Best of LinkedIn: Next-Gen Vehicle Intelligence CW 19/ 20

00:00:00: Brought to you by Thomas Allgaier and Frenes.

00:00:02: This edition highlights key LinkedIn posts on next-gen vehicle intelligence in weeks nineteen and twenty.

00:00:08: Frennes 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.

00:00:20: You can find more info.

00:00:21: the description

00:00:22: Imagine driving a two tone computer at seventy miles per hour.

00:00:27: Literally the only thing keeping you from a catastrophic crash is just one single line of code.

00:00:33: It's a terrifying thought, honestly

00:00:34: Right.

00:00:35: Well welcome to The Deep Dive and if your joining us You know that mobility space moving incredibly fast.

00:00:41: The stakes really have never been higher.

00:00:43: Today we're looking at what engineers, strategists And developers are actually saying on ground

00:00:48: Exactly!

00:00:48: We're extracting most critical insights From curated LinkedIn content across industry.

00:00:54: In todays journey covers alot We're going to explore how AI is completely redefining the cockpit experience.

00:01:00: Oh, yeah!

00:01:00: The dashboard is totally unrecognizable these days...

00:01:03: It really is.

00:01:04: and then we'll lift the hood To look at the underlying software architecture making that possible

00:01:09: because the hardware just isn't the star of the show anymore

00:01:12: right?

00:01:12: And finally will get into the massive challenge Of validating it.

00:01:15: all you know.

00:01:16: so these vehicles are actually safe to operate.

00:01:20: Okay let's unpack this starting with a cockpit Because to really understand where vehicle intelligence is heading, we have to start right when the human meets machine.

00:01:30: Yeah!

00:01:30: The dashboard is undergoing a massive identity crisis now.

00:01:34: I mean it's no longer just digital speedometer or like a place-to-tap your radio presets.

00:01:40: It has become primary interface for an entirely new computing platform.

00:01:45: We all remember dark days of early voice commands.

00:01:48: You'd be screaming, call mom four times at your steering wheel.

00:01:52: Oh

00:01:52: man!

00:01:53: And the system would somehow end up dialing a local pizza place instead?

00:01:56: Exactly I've definitely been there but looking at recent posts from Inabihi Ufo and Wasim Benz said about the new Rivian assistant rolling out to their R-One vehicles we are light years past that...and they made really crucial distinction here.

00:02:11: this is not just phone mirroring

00:02:12: Right, it's not just Apple CarPlay or Android Auto.

00:02:15: Exactly!

00:02:16: It is not just projecting that onto a piece of glass.

00:02:18: this as native AI agent built directly into the vehicle operating system.

00:02:23: What's fascinating here is the fundamental industry shift we are witnessing from a software-defined vehicle to an AI defined vehicle.

00:02:30: Sri Subramanian posted an excellent breakdown of this.

00:02:33: Oh, I saw that one!

00:02:34: He talked about multimodal systems right?

00:02:35: Yeah,

00:02:36: multimodals and agentic systems.

00:02:38: Multimodal meaning it can process voice visual context and sensor data simultaneously.

00:02:44: And agentic means well...it has actual agency.

00:02:46: It doesn't just wait for rigid command..It can reason

00:02:49: So its not Turn on the AC.

00:02:51: Right, you can tell a system like Rivian's to say move a meeting on your Google Calendar Navigate To The New Location and Text Your ETA All In One Breath.

00:03:01: Wow so it actually understands the context of the cabin.

00:03:03: Exactly!

00:03:04: The AI Understands The Context And Executes Complex Multi-Step Workflows.

00:03:08: You see different automakers taking slightly different paths to get there though.

00:03:12: Take Mercedes Benz for example.

00:03:14: Oh yeah they're going heavily into enterprise space.

00:03:17: Marcus Gobel shared a great interview with Magna Sosberg, the Chief Software Officer at Mercedes.

00:03:23: Their approach is essentially turning the car into what they call The Third Workspace...

00:03:28: Which is an ambitious label!

00:03:30: It IS.

00:03:30: They are natively integrating Microsoft Teams, Chat GPT via Azure Open AI and Intune.

00:03:37: And for anyone who hasn't spent time in Enterprise IT, Intune is basically Microsoft's corporate security tool.

00:03:44: It lets a company manage and secure your data remotely.

00:03:46: Exactly, so Mercedes is building the system where digital transition from office desk right into driver's seat completely seamless and corporately secure.

