dreaming, right? But like it would make sense that there should be some benefit for everyone around that. And that could be one way where AI is something that is positive. Like there's doom and gloom.

18:51There's so much doom and gloom. There was a recent talk by this. There's an AI evangelist. His name is DHH, which is David something. I always mess up his name. But really look him up because if you want to know what the end result of programming is, He kind of has a vision that's a nightmare and a dream. But he recently did a talk in China, and he said 90 % of the people in China are super excited and optimistic about AI. In America, it's 30%, which I don't know where that data comes from.

19:22Gary Vasilash: Well, there's a lot of fear in this country and not just this country. AI is going to take their jobs away and maybe end up destroying the world. Yeah, which, you know, I mean, I don't know if that's going to happen. I'm just saying who knows what's going to happen. But I watched, there's historical films on YouTube, right? And I watched this one on when it was 1967, Periscope is the name of the channel. And they were just introducing computers and they were interviewing all these people. Computers are going to take our jobs. There's not going to be any jobs left. You know, it's just like. Same with robots.

19:57I just feel like we just create jobs. I mean, everybody's going to want to do more. and you can do more now. So that is actually a danger of AI. Before we would do this, but now that we have AI, let's do 100 things, which is maybe not a good idea for a company. So people who are in the auto industry now or people who are thinking about going into the auto industry in terms of the electronics area, what skill set do they need to make this happen? I mean, you just need to be a good engineer. year i think and know how to use ai yeah so i so intrepid we we uh we we sponsor uh these hackathons so last year we sponsored all four that are michigan so that's uh oakland university university michigan dearborn university mission sparta hack and like we did we just did the u of m uh the one this year and i was just thinking like like what it must feel like to be uh you know a graduate it you know like you you know you don't they don't need you or whatever but uh so what we did is we provide this this hardware so hardware is the new you know the new hot thing you know because software is like well anyone can do software but uh so um but uh my realization at the end of the day with what these students put together which was amazing is that uh before i was a capable engineer to be able to do what no one could do.

21:26But these students, you know, they're a genius, you know, they're, they're smart, you know, I'm, you know, I'll, you know, anyone at that level is, you know, in, in, in engineering. And they, they produce amazing things in a day that, you know, that, you know, you never would be able to. So I think the argument that says that if you're going into college, and you don't, or you're not valuable, you could argue you're more valuable because you cost less than a 30-year-old engineer. So, I mean, I don't know what's going to happen, but we have a bunch of awesome guys at our place, and there's this one guy that his code is not as good as mine, and I look down on him kind of, just not look down on him because I love him, but we're working on a project together.

22:13His code is as good as mine now, so I can't say anything about it. but is it as good because of ai or yes yeah yeah because you you basically look at it and uh uh you look at the code it produces and it's good it's good yeah uh it can be wrong you know so like if you depending on what you're working on an adas system it's got to be more surgical where you you're inspecting everything uh but i would say the code that it creates is is much better uh than most people are capable of, were capable of. So like, they say AI slop, but like there's a lot of human slop, a lot of human slop. Yeah. Do you use different agents to check each other?

22:58Yeah, yep, yeah. So that's part of the learning curve of AI. So that's why you need to start, you know, it takes time to really understand how to use it productively.

23:09Gary Vasilash: How would you advise people in the industry going, wait, wait, I need to learn more. How should they go about doing that? Get a Claude or OpenAI subscription and just pick an app that you have on your computer and ask it to rewrite it and just see where it goes. Or if you have any sort of personal interest on anything that uses computers or software, just point it at that and just have some fun. It kind of blows your mind. You're like, oh, my God, it can do this. Oh, my God. You know, like 3D printing and, you know, just I mean, everyone has their own experience. but like really you should use the frontier models because they they they're just doing stuff that's just unbelievable what about what about companies though should are companies coming to you to get this sort of learning or where do you think starting but but i see that as a big thing so so uh like i'm the i'm i'm the ceo at intrepid and um you know people were talking to me like ai this ai that i'm like whatever you know blah blah blah it's you know because i use it and it's like, well, it's cute, you know?

