Best of LinkedIn: Software-Defined & AI-Defined Vehicles CW 37/ 38

00:00:00: provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts about software-defined an AI defined vehicles in calendaring its thirty seven and thirty eight.

00:00:09: Frenness supports tier one automotive suppliers with early stage market validation for their R&D efforts by combining secondary research direct OEM expert interviews and facilitated customer meetings.

00:00:21: you can find more info in the description.

00:00:23: so imagine first second that your trusting a multi ton commercial truck Right, just rocketing down the highway.

00:00:30: Sounds terrifying already!

00:00:32: Right

00:00:32: and imagine trusting it to be exact same open source software philosophy that runs like your desktop web browser.

00:00:39: Oh wow

00:00:40: yeah

00:00:40: I mean on paper It sounds like an absolute liability nightmare.

00:00:43: Totally But right now The biggest commercial vehicle manufacturers On Earth are doing exactly

00:00:49: that.

00:00:49: it's a staggering shift in how this industry operates because the insights we've seen circulating among mobility professionals over the last couple of weeks, they all point to the same unavoidable reality.

00:01:01: The physical underlying architecture vehicles is just fundamentally changing like... We are moving away from this scattered isolated hardware into a highly consolidated software-defined space.

00:01:12: Yeah, and you know that consolidation is laying the absolute groundwork for AI to literally take over the vehicle's core functions

00:01:22: which Is exactly what we're taking deep dive in today.

00:01:25: We really want to cut through the industry buzzwords And look at the actual mechanisms at play here like how does digital plumbing actually change, and how does AI evolve from being just a helpful voice assistant on your dashboard to becoming the vehicle's central nervous system.

00:01:42: And perhaps most importantly what happens to the economics when these new architectures finally hit the real world in the form of autonomous fleets?

00:01:50: To understand the AI we really have to start with physical hardware first because for decades vehicles were essentially, they were like a loose federation of separate computers.

00:01:59: Just

00:01:59: a bunch of black boxes?

00:02:00: Exactly!

00:02:01: They relied on highly distributed domain-based platforms.

00:02:05: so you had dozens of isolated electronic control units or ECUs just scattered throughout the chassis

00:02:12: Right.

00:02:12: So Like The brakes have their own brain and Windows has its own brain.

00:02:16: Yeah Powertrain Has Its Own Brain And None Of Them Were Really Designed To Talk to Each Other Outside Their Specific Silos

00:02:23: Which Is Hitting a brick wall economically right now.

00:02:26: I mean automakers are totally abandoning that legacy setup in favor of centralized compute and you know zonal architectures

00:02:34: And there was a great point on this recently from Alexander Shabe.

00:02:37: Yes,

00:02:38: she pointed out the core reason why this is happening.

00:02:41: automotive software has a massive scale problem Right?

00:02:44: You have every single automaker out there trying to independently develop their own software factories

00:02:49: Which is crazy when you think about it.

00:02:51: It is

00:02:51: its.

00:02:51: they're CICD infrastructure You know, the automated pipeline they use to test integrate and deploy code.

00:02:57: Doing that from scratch for every single brand it's financial

00:03:00: suicide.".

00:03:01: Yeah, Shaw basically argues that for successful partnerships moving forward there has be this harsh separation like a wall between commodity technology... The invisible stuff that doesn't make a customer actually buy your car.

00:03:14: and the brand differentiating software

00:03:16: Exactly, which brings us right back to that open-source truck example you mentioned earlier.

00:03:21: Oh Right yeah Francis Chow and Vinicius to design highlighted this massive move by the Triton group.

00:03:27: They are actually integrating red hats Linux based in vehicle operating system into What they're calling their one OS?

00:03:36: And there's scaling it across all their global truck

00:03:38: brands.

00:03:39: so we're talking Scania international.

00:03:41: yep

00:03:41: all of them right.

00:03:42: they are adopting a fully open-source container based approach across an entire commercial ecosystem.

00:03:48: and that is exactly what Shabh is advocating for, because you do not differentiate a commercial truck by how its base operating system schedules background tasks obviously

00:03:58: not.

00:03:58: You differentiate it by the application layer...the predictive maintenance..the route optimization....that's

00:04:05: Wait, hold on though.

00:04:05: Let me challenge the practicality of this for a second if Trayton puts open source Linux into a Scania truck?

