Michael Wiesinger, Kodiak AI | Kodiak AI’s Sensor Fusion Future

(upbeat music) - Hello and welcome to the Freightvine Podcast. Your source for all things freight transportation. I'm Chris Kaplus, Chief Scientist here at DAT Freighton Analytics, and this is our second episode in the Autonomous Freight Transportation Series this fall. Today, I'm looking to be joined by Michael Weisinger, the COO of Kodiak AI. Now, founded in 2018, Kodiak took a different approach than most other AV startups. Rather than relying on high-definition mapping, they developed a single AI driver system that fully utilizes and fuses together an array of different sensors to include cameras, radar and lidar. This essentially lets them operate anywhere. And so they were able to expand into three different verticals, long haul trucking, which is the primary one most AV startups started with, but also defense applications and industrial operations like sand hauling in the Permian Basin. The industrial and defense verticals obviously need to be able to operate environments with limited and constantly changing infrastructure where high-defin mapping simply isn't possible. Now, there are trucks in the industrial sector are operating without any humans in the vehicle while the over-the-road trucking segments still require safety observers. In our conversation, we discussed Kodiak's drivers a service model when this was what allows fleet owners to integrate virtual drivers into their existing network because Kodiak believes that autonomous systems will augment rather than entirely replace the current driver workforce. Michael also explains their statistical safety case framework as they prepare to take safety observers out of the cabs in their long haul trucking sector. So after my conversation with Michael, I'll provide a truckload market update. So let's get started. I'm Michael, welcome to the freight fine podcast. - Thanks, Chris. Thanks for having me. I'm excited to be on your podcast. - You've done a million of these. You know, being in your space, there's a lot of interest in what you guys are up to. And so what I wanted to do is first before we talk about Kodiak, let's talk about you. How did your path get you to where you are now with Kodiak? - Yeah, absolutely. So I actually spent most of my career in the automotive industry started very early on working for a digital. Everyone in the industry will know that digital is a subsidiary of Daimler Trucks. And so I worked there, then moved from operational into consulting, consulted for a Boston consulting group. And during that time, was always affiliated to automotive players, mostly in cars, not in trucks, but I consulted for most Germans, really most European OEMs in Europe, North America, South America, and mostly on network optimization and supply chain topics. So naturally you already see that there's a fit with autonomous trucking, right? Then before I actually moved to Kodiak, I also launched a startup for UPS with another and with a team from PCT Digital Ventures, where we actually stood up by the housing network and marketplace that matches merchants with warehouses. So again, in the transportation, kind of there's supply chain space. And during that time, I got extremely excited about building, about building something from scratch. And then the opportunity came up to train Kodiak and this amazing team and there wasn't able to pass it up. And then we are very excited that I made this issue back then. - That's, you've had a couple different transitions, but first I have to ask, what made you want to be a paramedic? - Yeah. - So Austria still has mandatory, either civil service or military service. - Oh, I didn't know that. - And back in the day, I was actually one of the last cohorts that had like a 12 months mandatory service. So I was like, you know, had to decide between military and civil service and I decided for a civil service. And I will say, and I did this for one year full time right those 12 months. And then I kept doing it for 12 years, while on chair really, if a certain amount of hours that you have to do per year, a certain amount of training you have to do per year. And to me, it has been a great school for life. Mostly because you interact with people from different kinds of lives, from different, you know, completely from kids to senior people, rich people, poor people. And I think it really teaches you a big deal of empathy. And I always look back, you know, since moving to the US, it's a different system here. I didn't continue it, but I always look back at that time. - That's interesting. I have a brother-in-law who's a surgeon. And he talks about, do you ever see a TV show, the pit? - Yeah. Can you watch it? 'Cause he says, you know, it's a little too real for him. You know, you have to need that after he's 20. He sees it in his real job. - Yeah, I tell everybody, don't watch those things because if you understand a little bit, you also know that most of the things are not actually accurate. And so it's a little bit hard to deal with it. - Yeah, yeah, yeah. So, but another transition you did is going from consulting to a manufacturing company. How hard was that? 'Cause it's sometimes people find it as a hard transition. There's a lot of golden handcuffs when you leave consulting, but then again, you're not on the road four days a week. - Yeah, I mean, it's just a different lifestyle for a shoe, but it's also a different mentality, right? Consulting is, and you know, I look back to my PCG times with lots of gratitude and that I learned a lot and it was a great time. But at the same time, you're recommending, right? And you're not actually implementing your bill. And so what I really realized, I want to build and I want to be responsible and accountable for the things that I'm building and not just, say, yeah, this was a good recommendation, or this was a bad recommendation in the end, right? And that really made me excited, excited. Now, what's different? - It's definitely, you know, PCG's large company at this point, lots of structure, lots of support tools, versus they're moving to a startup back in the day in 2019 when I moved. Codec was still fairly small, right? So-- - It was only one year old, right? Was it started the year before by Don, Bernat? - Exactly, it was founded in April 2018. And, you know, one year in, you don't have a lot of processes, you don't have a lot of tools, you don't have a lot of infrastructure. So you can't imagine, I came in and was like, okay, let's build those things, right? Let's send up infrastructure. Let's make sure we actually make ourselves ready for employing a product and scaling a product in the long run. - So one last transition question, 'cause now you're the COO for seven