24:13So last April, we have like a global sales conference where we get together and everyone talks about what they're working on with customers. And our UK office had a presentation about hardware and loop testing, which is a business that we really don't do. it's like a big business for our competitors and they were able to use AI to implement an entire system and win a big chunk of business and

24:53how it worked was they went into the customer, you know the customers don't even necessarily like what they're using and the prices and everything you know and uh the customer is like well i don't want to do it unless it can do my tests and here's 500 tests and then two days later they came in with them all working wow and uh so once i saw that i was just like oh yeah i better figure out what this is about so literally after that meeting i went back to our hotel over japan i ordered a six thousand dollar laptop you know as i could run some of the local ai models and and uh and then i just did a deep dive all summer and uh you know when you want to change your organization what we what we did is we kind of created this ai group so we have a ca ca ai officer chief of ai uh samir um and then and then we're doing weekly trainings on ai so we're so we have an ai chat group we share internally what people have done and wins at customers with it.

25:59And the training has helped a lot because there's a lot of fears. A lot of people don't use it right. And they use like Copilot, which is just, you know, it's a wrong turn, data, and U-turn. But the Frontier Miles, you know, so that's really helped us. So like our plan is to like transform our internal company, you know, every part of the company can use it, even like accounting. and, you know, our production. We actually have a production facility in Troy. And through that training, we are at the point where we want to do that for our customers. So I think we could train all our customers next year.

26:38And the way we're going to do it is we have, well, this particular hardware that we released, which we're calling the first AI hardware that, you know, in automotive. It's just unbelievable. So what do you call it? You have some name for that. This is called Rad La Serta. What does it do? Well, what doesn't it do? What do you want it to do? Well, I mean. It's the genie in the bottle, Gary. Yeah. So what blows my mind about it, and I just can't believe it works even still. But, you know, we experiment, and then it's like, wow, it does work. So the mode you would do, you know, like it's got audio and stuff like that.

27:20So we had a guitar guy at the office, and he made a guitar pedal and a voice changer and things like that. And he's not really a super coder, but the way the workflow works is you fire up Claude or whatever. Oh, quick question. Have you heard of Claude before? Yes. A lot of people are wondering if it's a girl or a guy. Do you guys know who it's named after? Could be either way. Claude Shannon? Yes. do you know where he was born no michigan gaylord michigan oh all right and his undergrad was at university of michigan and they actually have a statue in front of the electrical engineering theater science building um where i went to school so but i never knew obviously i just like it's cool to know that claude is michigan you know all right anyway you'll fire up claude or um codex or grok uh and you say i want to make a guitar tuner with this hardware and i pointed at what's called a repo and a repo is just uh there's something called github where it's just like kind of a collection of code uh so anybody here um you know if you're doing you know you do audio or video you know or whatever and you had an idea for a product you'd be like well i want to make this and here is a repo and then go and then it just works and it will make something you know and you can obviously explain it more uh but uh you know eventually it will come out so uh i put together kind of an example i didn't use this hardware but uh but but that works that just works with this hardware the screen the buttons and i did it a video of doing a full diagnostic tool and it took about an hour and a half with a diagnostic simulator and a diagnostic tool I mean the software the screens look better than stuff that we've ever built and any tool you probably used because you know just yeah but I did a simulated video of this AI tool that we wrote called AI Squared where first I cloned you guys and I cloned your voice and then we worked together on a project where we i what i did was i took i took uh so the tesla infotainment it's it's a pretty good design uh i took a picture of it my camera my regular camera uh of two screens in it and then in the video that i got i used you guys i told you guys to write the code to implement the tesla or T-SLA.