00:04:13: Sure they hold the ultimate legal liability for that vehicle but practically speaking how do they validate millions?

00:04:23: Yeah, it's a huge question.

00:04:25: Like you can't just cross your fingers and hope a community patch doesn't accidentally interfere with the braking system?

00:04:30: Bro!

00:04:31: You definitely can And that is basically the core friction point of this entire transition.

00:04:36: But it really comes down to how The modern software stack has layered.

00:04:40: Henrik Manguino made A really sharp observation about This

00:04:43: dynamic.

00:04:43: Okay what was his take?

00:04:45: Well he noted That of the software defined stack, things like the OS abstraction layer and inter process communication.

00:04:52: It's shifting away from in-house proprietary builds.

00:04:56: okay it's moving towards standardized vendor solutions From like Nvidia or Qualcomm Or open source models Like Eclipse S core.

00:05:03: Wait so what is eclipse s core doing In that specific context?

00:05:07: Its essentially an Open Standard for The Plumbing.

00:05:09: It defines the grammar And rules So That Different Software Blocks Can Speak To Each Other Safely.

00:05:15: Oh I see.

00:05:16: So, Manguino's takeaway is that the automakers are no longer selling that middleware.

00:05:21: What they're actually selling and what their validating is integration.

00:05:24: So, integration is product?

00:05:26: Exactly!

00:05:27: integration and safety argumentation at scale are now the actual products.

00:05:31: You don't validate every single line of the open source clay.

00:05:34: in isolation, you validate how the entire integrated system behaves within like strict fenced off safety parameters...

00:05:42: And when an automaker actually owns that integration layer it completely changes what the vehicle is capable

00:05:48: of.

00:05:49: Oh, totally!

00:05:49: Like Whitey Bluestine shared his experience recently with the Rivian OS-II update...

00:05:53: Right?

00:05:54: The Gen II architecture.

00:05:55: Yeah.

00:05:55: Rivian consolidated their Gen II R-I architecture and they brought the ECUs down from seventeen to just seven.

00:06:03: That's a massive hardware reduction

00:06:05: Massive.

00:06:06: And because that an over-the-air update isn't just tweaking user interface on screen anymore.

00:06:12: Blue Steam noted that this specific update changed deep hardware behaviors like the twelve volt battery management and low voltage fault recovery

00:06:21: Which is I mean, that's incredibly difficult for legacy automakers to replicate

00:06:25: impossible in many cases.

00:06:27: right

00:06:27: a legacy OEM has like Ten different tier one suppliers providing those modules as proprietary black boxes.

00:06:34: They are legally and technically locked out from altering that code.

00:06:38: Yeah, they can't touch it.

00:06:39: But Rivian can beam down an update That changes how the battery charges because they own that consolidated architecture From top to bottom.

00:06:46: And this is a but we have to acknowledge The immense hidden costs of getting To that elegant consolidated state.

00:06:53: Oh

00:06:53: for sure!

00:06:53: It's not cheap.

00:06:54: No Birkin Atlem has brought up A really harsh reality regarding these zonal architectures.

00:06:59: When you consolidate the physical nodes left in the extreme ends of the car or the zonal controllers, they stop being decision makers.

00:07:06: Okay so what do they become?

00:07:08: They essentially just become IO aggregators

00:07:10: meaning they just collect input and output data like instead of a smart module making a decision at the wheel.

00:07:16: it's just a dumb messenger passing sensor data up to the central brain.

00:07:21: precisely yeah which strips out miles of heavy wiring Which is great for manufacturing.

00:07:26: right.

00:07:27: but Atlamas points out that this massively spikes your verification costs because now a single zonal controller is handling the functional safety data of three or four previously independent systems.

00:07:39: Oh, wow!

00:07:40: I didn't think about that.

00:07:41: Yeah

00:07:42: When you perform a fault tree analysis to ensure safety You can't just isolate one physical wire anymore.

00:07:48: The complexity didn't disappear It just migrated from the physical wiring harness into the software verification process.

00:07:55: So it's a software headache no?

00:07:57: Exactly, and small tier one suppliers are finding that verification budget absolutely terrifying.

00:08:02: Which honestly explains why some automakers were kind of dragging their feet on this transition like Samay Yassen highlighted a really fascinating critique Of the new Audi A-II e-tron.