years, your VP commercialization, you're on the commercial side. I always think of commercialization external up to COO internal. Did you see a transition there? Was it big change in your responsibilities, or is it just more of the same? - Yeah, so maybe to quickly describe that you're track to be, I actually started the CODEC as a product manager. - Okay. - And so back then, I did product management, but I also of course supported Don and the executive team with strategy and topics and so on, coming with my background, right? That was natural. And then at some point, I don't remember which year it was, but, you know, I, Don and I had a conversation, it's like, hey, we need to start commercializing, right? Because we had great technology, but we weren't yet like selling to customers. I said, okay, we have to, as Don was like, okay, build it. I'm right, do it. And I basically built that commercialization function hired an amazing team while still being responsible for product. And so as we went along and as we grew, more responsibilities were assigned to me. And so at this point, I actually still lead the commercial side, on commercial side, the product side, program side, as well as the operations, which were then added at some point, you might remember James Reed, who was our COO at some point, and moved on, and now it's actually the chairman of our board. And so I was reporting to him and when he moved on, I basically moved into the chairman position. I basically took on the responsibilities from him as he has built the trust with me. And yeah, so it was really more of like a natural progression than it actually if didn't responsibilities all at once. - So it sounds like it's more, it's additive. - Yes, I agree with you, okay. It sounds like a startup, sounds like a small company. - Six years, eight years in now. - Eight years. - Yeah, eight years in. I mean, you know, we are public companies in September 2025, but we still operate extremely nimble, extremely fast. We don't want to kind of lose that culture of being capital efficient, of being fast and developing things. Of course, doing it the right way with structure, but still maintaining that speed from a startup. - Okay, so give us an overview of where Kodiak is now. Where do you stand in terms of the commercialization and eventually the scaling of the technology? - Yeah, 2026 is an extremely exciting year for autonomy, right? If you look across the board and specifically at Kodiak, so Kodiak operates in three radicals, long haul over the road trucking industrial, some people call it vocational, but for us we call it industrial and then in the defense space. And I'll walk through this three sectors to give you a good overview. Long haul trucking is, we have been operating a fleet of trucks since 2019 for customers like one or two behind Martin Brower, real transport, and many, many more that we actually have been holding faithfully. That's the segment. We are we still operate with a safety observant in the cap, but we are getting very close to actually taking that out and we can talk about that later. - Yes, yes. - And in the second segment, industrial, this is where we have operated since December 2024, fully without any human in the cap. So what does that mean? We actually went out and kind of signed a contract with Atlas Energy Solutions, which operates trucks in the premium base in two hull and sand for fracking. So basically bringing sand to the outside. And in 2024, as I said, we actually deployed the first trucks to them, but what's different, and really this is absolutely different compared to what everyone else is doing, This is actually a model of them owning and-- operating the trucks. So it's true. Sometimes people talk about transport, there's a service, and driver's a service. This is a true driver's a service. We are day maintained, day dispatch, it's their DOT number and so on on the side of the trucks. We provide the virtual driver to them that does the driving. And so we deployed the first two trucks in December 2024 and have now scaled to 35 trucks, which to our knowledge is the largest deployment, real deployment of driverless trucks in the world. And then in the defense space, we have been working with several branches, you know, started early on with the Army, now also working with the Marines on the program called Nemesis, but also have forged partnerships with some of the key players, think about general dynamics land systems, which we announced earlier this year to really build a foundation to deploy our system in ground autonomy applications in the future. Okay, that's great. So the three big verticals, let me talk, ask questions about them backwards. So I'm going to start with the defense. Is that the least developed as far as commercialization where it is now? It's more prototyping at this point or do you have anything actually in production or in commercialization? Yeah, I think if you think about where the Department of War and the different branches actually are right now in the autonomy journey, it's still early on, right? They're still like they're trying to find the right use case. We do they want to deploy first what the product will like exactly look like. And so yeah, I would agree that the contracts are still early days, right? It's not yet okay. We're scaling to the hundreds for thousands, but there is lots of traction. If you look at the budget for 2027, right, there's a huge jump in in budget that this actually dedicated towards autonomy for for military applications. So it is an exciting time. We repeated we have a opportunistic because it's harder to plan. It depends a lot on, you know, what happens, how decisions are made and lots of changes happen. So it is exciting, but again, we have a opportunistic. Yeah, and it's it's kind of the opposite of the other two, which I certainly for long haul trucking, which you're going to be on highways, well structured, well known for the military. I've done a we've had a fellows program here for 20 years. We have kernels come in and they talk about the need for autonomous, especially in the theater and because but in those situations, there is no infrastructure. So it's a very different problem, right? It's more sensor-driven. You can't have a map of the area. So is it a different technology or is it more of the same technology just slightly different parameters? Now, I love what you picked up on there. Actually, as Kodiak, we operate one single AI driver for all use cases. And that is actually one key differentiated and a key enabler for us and why we are able to serve three use cases or verticals at the same time. So we are very early on dawn and and the team of founding engineers have actually made a variable decision, which