30:03T-SLA. T-SLA. Because you didn't want to use the Tesla. Well, it's just a fun thing. So implement that user interface in our software. And then I think it runs for like a half. It took about a half hour, but you can see the result in the video where it looks exactly like the Tesla UE.

30:24Gary Vasilash: Okay. So we should take a quick commercial break right now. We should come back. I think we can run that thing that you put together. But first, let's give a shout out to our great sponsor, Bridgestone.

30:37Gary Vasilash: Making a life full of memories, one road trip at a time. That's what really matters. Bridgestone Weather Peak tires with a 70 ,000 mile limited warranty. All right, we're back talking about the death of software and that AI killed it. that Dave's got this incredible video that he cloned, what, me and Sean? Death of Software, our chief AI officer, Samir Bagwai, coined this phrase, there is no more software, there's only customers. I mean, what that means to me is that, you know, if you are a business and you have customers, that you need to get closer to the customers. Cool. Seamus, do you think you can run that now?

31:22Yeah.

31:29I don't know if the audio, was there audio with it? There's an intro video where I have you guys talking to each other, and then there's a video of you doing the project.

31:41All right, yeah, we can start it back up again then.

32:05Gary Vasilash: I'm John, the default Claude agent on TSLAW, here to write code, debug, and get things done.

32:14Dave, I'm Sean, your curious automotive investigator, here to dig into the code, verify the facts, and explain the results clearly. I'm Dave, Graph4C7 on DSLAW, and I write, debug, and verify the code in this workspace.

32:39Yeah, so the second video, there's no audio, but it's a time-lapse recording of you guys working on the project and me checking your work. so they're they're they're they're agent they're sub-agent you guys are agent agent you guys and just by you know copying your face and your voice but but uh because you know for those who are listening well for those who are listening though i mean what what is are they like visual representations so we were just these guys we were ai agents in there right and you just made us is that right well um the software that we wrote to do ai is called ai squared and in that software you can build you can assign personas to the people the agents that are working yeah it's kind of a fun way to work you know where you can clone your friends or the default set of agents is a team so i have hannibal and and uh you know face man and you know and they you know they talk to you so it's kind of like you have it's kind of like you're a computer nerd that has friends but your friends can be whoever you want so like i did dan campbell because his voice is so motivating.

33:47I mean, I wouldn't show it on here. Dan Keppel, for those who don't know, head coach of the Detroit Lions, right? Yes, yes, Lions. Very important to me. Are we waiting for the video? No, no. There's a second video. There's a one-minute video that shows you guys doing the work. But in that video, it'll show the screenshots of the end result of the Tesla infotainment system. So, you know, it's just something that, you know, I mean, I don't really know what the VW thing was where they wanted to work with Rivian for like tons and tons of money. Well, you know, I'll give you the background.

34:24Gary Vasilash: They spent 10 billion euros standing up a software company called Cariad, which wasn't able to get the job done. And now they're paying Rivian five to six billion dollars for its system. so um that was you know last year and this year you know there's different options they don't need that i mean i'm not i'm not saying that for sure yeah like i probably could go into rivian and do the same thing which i demonstrated you know i took pictures of the tesla and then it remade them um so i could do that with rivian probably and you know it it would probably take more than a day but i mean is it shippable i don't you know that's a whole other thing but But I feel like that they wouldn't have failed if they had this technology.

35:10Yeah.

35:11Gary Vasilash: What you're talking about here is revolutionary for the auto industry. And it goes beyond the auto industry, obviously. But, you know, we talk automotive here. So, I mean, what you're talking about here is amazing. Computer on wheels. What does, you know, make software? You know, like, I mean, if I was running an OEM, you know, this DDR memory, right? They went to Micron and wherever, and they got five years of supply. I would go to these frontier models and just say, I need five years at this pay. Maybe they can get a good deal before the IPO, because it can change everything. So it would be five years of tokens is what you're saying?

35:51I mean, I don't know. Right now, so right now that they have these plans, well, the script fishing plans, and they're heavily subsidized.