00:08:11: Oh yeah The New Audi

00:08:12: Right.

00:08:13: so as physical product the buzz is great right it's efficient It has a spacious interior But underneath the sheet metal, it is heavily relying on legacy carrier software and a four hundred volt MB platform rather than a true zonal next generation setup.

00:08:29: Yeah It's deliberate calculated trade-off.

00:08:31: right.

00:08:32: they optimize The hardware using a legacy platforms so that can physically launch the car in compute for market share today.

00:08:38: but structurally They are instantly accumulating a mountain of long-term technical debt?

00:08:44: Absolutely.

00:08:45: They are shipping a brand new car on a technological foundation that is essentially already aging out

00:08:50: and That technical debt Is going to be a massive anger.

00:08:52: because automakers or taking all this architectural pain and verification nightmare for A very specific reason which is you cannot run an AI agent inside a car, but has a hundred scattered brains that refuse To communicate

00:09:05: Right.

00:09:05: The architectural consolidation we just discussed.

00:09:07: that is the mandatory prerequisite for what comes next, this shift from a software-defined vehicle to an AI defined vehicle?

00:09:14: So let's get into the AI layer then because the narrative is moving incredibly fast here.

00:09:19: like Bill Russo observed the Chinese automotive market recently and noted competition has already blown past.

00:09:26: concept of

00:09:28: smart cockpit is old news.

00:09:30: Right, the new battleground as the agentech

00:09:33: cabin.

00:09:33: Agentech meaning that vehicle actually has agency doesn't just passively wait for you to issue a voice command to change the radio station right

00:09:42: proactive exactly.

00:09:43: it understands your context It anticipates your physical needs and it coordinates complex actions across your entire journey.

00:09:52: Which obviously forces the operating system to break out of the gash bore, like Siaga Rodriguez made a point that for years Vehicle OS just meant the infotainment screen

00:10:01: Glorified iPad.

00:10:02: Yeah, it was essentially a tablet glued to the center console.

00:10:06: But now the OS is bleeding into the vehicle dynamics The body controls and the automated driving systems.

00:10:12: It basically manages the entire physical space.

00:10:15: but you know let me play Dell's advocate for second here before.

00:10:17: we just spent years in this industry agonizing over-the-shift To software defined vehicles.

00:10:24: Is AI?

00:10:25: Just the marketing departments, shiny new buzzword to replace SDV.

00:10:29: That's a fair question!

00:10:30: Like are we just pivoting?

00:10:31: because AI is what investors want to hear right now?

00:10:35: I mean it is a cynical way to look at it but Peter Serrino framed their relationship perfectly.

00:10:39: i think he argued that AI does not replace the software-defined vehicle.

00:10:44: It rewards it.

00:10:45: Okay unpacked for me.

00:10:47: how does reward?

00:10:48: Well, if you built your stack properly like... If you decoupled the hardware from the software and centralized your compute.

00:10:53: Like we were just talking about You are now structurally positioned to run heavy AI workloads

00:10:59: Because data pipelines already there

00:11:01: Exactly!

00:11:02: The data pipelines are already flowing into a central processor.

00:11:05: So the SDV was simply foundation Right?

00:11:07: AI is house on top of

00:11:09: it.

00:11:09: That makes perfect sense.

00:11:11: And when those systems can finally talk with each other presents a cover of holocausts in car.

00:11:16: AI orchestration.

00:11:18: Oh, I love this example!

00:11:19: It's brilliant scenarios as imagine telling your car...I'm tired.

00:11:24: find the place to stop without adding fifteen minutes for my trip For a legacy vehicle that is an impossible command

00:11:31: Because data is totally siloed.

00:11:33: Exactly To execute single request The conversational AI has to part your intent.

00:11:41: Then It has to ping the battery management system for your state of charge, it has query the navigation system for topological route data.

00:11:49: And that has check driver monitoring camera to verify fatigue levels?

00:11:53: Yes!

00:11:54: All those discrete systems have to exchange data seamlessly in milliseconds.

00:11:58: That is real-time orchestration

00:12:00: And automakers are adopting very deliberate strategies to actually build those platforms.

00:12:04: right now, like Ekrem Mookbel highlighted Hyundai's dual-track approach.

00:12:08: Which is really smart!