was to not use high-definition maps. They were several decisions that was one key one, not using high-definition maps for any use case or for any vertical. And back in the day, of course, high-definition maps were coming from robot taxi developments, where everyone was using it because, you know, that's how things were done in 2013. But they said, let's not do it that way. And that has enabled us to be much more flexible and be much more nimble in what we do, where we can deploy and not have like a rigorous six-month process where you have to map something to then to then basically put a new product out, right? And that is actually across all three verticals is really paying dividends today. But it is in industrial and defense, of course, a massive game changer. Think about industrial, where we have again, hauling sands to well set. Well, it's typically our operational for somewhere between three to five weeks, right? So every three to five weeks, you need to change your route. And obviously at this point, we operate with a fleet of 35 trucks. We or Atlas operate several routes at the same time. And sometimes they say, hey, in a couple days, we need to have route X ready to go travelers, right? So if we would then say, oh, yeah, let's just create this high-definition map or England's come back in five months while we would have missed the whole route at that four months. And so this approach is really the enabler for actually operating in the pyramid. Same in defense. Think about, you know, going to Ukraine today and saying, never mind, we are just mapping here and we will come back in two years with our fully defined map and then we will actually, it just doesn't work, right? So really not being reliant on those high-definition maps is a key enabler and key differentiator for us. Is that does that cause a corresponding negative or a downside? Is it mean do you still have that? Because I'm trying to think for maps for the highways that you're you're able to use that very quickly. Does that slow the adoption rate for like the long haul over the road or is it not having effect? No, not at all. Quite the opposite. And people ask us like we have customers that ask us, hey, can you do this new routing two weeks? I think we are probably the only company that can say, yes, we are actually able to do that. So just to finish that thought, early on, high-definition maps were to some extent useful localization, right? And back in the day, you know, we weren't very out today with all the advancements that have been made. And so we had to do localization slightly different. And it was a little bit of a gamble, maybe early on, right, to go a different route. But again, it worked out. It pays is paying the dividend across all sectors right now. So, so what is the technology then on the vehicle itself? Do you have video, LiDAR, and radar? Do you have a combination of those? Absolutely. So we deploy something that's called sense of fusion. Sense of fusion basically means you bring all kinds of sense of modalities together, which is cameras, which is radars, which is lighters. All of them have benefits and and some areas where they are not as strong, right? Different weather, for example, different distances. But you want to bring all of them together if you want to develop and deploy a real driverless product. Again, you want to leverage each sense of strength. So you called it sense of sense of fusion? Is that okay? Okay. Sense of fusion is taking the inputs from cameras, radars, lighters, and kind of using it together to an accurate picture of the environment. That's a really interesting technological problem then, because you've got three sources of the truth, and you've got to decide which one is true. That's got to be interesting. Is that done? I assume in the software. Exactly. They're on onboard compute handles all of that. Yeah, and so the rise of the improvements in AI and the just compute powers certainly made that something that is more usable, but same with the defense then. So do you see this being used more in like an industrial back, back of lines type operation, or do you see it use on the in the theater of content in the on the battlefield? It definitely has applications in both. I know you're going to say that because you're selling, you're commercializing, right? Yeah, I mean, if you if you talk to the customers there, they think about tactical use cases. They think about more like logistical supply operations use cases. As I said, I think the armies with general old branches are still exploring there. They actually believe it has the biggest impact quickly. That's like something that they try to understand through all of these programs, but certainly you can think about autonomy really everywhere in the world or everywhere in supply chains. And so then in the industrial for the Permian basin, was that something when the company was foreign formed in 2018, 2019 was the initial thought to go to these other things or they just opportunistic at the time? Yeah, it was more serendipitous and then a little bit coming to us. We did in 2019 focus on long haul or whatever trucking, right? In 2021 was the first time we started working in the defense or public sector that back in the day that was actually an EF was project. And then at some point Atlas came to us saying, hey, we see what you're doing on the highway. We see what you're doing like off road really? We are somewhere in the middle. And so can you make something work? Because we do have roads. It's like a structured network of roads. Some of them are paved, some of them are collegiate clay, some of them change every day, some of them are kind of more steady. And they said, can you make this work? And we said, well, let's take a look. We certainly have the core foundations. We certainly have set up our system that way. And that's what we basically did in 2024. 