36:03I spent 150 billion tokens this summer, which if you look at the numbers, what that costs is like$6 million, which is unbelievable. I mean, I made all kinds of stupid software too, like Flappy Bird. I made like 20 versions of Flappy Bird. That's just a fun demo. I mean, I can make one right now where this would run Flappy Bird where you guys are jumping over cars or something. Yeah, I find this interesting because so much of what we talk about on the shows here is how the auto industry needs to get more lean and move faster. And so much of that we thought needed to be collaboration between all these automakers.

36:45Maybe they don't need to collaborate as much as before.

36:50Gary Vasilash: I think they do when it comes to capital investment. So, for example, one of the big things I've been talking up, and it's not just me, is how the industry needs to do more reuse and more component sharing across the industry, which is exactly what the Chinese do. And what would be very interesting is taking what you're talking about there, Dave, and just saying, what if GM Ford and Stellantis said, let's just throw our entire parts bin into an agent. and it picks out the best common components that are commodities that the customer doesn't care about has nothing to do with brand identity door hinges windshield wiper motors fuel filler necks that sort of thing uh you could save a fortune yeah i mean i don't really know anything about that part of the business but uh i mean i would say every part of the business you can you can build tools around stuff like that that give you insight that you didn't have before and it doesn't necessarily have to burn tokens what I like about it is it kind of just say there's someone that has that idea at OEM

38:06like last year they would have to be like well I have this idea do they stick it in the idea box do they have to go to IT and then someone has to decide whether that's important or not and they don't really have an idea of how important it could be it might be experimental it might not worth even trying but now you are empowered to open up an agent and then create a proof of concept or maybe even the final product some of our customers still use a lot of spreadsheets so I mean just that just going from spreadsheets to like whatever we want that solves the problem it really just enables it really enables you to take a computer and make it do whatever you want like in the easiest possible way you know so everything that you've ever used a computer for can be done faster and easier and only only what you need you don't need to like worry about somebody at microsoft and how they thought you should work or like someone on google like you know they have no clue and the software is hard to use because of it.

39:11It doesn't have to be hard to use. It can just be what you want. Yeah, I also think about the stories of, you know, suppliers, supplier A's software didn't communicate well with supplier B. Now you could just go in and say, hey, why aren't these communicating well, it seems like? Could you tell it to communicate? I mean, tell it to combine so it does communicate? Yes. Okay, so that, which is fix yourself, right? So, you know, we developed this. This is kind of the learning thing. It is, you know, we, you know, like, first of all, we were writing everything by hand. And then it's like, well, this doesn't, you have it do this, write something, and then it doesn't work.

39:59And you're just, well, we'll fix yourself. This particular hardware, we actually integrated the hardware to actually test everything, too. so when you work with AI if you can provide a tool to have insight on what's going on it will use it so for example this particular device has LEDs and a screen so at some point someone's like well why don't we just point a camera at the screen and tell it to look at the screen and it just worked it would read what's on there and uh so so that's called end-to-end testing so like um you know i would envision for our customers like this is like someone writes a specification that's detailed enough that um it also explains exactly how to make sure it's correct uh uh and and then then and then you provide the tools you know which would should should be our tools of course uh but you know the camera works right now the audio works like if you have sounds like so you got a chime it will just use your microphone on your computer to learn it and it can tell you everything about that chime like it's repetition rate it's frequency everything like for validation so that's why you know when you talk about like can it be your plumber absolutely it can be your plumber but it just needs to have arms you know it doesn't have arms yet it does have eyes and ears and it works it works great I mean, right now the models are slow.

41:34So it would look at your pipe, and then water would just be going everywhere for a half hour. And then it would be like, boop. And then maybe it would be frustrating. But they have technologies they're working on that are going to make the models a whole lot faster. So there's a lot of silicon. Silicon is like, right now there's a lot of software in the models where they load it into GPU and RAM. and they're working on hardware versions of that, like complete hardware chips that are real fast. So that will be like when robots become, you know, walking around, you know, that'll be something like that.