00:12:09: It

00:12:09: IS on one hand they're integrating NVIDIA's validated vehicle AI platform to accelerate their immediate production timelines but simultaneously They're developing their own proprietary atria ai capabilities in house

00:12:22: So there basically hedging their bets to stay competitive Right Now while targeting level two plus autonomous capabilities by twenty twenty

00:12:28: eight Exactly, and the foundation models powering these capabilities are getting wild.

00:12:34: Like Dave Taukic brought up NVIDIA's Alpamao Models.

00:12:38: VLA Models?

00:12:39: Yes which?

00:12:40: our Open Vision Language Action Models or VLAs.

00:12:43: And that is a critical evolution because to understand a VLA model you just have think about it.

00:12:47: human driver.

00:12:48: Okay If a ball suddenly bounces into the street in front of your car, you don't wait for the language center to form this sentence.

00:12:55: There is a ball—a child might be following it!

00:12:57: I should press the brake

00:12:58: pedal.".

00:12:59: No...of course not…you just hit the brakes.

00:13:01: Right.

00:13:01: Your visual cortex maps directly to physical action —reflexes.

00:13:05: You react before even having words?

00:13:07: Exactly.

00:13:08: VLA models do exactly that for autonomous systems.

00:13:11: They map visual inputs from cameras directly driving actions.

00:13:16: So they fundamentally reason about the physical world in front of them, rather than having to translate everything into rigid pre-programmed code logic first.

00:13:24: That is a phenomenal way to explain it.

00:13:26: VLA models give the car reflexes.

00:13:29: and hey by the way if you're finding this breakdown of foundational AI models an architecture useful take a quick second to subscribe to The Deep Dive on whatever app your using right now.

00:13:39: It really is the easiest way.

00:13:40: make sure that all our future break downs where technology heading.

00:13:45: Because you know having an AI with reflexes is incredible in a simulator.

00:13:50: But what happens when that VLA model actually hits the absolute chaos of?

00:13:55: A real city street

00:13:56: right.

00:13:57: That is where the rubber meets The road literally.

00:13:59: so let's look at this sheer scale Of the deployments happening as we speak.

00:14:04: Jevgeny Kavanaugh announced that bolt and lucid are partnering to deploy twenty five thousand fully autonomous Robo taxes across Europe.

00:14:11: twenty-five thousand I mean, that is not a pilot program.

00:14:14: No!

00:14:15: That's as massive commercial fleet

00:14:17: It Is and the specific choice of Lucid is really telling.

00:14:20: Kabanoff pointed out that a robotaxial is completely different life than consumer vehicle.

00:14:24: The Fleet Operator carries the economics every single design decision.

00:14:28: Right because if you run a fleet your profit margin lives in dies by API access to the vehicle.

00:14:34: Absolutely You need know exactly how much power sensor cleaning system draws Or exactly how fast the automated doors open and close, because those split seconds dictate your utilization rates.

00:14:45: Which means Bolt needed a hardware partner that gave them access to lowest levels of software-defined architecture.

00:14:52: And it isn't just an European phenomenon either.

00:14:55: Shane Boguey noted that Nevada has now authorized up to seven thousand fully autonomous vehicles for paid passenger transportation.

00:15:03: So their regulatory gates are definitely opening.

00:15:05: Yeah But you know...that brings up a glaring question.

00:15:08: If the software architecture is consolidated and that AI has reflexes, And the fleet deals are actually signed.

00:15:15: What does the actual hold up?

00:15:16: Like why aren't they everywhere?

00:15:17: yeah Why aren't our streets flooded with these today?

00:15:20: well The bottleneck Is no longer the intelligence according to a case study shared by Joe de Becker regarding a code is the biggest barrier To scaling.

00:15:28: is validation

00:15:29: Validation?

00:15:30: Yeah

00:15:30: you can simulate millions of miles in A server farm sure but the real world Has unpredictable construction zones, erratic pedestrians and bizarre weather patterns.

00:15:41: Validating the AI's response to that sheer unpredictability And creating a scalable safety foundation based on that validation.

00:15:49: That is the hardest engineering problem in the world right now

00:15:51: which ties directly into A super provocative point Andrea Leitner made about the Chinese automotive sector.

00:15:58: Oh,

00:15:58: this was really interesting

00:16:00: because we all know they are moving at breakneck speed and there is a very common assumption in The West that Chinese OEMs are moving fast simply Because They're cutting corners And lowering their safety standards

00:16:12: which lightener argues completely false

00:16:14: complete false.