00:03:56: But I look at this and have to wonder... Is it really the leap forward we think?

00:04:01: What do you mean?

00:04:02: Well are we just bolting an office cubicle in smart phone onto dashboard?

00:04:07: It feels like we might just be distracting the driver with better productivity software instead of fundamentally changing how the car operates as a machine.

00:04:14: I mean, that is definitely a trap some automakers are going to fall into just slapping a tablet on The Dash and calling it a day.

00:04:21: But the leaders are going much deeper than just an infotainment overlay.

00:04:25: Like who?

00:04:26: Well...to see the difference look at their recent Beijing auto show.

00:04:30: Guangyang highlighted a really revealing demo from SenseAuto.

00:04:35: They showed an in-car AI agent tied directly into the vehicle's physical sensor suite.

00:04:40: Wait, tied into the physical sensors.

00:04:42: How does that work?

00:04:43: Yeah so a driver used a simple conversational voice command to search through their dashcam history.

00:04:48: They just asked the car show them the video clip of a near miss at specific intersection.

00:04:53: Oh

00:04:53: wow!

00:04:53: So no scrolling thru sub menus or checking timestamps.

00:04:56: None.

00:04:57: The AI understands the semantic context what a near-miss looks like searches vehicles memory banks and pulls the exact clip.

00:05:05: That is contextual companionship

00:05:07: Because it's an active participant in drive.

00:05:08: Exactly The AI is aware of the car's physical environment, its history and it censor states.

00:05:16: But you know if you want a car that can instantly pull dash cam footage understand traffic patterns and process your calendar seamlessly.

00:05:23: You can't run on a fractured disconnected computer system.

00:05:27: No absolutely not!

00:05:30: which brings us to the actual architecture and how these companies are organized.

00:05:34: Yeah, And Anuridha Dorlyani posted a brilliant application of Conway's law to automotive engineering To explain why this is just so difficult.

00:05:43: Remind me, Conway's law is about organizational structures right?

00:05:46: Spot on!

00:05:47: It states that organizations design systems that mirror their own communication structure.

00:05:52: Legacy vehicles ended up with over a hundred separate isolated electronic control units or ECUs... Because

00:05:59: traditional automakers had completely seperate departments.

00:06:01: Exactly The power train department didn't talk to braking and breaking didn't talked infotainment.

00:06:07: It's like trying to build a world-class symphony orchestra, but locking the string section and the brass section in two different buildings giving them different sheet music.

00:06:16: And hoping it sounds good when they finally play together.

00:06:19: That

00:06:19: is a perfect analogy.

00:06:21: If your company has siloed Your car's brain will be silo'd.

00:06:24: You simply cannot build a unified centralized software defined vehicle with a hardware defined silo'ed organization.

00:06:32: The company structure actually has to change And the architecture itself is going through a massive evolution right down to the fundamental code being written.

00:06:39: Like you brought up what Hans post about The silent shift happening in the industry from the programming language C++ to Rust.

00:06:46: Oh, that's a massive shift In the industry Right now.

00:06:49: and he made point That this isn't just some trendy developer preference.

00:06:53: It is critical safety architecture decision.

00:06:56: C++ has been the industry standard for decades Because it allows developers To build really fast efficient software

00:07:02: Very fast, but the downside is that it leaves the door open for memory leaks.

00:07:07: Exactly!

00:07:07: Memory leaks and buffer overflows only show up at runtime meaning all of this software's actually operating In a laptop.

00:07:14: A memory leak is an annoying crash in a vehicle moving on highway speeds.

00:07:19: A memory error affecting the braking system Is catastrophic safety failure?

00:07:24: Yeah And Rust solves this by eliminating memory-safety risks At compile time.

00:07:29: How does this really clever?

00:07:32: Rust forces strict ownership rules on every piece of data.

00:07:36: So

00:07:36: it won't even let the code run if its flawed?

00:07:38: Exactly!

00:07:39: If the code isn't perfectly memory safe, then compiler literally refuses to turn into an executable program.

00:07:46: The bugs are caught before the software is allowed to exist.

00:07:49: That's

00:07:49: incredible But having a safe bug-free code only has half the battle Right...if

00:07:53: we connect this with the bigger picture It all about how these disparate systems integrate continuously.

00:07:59: Marcus Rettstat brought up a great point regarding Eclipse S-Core and AutoSAR.