42:13Could you tell to develop a car? I don't know. You know, that's great. I, you know, you should try it. Say you want to develop a car and see what happens. Yeah.

42:24Gary Vasilash: No, no, I think this was a great, you know, to set certain parameters. We want to go into this segment, you know, You know, here's the total addressable market. Here's the price point. Here's our parts bin. Here's the plants. Here's the supply chain. You know, what should we do? Yeah, I mean, I don't think it would work, but honestly, every time I try something, I'm just like, that's so creepy. Like, how does it know this or whatever? You know, like, I was trying to think. Well, I was doing something with NFC. So, like, when you use your credit card, it has a radio in it. and uh i was like well let's work on this together and it's just like oh well i started simulating all the waveforms and then i did like fft to verify the waveform and i designed it you know the the coil to capture it and i'm just like you know like i don't know how to do any of this stuff but you know so i i don't know if you could say design a car but like you can you can you can certainly ask it to and see what happens in it and i think the results would be fascinating maybe.

43:25So, you know, so you mean the style of a car, Gary? I mean, I mean, conceivably, I mean, if it knew everything, it could develop the whole damn thing.

43:35Gary Vasilash: Well, look, I've talked to designers, you know, and their, their teams are absolutely using AI, but they're using it as an idea generator. They're not using it to actually design the car. They're, they're coming up with iterations and, and other ideas, but it's probably only a matter of time before they go boom we're going with the ai design yeah i've seen some ai car designs in there they're generally pretty good until you start checking the fine details on it where there's you know door handles where door handles should not be and right that sort of stuff so then you just turn it over the meat meaning us and they can what you do when you when you go there with that right you um the models are kind of dumb they don't know i mean they know a lot but like they don't know what you're doing so what you do is you kind of explain you write down what it takes what do you want like how do i like if i designed a car i probably have a bunch of rules of thumb that i did like you can't have angles like this or angles like that and then you put it and they call it a skill but all it really is is it's instructions to the model to follow, you know, and it knows a lot, but then you kind of customize its intelligence by providing all that information.

44:54And each model has kind of a, they call it a context window, which you can fill up with all these instructions like that. So like if you were an expert at design and, you know, maybe manufacturability of the design, you could author a document. And that's kind of like your IP, I think, on how to, you know, guardrails basically for the for the ai to follow to to you know to build like um for example this is a this particular printed circuit board is completely ai generated and built in china i did not build it i didn't solder it i just had ai i told it what i wanted it's actually the simplest board you could possibly make because it just connects the wires and the connectors but i just wanted to start out with a win uh so in the in the document that i provided to the ai to design this uh i provided these instructions that you know our our technical people would know like you know no right angles or you know this size board or um you know uh double check with a like a different agent, things like that.

46:05So you can develop that to apply to what you're working on. And it can be shared within an organization. How would that have been made otherwise? I would pull up CAD, and then I would pick these components, and then I would draw the connections like in a schematic, which would be like a CAD software. and then I would size the board and then I would... They don't really have a good auto router in the software I use. It's a freeware called KECAD. But you just, you know, you draw where the connections go on the board and then you send it to a board house. Usually it's... Most people use a China board house if they don't care about the IP or, you know.

46:56but uh this is a simple board but yeah before it was that i just did it because i i needed it and i just was you know like can it design a car you know that that's what you ask um you just kind of poke it you poke it like can you you know design a car for me and just see what happens and like is then you know like and then you kind of learn the limits you build the trust with it you know if it does a terrible job you know there's there's a lot of great stuff on on x where they're you know they have a board where the connector is backwards and you know that, you know, they manufactured it. Uh, but that's, those are mistakes that I did, you know, and when I designed boards when I was younger, uh, which I don't really do anymore.