2024. We brought a truck there. We tested it. It really worked out of the box, which we were all including Atlas really excited about. And then we said, okay, how do we make this product? How do we deploy? And we focused the next 12 months on actually doing that until we deployed the first two trucks in December 24th? And you described as a driver as a service. So Atlas owns the vehicle, has the DOT license, all that stuff. You're prying the software. That makes it pretty sticky at this point for you guys. I would assume it would be very hard for Atlas to switch to another provider. Is that a fair statement? I mean, yes, that is absolutely fair and accurate statement from our perspective. And think about not just Atlas. Think about all the road trucking. There's a huge driver shortage. We all know about that. 2026 with, you know, non-domicile driver issue. There's actually the demand for autonomous trucks is truly rising by by the day almost. And so, yeah, do we believe autonomous trucks and our system in general will be extremely sticky? Absolutely. Do we also believe that the drivers a service model, as you described it, will help with that even further because they own the truck, they operate them trucks. They need the drivers and, you know, we provide the driver. Yeah, we absolutely think that way. So let's say Atlas has a truck and then you're providing the drivers of service, can a human actually go in and use that truck? Absolutely. You know, right? Okay. So it's not, the hardware is not designed to not allow a human in there. It's not cabbless. No. It wouldn't make sense as every stand today, right? There are scenarios, let's say they want to take the truck and drive it into a maintenance fee, right? What they want to take the truck and do something completely different than we designed together. They want to have the ability. I know, you know, back in the day, there was like concepts maybe five or ten years ago at this point of a cabbless autonomous truck. It has never materialized because I think everyone understood that there are just situations where you still want to put a driver in. And so we believe right now, this is still the best approach. Yeah, it's funny. As I talk to more and more companies in the space, they realize and everyone's realized drivers do more than drive, right? There's these other activities you need to do. How do you get those do the robot can't do it by itself? So what about for other like mining operations, because that seems like another natural fit as well as agriculture, because those are, if you go from a battlefield where you can do anything, right, to a controlled environment like what you described in the Permian Basin or mining or agriculture, and this seems the most, the riskiest as far as involving other people would be the long haul on the highways. Have you looked into mining as well, or agriculture aspects, or are those two different? Yeah, when we talk about our industrial vertical, we generally think about three use cases, and I will explain why. One is what we call whirl and gas kind of fields and everything that's happening in India, right? The second one that we have actually worked on and I have deployed a pilot in is logging transportation. Yes, it's actually huge demand, because those are extremely remote locations where we are hard to find drivers, reliable drivers, right? So this is not a use case that is huge. And then third, we think about minimal transport. So minimal transport comes close to what you consider mining, but isn't necessarily mining. Yes. Think about taking goods from a mine to a port, from a mine to a real yard, right? Those sometimes are 50 miles, can be 150 miles, can be longer distances, think about Australia, right? I want to say Australia seems to be a huge market for you guys. Absolutely. Yeah, we are exploring all of that. Now, to answer your question about mining, mining has a very different software requirements. It's actually very simple, right? And you have seen some companies deploy some version of its remote control, but anyways, write some version of autonomy, and we could absolutely deploy it there. Our stack is almost oversophisticated for actually deploying on mine. And so we need to be smart and really think about, where do we get the biggest bang for the buck? And that is in other vehicles, in other use cases, and it is specifically in long-haul trucking, right? The stack that we have built, the sophistication that we have built is actually good enough to work on highways. And so this is the biggest ham, so naturally, this is where we will put our focus. That makes sense. So last thing for an industrial before we get to the long haul, the main reason why I'm here is for the industrial for when they're in the permeability, do you have a remote driver? Is there someone monitoring it or is it pretty much you say got to get from A to B and you let it go? Now, it's actually the truck is truly on the own. We do have a 24/7 command center and the truck can call for help. If there's only, if there's any issue, if the truck gets stuck or let's just assume there's a tire blowout or something like that, right? We have a 24/7 command center that can respond to any kind of support the truck needs. And we can then triage, is it something on the truck and we will tell Atlas about it, is it something with our system, we can kind of take care of that, right? Yeah, yeah. So there is a command center, but it's not like there's a remote driver, that is important. It is truly wild. I have been out there a couple of weeks ago and, you know, we obviously I know that we are working on this every single day, but then you see a truck driving by no one in the cap. It doesn't get old. I'll tell you this. You should absolutely comment and visit does because it's so cool. Next time I'm in the Permian Basin, you know, if you ever kind of trying to go on vacation in Permian Basin exactly, it's a vacation spot. No, but it's so, how does that work date? So it has to have some kind of map, right? It has to have, it doesn't have to be high depth, but it has to have some kind of map to route or else it's just going to go across the countryside and the point. So is it a rudimentary map that it then uses its sensors to stay roughly in line with? Yeah, we have a very accurate description actually. It is a, we call it a light map, but we sometimes describe it as mostly a routing layer, right? That's what people understand. Think about Google Maps, right? Yeah, it knows, it knows where the roads roughly are. But the thing is, especially in the Permian, he cannot fully rely on it because the map should like we changes, right? So we still need to use all the sensors and pick up if there were any changes and we can actually live update those, right? Because it is important, even though the map or the kind of the road might have had this edge, as we've said yesterday, today might be different because there was a wind storm or whatever it was. So yeah, there's a rudimentary map, but really the truck is acting based on what it's seeing life. That makes sense. And I don't know, we're focusing on TV shows here. Do you watch Landman? Have you seen that show? Of course. I'm not much of a TV watcher, but getting into the Atlas business and working with them, I felt obligated to do it. And I will tell you, I've heard everything from some people say, "Yeah, it's pretty accurate." Some people say, "Yeah, 85% is made up." But I think it's an angel regardless. Yeah, yeah. So that's really interesting. Okay, let's get to the main reason for this podcast over the road, Longhall. So what