00:08:03: And for those who aren't deep in the weeds, AutoSar is basically this standardized rulebook of how automotive software should be written or structured so different parts can talk to each other.

00:08:12: Correct!

00:08:12: Marcus noted that you could have best individual rust based software stacks but if they don't integrate seamlessly via these standards at runtime... You don't have a platform, you just have islands of technology.

00:08:25: And to bridge those islands the industry is moving heavily toward virtualization.

00:08:29: Yeah!

00:08:29: Harshad Bhuttata noted how companies like Qualcomm and Google are virtualizing software-defined vehicles To decouple the software from hardware life cycles.

00:08:39: So they're running consistent automotive software across both physical hardware AND virtual system on ships or socks.

00:08:47: Right, and Unsock is essentially an entire computer system.

00:08:51: The processor the memory the graphics packed onto a single microchip.

00:08:56: So virtualization means they're creating a digital ghost of that physical hardware in the cloud.

00:09:01: Yes

00:09:03: That way developers can test and innovate at the speed of software.

00:09:06: They don't have to wait months for a new physical micro chip To be manufactured and installed in a test vehicle.

00:09:12: You just test the code on the virtual chip And when it works?

00:09:16: But hold on, I have a question about that.

00:09:17: If everything is virtual and we are constantly updating these complex systems over the air doesn't make this car incredibly fragile?

00:09:26: We've all had our smartphones completely bricked by bad iOS or Android update.

00:09:31: Oh for sure.

00:09:32: So how does an automaker scale a constantly updating system across millions of vehicles without triggering mass outage?

00:09:41: Well, standardized platform continuity is really the only way to manage that kind of complexity without it collapsing.

00:09:46: You don't build a bespoke software architecture for every single car model

00:09:51: anymore Right That would be a nightmare to maintain.

00:09:53: Exactly.

00:09:53: Peter Bosch posted about exactly this.

00:09:55: regarding Cariad The Software Subsidiary For Volkswagen Group They are scaling their software platform from high-end premium vehicles all the down to entry level Skoda Epic.

00:10:06: So they use same foundational architecture across entire fleet.

00:10:09: Yes

00:10:10: When you use the same architecture, you aren't managing a million different fragile webs.

00:10:14: You are managing one robust heavily tested platform that just turns certain features on or off depending on the car it's in.

00:10:22: It is the difference between artisanal custom engineering for every single car and building true scalable factory software

00:10:30: Exactly.

00:10:31: But this leads us right into the biggest hurdle of all.

00:10:34: If this virtualized Continuously updating architecture is the vehicle's new nervous system.

00:10:40: How on earth do we validate it?

00:10:42: how Do We prove an AI Is safe enough to let It drive The car itself?

00:10:46: That, is the trillion dollar question and the actual bottleneck holding everything up might surprise you.

00:10:52: Here's where it gets really interesting.

00:10:54: Sandeep Coray shared An incredible insight.

00:10:57: he pointed out that sensor fusion is basically an architecturally solved problem.

00:11:01: now

00:11:01: wait Really the hardware is fully there.

00:11:04: Yeah, the physical act of combining cameras LIDAR and radar into a unified three-D view of the world.

00:11:09: The hardware can do it!

00:11:10: The real bottleneck is AI perception validation.

00:11:13: Oh that makes sense because traditional automotive safety has built on deterministic hardware equations.

00:11:19: you test mechanical brake line thousands of times And you can calculate exactly when will fail due to wear or tear.

00:11:25: But When your sensor actually learned neural network interpreting kixels standard safety equation fails.

00:11:32: An AI doesn't have a mechanical wear rate, it gives a probabilistic output.

00:11:36: It says I am ninety-eight percent sure that is a stop sign and not a pedestrian wearing red shirt

00:11:43: Right.

00:11:43: And the safety tool chains for regulatory approval simply aren't built to handle probability like that

00:11:48: Which creates what George Pascoe refers as the SOTEF validation gap.

00:11:53: And so, Tev stands for safety of the intended functionality.

00:11:56: His

00:11:56: post was fascinating!

00:11:58: He broke down this sheer mathematical impossibility... ...of real-world testing.

00:12:02: The numbers are staggering.

00:12:04: To prove statistically that an autonomous vehicle is safer than a human driver, you would need to drive it for roughly fourteen billion kilometers.

00:12:12: Let's just dwell on the number first.

00:12:13: second...

00:12:14: It's mind-boggling!