47:36This is just kind of something I just, something I'm aware of that, that I poked AI with to see what would happen. Gary, my mind's going back to industry days at auto shows where you had hordes of engineers with tape measures and stuff like that underneath cars. Imagine with a camera, a microphone and one of these things here the kind of damage that someone could do i feel at an auto show learning about other vehicles improvement is not damage yeah what kind of info they could learn with a camera microphone in one of those things that you know going around an auto show and getting under it's amazing what you'd get out of a picture like that video i showed you i took a picture of their infotainment boom yeah instead of you know maybe going after

48:22Gary Vasilash: an entire car go after chunks at a time like i i still like this idea of commonizing commodity components but even in any automaker could dump its whole bill of material into an ai agent and say we're developing a new car pick out the parts that's in the parts bin and start putting stuff together for us i mean it you know just take chunks at a time and then end up doing an entire car yeah I mean in our production you know that our vendors have APIs which you know a good example would be there's an electronic component vendor named DigiKey and so when we do like a build we built like a thousand of these we have to buy we have like a list of bill materials and you know one of the things we're doing is we're building software that connects to an API it's a way where that vendor provided a way to do everything you would did if you called and ordered something, you know, but through a computer, that's, I would, that's application, you know, interface to them.

49:31So with AI, we can build, you know, when you do a build, we can, we can actually integrate their system into our system and buy stuff, you know, or look at other systems. So that, that would be an example of, you know, So I guess.

49:48Gary Vasilash: So where do you see this going, Dave? This is unreal, but you're lacking here. Well, I don't know where it's going. I don't know. And that's a little scary. For us, I think what I see us doing is becoming closer with our customers and being able to provide software. a lot of our customers are just too busy maybe to learn AI. Maybe the first step is we do training and we do what they want. Applications engineering is going to be more important where a customer will be like, I'm working on the seat. I don't have time to mess with your gizmo. Make it work for me now. We do it.

50:38We'll have a lot of value around that. what would be great is if our customers really embraced it and they used our software so our software VehicleSpy it's used by more than 10 ,000 automotive engineers and you know one thing we would think about is well software is dead, how are we going to make money for that part of our business and we're basically turning VehicleSpy into a library for AI so we provide We still provide vehicle spy because it's still important for a lot of people. This is going to be a transition. So we have from where we are now and then we have where we're going to be. And then there's all this chaos in the middle.

51:22So you don't want to create more chaos than there already is. So we provide a path for our customers to use what we have now and also to jump in to AI. and like I know the day I really jumped in is I was trying to do something I was I couldn't get it work it was it was just like building some software which that means is I have all the list of the software code and I have to make it into actual software and I just couldn't figure it out for two days and I'm just like you know screw it pull up Claude for the first time and then it just working and I was just like the light I just you know the light turns on everyone is going to have a moment where that light turns on and it's great to see it I've seen it with a bunch of people people have seen it happen to me people were talking to me about AI before it turned on for me the students we sponsor a lab at Oakland University Dr.

52:24Hanna one of the students there, Adam, he was just great at it he was showing me stuff just blowing my mind every software he works he just replaces he just replaces and so i kind of got that idea from and uh and uh my feelings originally um because we we had an event uh so so we were building this with the lab and internally and we actually did a work retreat in miami and the goal of our student team i was there with them and uh we were in a bigger bb is to to write a video game for the or a video game system for this by hand uh based on uh this existing uh video game system and it's it's a decent project fun project you know video games and then adam just you know gets it done the first day it's like now what and i'm just like i was so angry so like you know because it's like i didn't really what the light wasn't on my head it was like it was like it's kind of feels like cheating and like you know i see that in the people the developers that entropid they're just like you know i've seen it like i i understand it i understand it uh and uh yeah you just have to that everyone has going to have to go through that moment those moments where like there's like disbelief and then it's like it can't be real and but i'm looking at it it looks real to me you know i don't know just