was the big challenges? Because it's kind of what you were originally starting to do, but these two other ones were the versions that very profitable. I assume that it's a revenue source and it's really interesting. It's helping you be better on the Longhall trucking. What are the big challenges that you're facing now for Longhall trucking? Because I think it's the only one you said that still has the observer. Correct. We still have to serve in the cab for our Longhall operations. Right. So, you know, we are currently working through what is called the safety case, we just announced an update to our autonomous readiness measure. Which is basically how we measure ourselves to be ready, or to have closed our safety case. We announced 93% per day end of august up from 91% by the end of July. So you see a really, really solid progression. We have communicated that we are targeting the end of the year to actually undo the first runs without the safety observing the cap and we believe based on that trajectory that I just laid out that we are really on track. So what is the safety case, though? I didn't quite understand exactly what that means. Yeah, safety case is a buzzword that's used a lot and think about it in very simple terms. It's a statistical argument that says how safe your system is. So you kind of have a top claim, top claim can be, we want our system to be at least as safe as the average human truck driver. And then you need to find statistics to actually back that up. So there's different things you do, safety case has like all kinds of hardware related things. From the truck perspective, think about functional safety, from a software related thing, it's safety of the intended function. And many, many other aspects ready, also need to have the operational process, the cybersecurity processes and all of that. But generally speaking, what we deploy, I think this is something that resonates and people can easily understand a probabilistic risk assessment, meaning when we assess how well we handle a certain situation, we've worked with a model that's used and was developed actually by NASA, that's used for nuclear reactors and all kinds of safety critical industries. And basically says, okay, there are like thousands of things that could happen, right? And everything has a certain exposure and a certain severity and a certain probability. And so you basically have those values and those values you have comparison, right? From, for example, we work with the University of Michigan Transportation Research Institute and you basically then came up, come up with values that compare against human values. You take that from your on-rope data, your simulation data, your different OLED analysis, different other scenario databases, right? So really, you basically say, okay, I want to make sure that in all those cases, including some unknowns and knowns that we just don't know yet, that we are safe enough and that we meet our safety bar. That's really what the safety case does, that's a lot of work, right, working through all these scenarios, but that's exactly the work that we are currently working through. Now mind you, in the industrial space, we have worked through the same safety case, right? We have worked through getting to 100% in December 2024. So we have built the muscle on what we need to do, how we run through the process. And that's what we are now applying to the over-the-road business. And that's what we are aiming for 100% by the end of this year. So you're at 93%, and that's not like 93% of the times your trips won't have an accident. It's much higher than that, but the 93%, so that means 7% is there. Where is that 7% coming from, do you think, is it technology, is it the redundancy of the hardware, is it the regulatory, what is causing that 7% that you want to shrink down? What do you think the sources are if you did like a failure mode analysis? So what's important is that the 7% doesn't mean that there are 7% of things we wouldn't handle. That is not the case. It basically means that there are 7% of things left that we haven't fully validated yet. So the technology might actually already be there, but we still have validation work to do. I'll make an example on the validation work is most the edge cases. Think about, I don't know, there's a vehicle on the shoulder that has a ladder sticking out into the right lane, right? And our truck needs to handle it by, you know, I am making a lane change if possible, slowing down, coming to a stop. So that would be one of the edge cases there. There's other edge cases like that, but it's really not anything cool technology anymore. You mentioned redundancy. Our trucks have redundant brake gain redundant steering redundant power, right? There's no cool technology element. It's really just about validating all this in the area that could happen. So what there's a recent FMCSA ruling about the triangles, which is kind of a, it seems like a silly thing if there's an accident, who's going to put the triangle out? And no different companies ever are trying to arrange that they use some other signal. With your current safety operator there, they can take care of that. Is that part? Is that figure into the safety cases? Or is that something totally different? So the morning triangle, well, first of all, I mean, going back to your initial question about me being a paramedic, I'll tell you, all the accidents I have attended to have never seen safety triangles actually been put out, then they should be not as a second. I think it's unsafe a spot for anyone to be on the highway to walk like next to the fog line with this, this triangle in your hand, right? So, you know, we can all debate about the whole safety triangle on topic. But importantly, the autonomy industry really worked together. Everyone came together and said we need something better, because first of all, it doesn't make sense. Second, it doesn't actually work for a truck that doesn't have a human in the cab. So what we did is we developed an approach to basically put a warning beacon super bright on the top of the truck, in our cases, on the top of our sensor parts, and then applied for an exemption. And that exemption was granted now many, many months ago. And that's basically the way we, I don't love to say the circumvent, because I actually think it's the better approach. But I will call it circumvent for now. But that's how we don't actually need to have somebody to put out the warning triangle. That makes sense. Let's switch topics a little bit and talk about Bosch. So, okay. And can you talk about that partnership and where that fits in the evolution of what Kodiak wants to achieve? Yeah, absolutely. I mean, in general, we are a company that approaches things