00:12:15: If you drove single car twenty four hours of day at highway speeds…it'd take you roughly sixteen thousand years to hit fourteen billion km.

00:12:23: So you literally cannot drive your way to safety?

00:12:26: No – You can't.

00:12:27: by time finished hardware and software would be decades obsolete

00:12:30: …So if real world driving cant solve How do they validate the AI?

00:12:42: They use

00:12:46: procedural generation to simulate thousands of chaotic edge cases.

00:12:54: Pedestrian wearing a stop sign t-shirt is walking a dog on black ice with blinding sun glare hitting the camera lens.

00:13:00: The ultimate nightmare scenario for a camera?

00:13:02: Yes,

00:13:03: and they simulate these ultra specific chaotic moments ten thousand times an hour On massive GPU clusters to find the exact statistical boundary where the car's AI perception fails.

00:13:15: Wow You were essentially using one AI in a server farm to aggressively interrogate another AI and a virtual car.

00:13:22: It is just mind-blowing scale,

00:13:23: it really is.

00:13:24: but you know there's Another side of this safety coin And it's less about the car making an innocent mistake and more About malicious actors.

00:13:30: yeah talk cyber security.

00:13:31: oh This is a massive issue.

00:13:33: definitely I was reading a post by Antonio Gonzalez Pregueño about something called vex files vulnerability exploitability exchange files.

00:13:42: In the modern software supply chain, automakers get flooded with thousands of automated vulnerability alerts every single day.

00:13:49: Yeah it's a constant barrage.

00:13:51: and VEX files basically technical document that says yes this flaw exists in our code but here is exactly why cannot be exploited in our specific vehicles architecture.

00:14:01: But Brigania warned these VEX file can not just be bureaucratic shields.

00:14:06: Right companies cant use them to avoid the cost of patching or testing.

00:14:09: They must deeply technically defensible.

00:14:12: And that ties directly into a sobering point made by Michael Entner Gomez.

00:14:17: he emphasized, in the connected vehicle any over-the-air update channel or gateway module is massive attack surface.

00:14:24: Because once you have software to find architecture You aren't just driving mechanical vehicles anymore

00:14:29: Exactly!

00:14:29: You are driving two ton computer.

00:14:31: someone else can reach remotely.

00:14:34: The exact same infrastructure that seamlessly pushes your new infotainment update Or fixes bug could, under the wrong conditions be compromised by a bad actor.

00:14:43: When you put it like that... It really changes how we should view data these cars generate.

00:14:48: I mean for years.

00:14:49: The narrative was that vehicle data is something to be harvested and sold to advertisers or third parties.

00:14:55: Right!

00:14:55: The new oil?

00:14:56: Exactly Yeah.

00:14:57: But considering the safety in cyber security risks ...it sounds like We need stop thinking of vehicle data purely as gold mine to be sold and start treating it like a continuous EKG monitor for the car's health and security.

00:15:09: The data's primary purpose has to be diagnosing, and protecting...the host

00:15:13: organism.".

00:15:14: I think that analogy is spot-on actually!

00:15:16: And industry data backs you up.

00:15:18: Leslie Clavin highlighted in new reports showing OEMs are officially shifting their priorities…

00:15:23: So they're finally focusing on smart diagnostics & predictive maintenance?

00:15:27: Yes They've overtaken data monetization as top global priority for automakers.

00:15:33: They are realizing the true value of data isn't in selling it.

00:15:36: It's using it to spot anomalies, stop cyber threats reduce costly recalls and ensure that vehicle actually gets better and safer over its life cycle.

00:15:46: So what does this all mean?

00:15:48: We've gone on quite a journey today.

00:15:49: we started inside the cabin looking at how agentic AI is turning the dashboard into a context aware companion,

00:15:55: and then we dove under the hood to see how legacy organizational silos have to be torn down

00:16:00: right?

00:16:00: To build standardized virtualized software platforms using safer code like Rust.

00:16:06: And finally, we looked at this sheer mathematical impossibility of real world validation and the massive simulation in cybersecurity shifts required to keep these two ton computers safe from edge case accidents an active threats.

00:16:18: This raises an important question for you to consider, though.

00:16:21: As vehicles become these continuously updated AI agents monitoring our driving habits or calendars and the environment around us in real time at what point does the car stop being a piece of hardware you own?

00:16:33: And start becoming?

00:16:38: And as that boundary blurs, how will this shift redefine personal privacy and insurance liability in the coming

00:17:02: decade?