53:53Gary Vasilash: yeah yeah you know but going going back to what you were saying earlier in the show about the self-healing cars, self-repairing cars. Fix yourself. Fix yourself. I mean, this would alone be huge. You know, last year, just in the United States, automakers spent$30 billion on warranty work. $30 billion. And I'm not saying that this could get rid of all of it. You know, I mean, if a casting breaks, AI is not going to be able to fix that. But maybe it fixes the manufacturing process so that never happens but still even even if you took 10 out you're talking billions of dollars i mean our tools as they are can solve a ton of problems that are all related to software right now you know because they can capture everything on the car you just it's you have to add them to the car you know you know that i mean that's my like my dream is oems would add just add our the software that we've been making for years on the car because it's the you know it's the data that they use to fix the car.

54:53And with OTA, we can just fix it. Not only that, think of the tech shortage as well.

54:58Gary Vasilash: Oh, yeah, right. You know, I mean, trying to fix electrical problems probably takes up more, I guarantee you, it takes up more time than changing a set of brakes or whatever. If you can eliminate that, you might have enough techs to be able to do, you know, the nitty gritty kind of stuff. Right, no, no, even if it doesn't fix itself, it can tell the tech, hey, you know, I got a short or whatever here. Go fix it. But Dave, you're saying you have this right now. Well, our wireless Neobi, we can record. Well, the OEMs do use it in development. So all of them use it almost. But you want them on the car itself.

55:38Well, I think that is a solution. If you could fix the car in the driveway before. If you make a mistake and you fix it before someone notices, it didn't exist in fact what you're saying you don't even need over the air updates well you should you gotta have it you gotta have that I mean you gotta have that any software no software is perfect when you release it it's getting more perfect anytime software works the first time I feel weird inside that can't possibly be true it is happening more with AI than my software that I wrote But yeah, you got to have OTA. I mean, yeah, I just, how would, you know.

56:28One of the things you were saying, I think, before the show, which I thought was interesting, was that like what you're looking at is a way to not get rid of people, but make people more productive so they can do more things quickly and correctly. and so it isn't you know the scary boogeyman of of ai coming in and replacing people yeah i i try to be optimistic about it and and and my plan is that we do more for our customers and we do things that we weren't able to do and we're more competitive um you know with with our our competition and and and we help our our customers succeed like uh you know i i think that's our plan now can some i mean people talk about like you know can you create like take auto line for example like can you create a virtual version of yourself and just you know you could you could write software that just produces your content and see that that's my dream that's my dream actually well i we can do it like virtual john we gotta talk i mean you can i mean they have things called software factories like it's an idea where you know like in our software we can just put the customer types in the bug and that gets, goes up to the cloud.

57:46You know, the AI has access to our source code. It figures out the bug and figures out how to test it. And then it sends it back down, you know, like, I mean, that's totally possible. I mean, it's just a matter of doing the work. And, you know, it's possible for a car. I mean, why wouldn't it be the same thing?

58:08Gary Vasilash: We're going to have to wrap up very quickly. Anything else that you wanted to bring up? Have we missed anything here? No. I don't know. Dave, it's always fun to have you on here. I mean, it's... Yeah, I was so excited because I was like... You've got my mind going in a million directions. This is the biggest thing that ever happened for our customers, especially the ones in Detroit, because they can compete with the Chinese OEMs. You don't need 10 ,000 people or 100 ,000 or a million. What is it? I don't know. Yeah, BYD's got a million employees. You don't need that many people anymore to get what you want.

58:36Gary Vasilash: BYD has more engineers than there are in all of Japan. Yeah. And we're ready to help. We're ready to help our customers. So, you know, yeah, I'm excited to see what happens. Yeah. So cool. Okay, we're going to wrap it up. Sean, great having you on the set. Great being on. Gary, we'll keep on doing shows like this. This has been awesome. Want to thank all of you for having tuned in. AutoLine After Hours is brought to you by Bridgestone Tires, solutions for your journey.

59:10Thank you.