from an ecosystem perspective. We don't believe that we need to build everything ourselves. There are great companies that have built manufacturing capabilities, that have massive supply chains, that are really, really good in certain things. We obviously haven't had the time and the rigor to develop those yet, right? And Bosch is one of them. Bosch is one of the, if not the largest, automotive OEM, extremely good reputation. And so we said, hey, we know what we need to do in the future to deploy and scale. They know how to manufacture. They know how to scale. They know how to build the right supply chain for the components that we need. And so that's really the partnership. We basically work with them on evaluating the components they have and then making it into an autonomous, fully autonomous system to get on. Okay. So, if I look at some, the path for companies in the space, there's some proof of technology, proof of concept. Then there's pilots, the proof that it works in operation. Then there's commercialization. We actually get paid for it. And then there's scale for this. And so where do you, is Bosch the thing that's helping you go from commercialization to scale? Okay. And so where is the core expertise of Kodiak then? Is it in the software? Is it in the system that, what do you call the the sense sense or fusion? Is that the core essence of it? And then you're using partners for manufacturing of the equipment around it? Yeah, we'll say it's several things. One of course is of course the software, right? The AI driver, again, having built a single AI driver that works across use cases that can be trained. One use case helps the other helps the other and so on, right? That's of course core for you have done. Second is how we have actually designed the system around it, meaning the hardware, the sensors and so on. We have done things a little bit differently than most of our competitors with putting all the sensors in what we call the sensor parts. Meaning the cameras, the lighters, the radars that we discussed early on, they're all in one which is structure that can be easily taken off and replaced in case something would go wrong, which is good for uptime and reliability, right? And so that design, including the software, are really our kind of key IP if you want to. Sure, sure. But the manufacturing of those components and sub components and sub assemblies, that's not core of what we do, right? That's where we want to work with suppliers, Bosch, one of them, but of course we work with many, many others. So if I have a fleet of trucks and if I want to work with Kodiak, can you retrofit my existing vehicles to be autonomous? Yeah, our system is platform agnostic. So basically, we have designed the system to integrate into any truck into any model we have actually at this point deployed several different truck makes and models, including the cap, which we recently announced on the Western stuff within an X, which is a similar platform that Atlas picked for their next 65 trucks that they want. I just really designed it to be modular. And so yes, the answer, the answer is yes. Now, when you say retrofit, would I or we as a company want to take a five-year truck with 500,000 miles and then put autonomy on top? I don't think that's the best use of the autonomous system, right? I think it's always better in a new truck. But in theory, it's possible to upfit any truck without a system. So whenever you make a design, you're an industrial engineer, I'm an engineer. There are trade-offs involved. And so what is the major trade-off that you think the Kodiak made by making it so that you're agnostic to platforms? Does that mean? What does that mean? Is there a downside at all for that? I wouldn't call it that downside. I mean, I think it took, there was more owners on us to actually make this work, right? The reality is the OEMs don't yet have the ability to manufacture thousands of those or 10,000 of those trucks, right? And so since we live in this reality, but we couldn't wait for that to happen, we basically said we have to find our own path and we have to develop a system that we can integrate into any truck. Does that take work? Absolutely, right? You have to make different design decisions. Sure. Especially when you say we want to be able to integrate and take care. We want to be able to integrate in the European cab or what you mean, right? So yeah, you make some trade-off and some different decisions. Sometimes it takes more work to actually say, yeah, this is a good solution. But that's really the approach that we have to. Yeah. And so what is Kodiak's business model then going forward? Right now you describe drivers of service to Atlas. Do you see the same kind of model going to carriers? Now that we're going or would you sell this as a service? Do you see yourself Kodiak as a provider of the trucks and becoming a carrier? How do you see this falling out for the long haul trucking side? Yeah, we want to bring the driver's service model to all sectors that we're working. So we have deployed that driver's service model with Atlas since now, you know, over one and a half years, almost two years at this point. And we believe that's the right model for the future also for long haul trucking. So yeah, we work with some of the carriers. I mentioned them early on already, but we work with them and saying, how do we make this a product and the solution that works within your operations, right, within all the processes that you have already set up? We don't want to replace them, right? We don't want to become a competitor to them. I mean, they have decades of experience of how do we acquire freight, how to maintain trucks, how do dispatch trucks, right? I want to be able to partner with them so they can do their core competences and we can bring our core competences, which is the virtual driver and create a solution that is beneficial and driving value for all of us. So yeah, that driver's service is the model for us. Okay, so I'm a carrier and I've got, I don't know, 250 drivers. And so I wanted to start working with Kodiak. So I have some other trucks and I modify them to make it so that they can be autonomous. And then is it up to me to how I use them within my network? Is it essentially you just turn on the services and then they it's up to them to utilize and set things or are you the middle man for them to operate those trucks? It's a tight collaboration. I think autonomy is a product. It's not like you go to a supermarket then buy whatever product you want and you just consume it, right? Autonomies much more requires more sophistication. So the thing that we have actually built with carriers with Atlas is the partner deployment program, very basically have laid