00:16:15: She says Their Speed isn't about Lowering the bar it's About changing how?

00:16:18: They clear the Bar.

00:16:19: Hmm Because AI Defined Vehicles Require Continuous Learning from Data.

00:16:24: Chinese Oems Are Utilizing High automated validation loops, so instead of relying on slow manual traditional engineering gates they are continuously generating evidence and automating the validation process between trusted safety parameters.

00:16:39: Wow!

00:16:39: They're essentially industrializing the active validation itself.

00:16:43: but

00:16:43: even if you perfect a validation we have to look outside the vehicle too.

00:16:47: exactly because I want to challenge the premise that autonomy is purely a vehicle engineering problem.

00:16:53: right

00:16:53: let's say And the VLA model works flawlessly in a vacuum.

00:16:59: Does that mean it actually succeeds?

00:17:01: In a bustling city,

00:17:02: no It doesn't because true autonomy is an ecosystem problem.

00:17:06: right.

00:17:07: Jeff hood shared some critical takeaways from the accessibility piloted Detroit And he emphasized that safe autonomy requires resilient physical infrastructure.

00:17:16: Like the vehicle still needs clear lane markings?

00:17:19: Exactly, it need standardized road signs.

00:17:21: It needs traffic lights.

00:17:22: They can actually broadcast their state to vehicles' receivers.

00:17:26: So its not just a car being smart.

00:17:27: The city has be smart too.

00:17:29: Right and beyond concrete and paint It require seamless integration with municipal transit agencies.

00:17:35: A robot taxi fleet should complement existing public transport and not cannibalize it.

00:17:40: And above all, It requires human-centered public trust.

00:17:43: You do NOT build trust through an over the air update.

00:17:45: Definitely not!

00:17:46: you

00:17:46: earn it Through first hand experience Deploying safety operators and utilizing local transit ambassadors.

00:17:53: The smartest car in world is still just one node In a much larger municipal network.

00:17:58: And that ecosystem reality is fundamentally changing the commercial sector as well.

00:18:03: Kylis Nishit attended the IAA Transportation twenty-twenty six event and walked away with a pretty jarring conclusion.

00:18:10: Oh, yeah

00:18:11: Yeah He said The battle among heavy truck OEMs Is no longer about who builds the best physical truck.

00:18:17: Which Wild statement for a commercial trucking show,

00:18:19: right?

00:18:20: But you look at Mercedes Volvo Scania.

00:18:22: They aren't just pitching horsepower and aerodynamics anymore they are pitching integrated transportation operating systems.

00:18:28: Right.

00:18:28: so energy management charging infrastructure AI

00:18:31: driven fleet operations predictive uptime.

00:18:34: Nishit basically said the real competition is no longer truck versus truck.

00:18:38: It is ecosystem vs.

00:18:39: Ecosystem

00:18:40: which perfectly synthesizes The entire arc we've been tracking today.

00:18:43: honestly

00:18:43: it really does

00:18:45: to bring this all together for you The painful, expensive shift from scattered modules to centralized zonal architectures.

00:18:54: It is the mandatory entry ticket for the future.

00:18:57: You take on that technical debt and verification cost because you

00:19:00: have do Precisely Because once you lay that consolidated digital foundation AI stops being a novelty.

00:19:06: It becomes the central nervous system, granting the vehicle the reflexes and orchestration it needs.

00:19:12: And that integrated AI-driven architecture is the only way you can eventually scale and validate autonomous fleets in a messy unpredictable

00:19:20: world.".

00:19:30: If the physical hardware is rapidly becoming a commodity, and if software integration AI validation in ecosystem management are the only things creating durable value.

00:19:38: Right

00:19:39: How long until legacy automakers stop calling themselves manufacturers entirely?

00:19:44: And simply operate as giant software integration firms?

00:19:47: I mean it is the existential question keeping every legacy OEM executive awake at night.

00:19:52: If you enjoyed this episode, new episodes drop every two weeks.

00:19:56: Also check out our other editions on charging and battery tech.

00:19:59: commercial fleets an autonomous in mobility markets.

00:20:02: Thanks for joining us on this deep dive into the architecture of tomorrow.

00:20:06: see next time.

00:20:07: don't forget to subscribe.