out a program that is like sometimes a multi-year program where it says, okay, we start with analyzing via your network would autonomy actually make sense where does it bring the biggest value? Because there are lanes where it has more value, there are lanes where maybe I wouldn't start with, right? Second, then analyze, okay, what does the end to end process look like in the future, who is doing fueling, who is doing maintenance, who is launching and landing those trucks, right? It was supporting the launch and plan of those trucks, then lay out kind of how do we integrate with your systems, typically carry has a transport management system, they have a free fleet management system, right? So we need to integrate with all of those for them to actually have a solution that really works seamlessly and actually brings them the value. And then of course, define kit BIS, define metrics, define the roller, why do I think this is so critical, because I'm convinced that in the future, autonomy companies will not necessarily. the French each other by, oh, you have built cool technology and your technology is slightly better and that's an area where it's this from. I think they will differentiate by how reliable the system will work and how effectively it's integrated into the carriers and fleet operations. That, I think, is key. That's exactly where we have done the last, again, one and a half, almost two years. Every defined, what are all the things that we need to build around the technology to make this a product that we can actually deploy to our customers? And your customers are the carriers or for a shipper if they have a private fleet. Anyone really that operates a fleet is a potential customer. Got it. Got it. So if I look at my network, what are the types of lanes that are more attractive, you think, to autonomous trucking? Is it short-length the hall, long-length the hall? Is it the consistent freight versus the infrequent lanes, or is it drop-and-swap lanes, where you're just picking up and dropping off or live load? Is there a certain sweet spot of a type of lane that you see that a carrier would generally adopt autonomous? Definitely. And we have had many of those conversations with the carrier partners that we work with. It's not that the technology really cares about. The technology is like, yeah, I'm going to do the driving no matter what, but how the way we actually approach this is, what brings the most value to the carrier? And basically, if you look at the landscape, long lanes, it's really really hard to find team drivers, especially for time-sensitive freight. There are some carriers who have kind of implemented a relay concept, but it's always kind of nascent, right? Again, because it's hard. And so we believe autonomous trucks is just an additional modality. Think about intermodal, right? That was introduced. So autonomous trucks and additional modality just like that. And I believe the benefits will be highest. Everything that is long lanes. Think about everything above one hour of service. Because once you get above one hour of service, you either have to lay over a driver where you need a team and all of that costs either time or money, right? So naturally, to me, those are the longer lanes are absolutely the biggest value lanes. Does that mean we couldn't do shorter ones? No. But at the same time, we are working with the carriers to say, okay, you have drivers. They have core competencies. We have autonomy system. That has kind of its strengths. So how can we combine those? How can we create a network for you to gather there? You say, for example, you have drivers on kind of the shorter lanes, maybe even the store deliveries, right? For somebody who has like a ton of store deliveries. And then you use autonomous trucks on the long haul shuttles to kind of compliment that, allowing the drivers to be home every day. And so they don't need to be on the road. So it's really about understanding what makes most sense. There is no more ice fits all. So it's really a collaboration with. I think people are still trying to figure that out. But one thing you mentioned, the hours of service to me, that makes the most sense, right? If you can go longer and you don't need a driver needs to stop a day and rest and all that. But even the bar is even lower than that. We just finished a study looking at ELDs and the average, and I forget how many thousand of drivers we looked at, long haul trucking on average, they only drive six and a half to seven hours a day. And yeah, we say 11, but they don't. They just don't. It's very rare. And so the bar to make it sense, if you can fully drive all 11 hours, that is a game changer by itself. You don't need to go more than that. But then the short haul, I was wondering because I've done some drive along with companies with Schneider, where in Chicago, where you know they're driving five miles back and forth, just moving containers. How does, how does a ton, how do a ton of vehicles do with backing up? Do they? Yeah, I mean, it's absolutely possible to back up. I mean, think about it from a, from a sensor perspective, right? We have 360 degree coverage all around the truck. Again, sensor cameras, radars, lighters. But you've got a 40-foot long trailer behind you, right? I think you're a foot trailer. Right. And the human has the same. So, so no, but I've seen some, I've seen some mad skills with these drivers who, especially they've been doing it for a while back in and it's, it's insane what they can do. Is that a tech, I mean, it seems like a totally different skill. Yeah, the technology is absolutely capable of backing up. Now is, is, again, this multi-stop planes, lots of backing up, even sometimes drivers, is, should we start with that? Do we recommend that to our customers? No, absolutely not, I understand, understand. And, and so is it safe to say for the long haul trucking, you're more of the, the hub to hub, where, however you define your hubs for that, that it's mainly going to be the middle mile, the long middle mile. Yeah, I would say sweet spot. Yeah. That's an accurate statement to say the long amounts. When, when people talk about hub to hub, I want to be careful because sometimes they said hub to hub like several years ago, to mean, it has to be an all dedicated autonomy hub, to a dedicated autonomy hub, creating this like inefficient moves on, on kind of the first and last, but no, that's not what we do, right? We go absolutely dock to dock. We, we go really, you know, from a DC to a store, from a DC to another DC. So that is important that, that we don't misunderstand what hub to hub means. Sure. But yes, do I believe it's kind of the middle miles is where the autonomy will start to scale? Yes, absolutely. It seems like some of the sweet spot area would be for retailers bringing agriculture in from the, say, the Southwester from Southern California, all the way up because that's, I mean, like for apples coming in from the Northwest, they use unit trains to come because they will keep rolling all day long. But being able to bring that on a truck that can bring something and they can cut a day off. That's, that, that seems like that would be a sweet spot. So let me ask you one last question. You've been very generous with your time, Michael, I want to ask one last question. And I'm not going to hold you to this. But how many years do you expect it will be where at least 50% of all over the road, long haul full trucking miles are done autonomously? That's a, that's a good question. You know, I won't hold you to it again. Yeah, no, no, but the public markets might need to be careful in what I'm saying. But I, I'll, I'll, I'll, I'll offer you this. Okay. Just currently roughly three million class eight trucks, right? And that's just assumed for now, we focus on class eight trucks. Got it. It's roughly 200 class eight trucks produced every year. So let's just assume every of those trucks would be autonomous, right? It's more than a decade to replace all trucks. So let's just assume 10% of those that be, let's be conservative 10% of those. We are more than a century, right? So you can do the math on, on 50%. But again, it's, I think your question is maybe informed by some drivers being concerned that there won't be a driving job in the future. And I think that is absolutely not the case, right? Again, I, I see autonomous trucks as an additional modality that has strengths, but I also see the drivers having the strengths. And so I think we will see an augmented network that is really going to be more reliable, more efficient, faster, and basically serve the demands that the, that the supply chain has. And quite honestly, those demands are constantly rising. And if we can just feel kind of that supply versus demand gap without autonomous trucks, I think we have all done a, a, a, a, a, a great thing. You're, you're consulting background showed, you're giving us numbers without really answering the question. That's okay. But, but, but of course, you're, your analysis ignored that we could convert some of the more recent ones, right? And so, yeah. Because I, I, I'd assume, because the, the, the, the answer I took away was a century. Do you really think it's going to be a century? Is I want to be less than that? As I said, less than a year. Because I think the rate of new trucks will be higher than 10%. Right? If it was just, we sent every year, then it was a century. But I think the rate at some point, not today, but at some point will be much higher than 10% of, of replacement with autonomous trucks. And so I think it will be less than a century. But, you know, somewhere between the, the decade in the century. Be here a hundred years from now, unless something dramatic changes. But Michael, thank you so much. I really enjoyed talking to you today. I learned a lot about Kodiak and just the AV space. Thank you. Thank you, Chris. I was blessed I haven't been on. Okay, everyone, stay tuned for the truckload market update. Hi, welcome to the truckload market update for one October, 2026. As you can tell, once again, I'm in a hotel room. So I'm recording this from the inland 26th Journal of Commerce conference in Chicago. It's time of year, conference to conference. But let's get to the truckload market update. In the drive-in market, we saw that contract rates dropped 2%, spot rates rose 0.3%, and the current level of replacement rates for drive-in is 11.2%. That means new contract rates coming in are on average 11% higher than the rates they're replacing. The gap between spot and contract is 13 cents a mile, where spot is above contract, year-over-year change in spot rates, 37.2%, and 11.7% for contract. Temp control. We saw contract rates rise slightly 0.6%, spot rates dropped 1.6%, and the current level of replacement rates for temp controls positive 8.3%. The gap is 13 cents a mile between spot and contract, spot above contract, of course. And the year-over-year change in spot rates is about 30%, and about 12.5% for contract. For Intermodal, we saw contract rates go up 1.6%, we saw spot rates slightly rise 0.6%, we saw current level of replacement rates of positive 7.6% for Intermodal. Market gap is about 5 cents a mile. Again, spot Intermodal is pretty small, so I don't need much pay attention to that. Year-over-year change in spot rates about 20%, and about 11% for contract rate change. That was for Intermodal. Flatbed, finally. Contract rates went up 0.3%, spot rates up 1.7%, we saw the current level of replacement rates at 6.4%, market gap is 2 cents a mile negative, where spot is actually below contract for flatbed. It's so volatile. You never get that this bounces up and down with a plus or minus a penny or two each time period. A year-to-year change in spot rates for flatbed is 33% and it's 20% for contract rates. All right, more of the same, active contract rates for drive-in drop slightly, but everything else increased slightly, you know, one to two percent, somewhere in that range. Replacement rates for all the modes are positive, four to twelve percent, you know, the drive-in is down from what it was. They're all slightly down from what they were. Two weeks ago, the drive-in replacement rates was 14%, temp control was nine, and flatbed was nine. So those of all come down a handful of points. It's still high and the year-over-year increase in contract rates are in the 10 to 20% year-over-year. Spot rates increase slightly, 0.3 to 2%, and the year-over-year there between 20 and 40% above where they were. We're going to start hitting the time of year where the rates were increasing last year, so that year-over-year is going to start shrinking a little bit, I think. And the gap between spot and contract is positive between five and 13 cents a mile, for everything except flatbed, which I mentioned earlier dropped about two cents a mile. Anyway, that's the state of the market, the truckload market for drive-in, temp control, intermodal and flatbed. On one October, 2026, it's a tight market. It's going to continue to be tight. Capacity is scarce. Carriers are making up for the last three years of a really soft market. And I see these bids starting to strengthen up, so it's time for shippers to start strengthening the routing guide, building in some consistency. It's going to be a challenging market throughout the end of the year. Maybe there'll be some easing of this in the second or third quarter of next year of 2027. We'll see. And that's it for one October, 2026. [BLANK_AUDIO]