Best of LinkedIn: Software-Defined & AI-Defined Vehicles CW 31/ 32

00:00:00: Provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts about software-defined and AI defined vehicles in calendar weeks thirty one and thirty two.

00:00:10: Frenna 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: Welcome to this deep dive.

00:00:26: If you are navigating the next-gen mobility space right now, You already know... The volume of noise out there is just deafening.

00:00:33: Oh yeah!

00:00:33: Totally

00:00:34: So.

00:00:34: our agenda today is strictly signal.

00:00:37: We are breaking down the absolute top trends in software-defined and AI defined vehicles across LinkedIn, basically cutting through all that utopian marketing fluff to look at well...the brutal execution realities.

00:00:48: Right!

00:00:48: The stuff that's actually happening on the ground?

00:00:50: Exactly.

00:00:51: Shifting underlying architectures…and of course massive regulatory deadlines.

00:00:55: just staring

00:00:56: Yeah, because I mean for a century the industry basically treated cars like mechanical watches right?

00:01:01: You turn the ignition fuel combusts gears mesh.

00:01:05: it was this very closed predictable loop of cause and effect.

00:01:08: Very

00:01:08: mechanical...very predictable!

00:01:10: Right

00:01:11: but the transition we're observing in the data right now is the equivalent of demanding that that mechanical watch suddenly become uh well ...a living thinking organism.

00:01:20: It's

00:01:20: a huge leap

00:01:21: It is, and just as the industry has finally like wrapping its head around the baseline concept of the software-defined vehicle or SDV.

00:01:29: The goalposts have completely moved... ...the new paradigm dominating the conversation is the AI defined vehicle

00:01:35: which honestly represents a fundamental dismantling of the legacy vehicle architecture.

00:01:40: And Jin Chang actually laid out this new framework recently at the AutoSoft event detailing how the industry is moving, you know way beyond just slapping a computer in a car to run the brakes and infotainment screen.

00:01:51: Right it's much deeper than that now.

00:01:53: Yeah we're seeing this layered approach of agentic AI physical AI and sovereign AI being fused directly onto that SDV foundation.

00:02:03: Okay let me just make sure I'm wrapping my head around How those three Actually interact In the stack.

00:02:07: so agentic ai that handles the independent decision-making, right?

00:02:11: Yes.

00:02:11: Exactly!

00:02:12: And then physical AI grounds those decisions in actual spatial awareness and physics like gravity momentum friction all that...

00:02:19: Right.

00:02:20: ...and then sovereign AI ensures the compute happens locally securely write on the vehicle's own silicon rather than relying on some vulnerable cloud connection.

00:02:30: is it?

00:02:30: did I get that right?

00:02:31: That Is The Exact Triad.

00:02:32: Yeah The implication extends far beyond just the automotive sector.

00:02:38: This architecture is effectively a generalized, embodied intelligence

00:02:42: platform.".

00:02:42: Wow!

00:02:43: "...the whole goal is this converged ecosystem spanning vehicles, humanoid robots and eVTELs those electric vertical takeoff-and-landing aircraft.

00:02:53: they're all drawing from the exact same architectural blueprint...".

00:02:56: That perfectly frames an observation I saw from Stefan Lagressel that completely shifts how you view an automaker... He points out that the car is essentially the blueprint for tomorrow's robots.

00:03:06: Oh,

00:03:06: That's a great way to put it

00:03:07: right Because a modern passenger vehicle is already the most complex, highly regulated robot produced at a massive global scale.

00:03:16: So the logical conclusion is that whichever semiconductor companies manage to win the automotive space well they'll inevitably conquer the broader robotics category too.

00:03:25: Yeah I mean think about it.

00:03:26: The compute required in Navigator Toucan Vehicle at eighty miles per hour through blinding rainstorm Is just staggering.

00:03:33: Terrifying really

00:03:34: Exactly.

00:03:35: And if your silicon can solve for level of chaos and latency, dropping that exact same chip into a humanoid robot operating in nice climate controlled factory warehouse is frankly a trivial downgrading complexity.

00:03:49: Which explains all the tectonic activity on this silicon market right now.

00:03:52: like NXP has reportedly targeted the edge AI chip maker Amborella specifically to capture that you know millisecond localized decision-making capability.

00:04:00: Yep!

00:04:01: The land grab is ON

00:04:02: It really is.

00:04:02: And meanwhile, BMW has tapped Qualcomm's Snapdragon Digital Chassis and Ride platforms as their lead compute silicon.

00:04:10: We saw Cristiano R. Ammon & Nicole de Gaulle highlighting this as the foundational shift for next decade.

00:04:16: it a total land grab of the brain.

00:04:22: We should probably hit pause here and push back on this a little bit.

00:04:24: Okay, let's hear it

00:04:25: because the idea of a frictionless unified silicon brain powering our cars and our robots sounds incredible in a keynote presentation but Augustine Fiedel highlighted a recent interview with Schaeffler leadership And they described this current period as a highly stressful double pivot.

00:04:43: A double pivot?

00:04:44: Yeah

00:04:44: I mean, aren't we getting massively ahead of ourselves here?

00:04:47: Legacy companies are frantically trying to build these advanced AI-defined organizational capabilities while their basic SDV projects are still in the very early incredibly painful stages of development.

00:04:58: Oh wow!

00:04:59: Yeah so that utopian vision absolutely shatters when it hits the factory floor.

00:05:04: The dream of generalized AI is colliding with the brutal execution realities.

00:05:12: hardware first.

00:05:13: Building a mechanical car is an art form these companies mastered over a century, right?

00:05:18: They know how to bend metal.

00:05:19: Yeah but building a scalable secure software platform that just happens to have wheels is proving excruciating for them.

00:05:26: they're struggling to ship on time let alone actually turn a profit doing it.

00:05:30: so

00:05:31: where specifically Is the mechanism breaking down?

00:05:34: I mean is at the coding itself or as of the architecture.

00:05:37: While it's the architecture attempting to do way too much, Fadi Labib gave a very sharp assessment of the autosar standards.

00:05:43: that illustrates this perfectly.

00:05:45: So Autosar Classic—the older standard still ships reliably today.

00:05:48: Okay why is that?

00:05:50: Because it uses a signal-based architecture to solve a bounded, highly specific problem.

00:05:54: It does one task...predictably.

00:05:57: But Ottersar Adaptive which relies on a service oriented architecture designed to fulfill this like unbounded generality of the SDV dream that is heavily stalling

00:06:07: because its trying anticipate every possible future workload.

00:06:11: Exactly!

00:06:11: Its sounds like building massive sprawling foundation for a skyscraper when you aren't even sure if your building an office building or shopping mall yet.

00:06:19: Oh, man.

00:06:20: Yeah the middleware gets so heavy and bloated with abstractions that the product teams can't actually compile and ship a working feature today.

00:06:27: Wow yeah That unbounded generality requires constant refactoring.

00:06:31: I mean you are basically building Middleware for sensors that haven't even been invented yet Exactly.

00:06:36: And financial toll of chasing that dream is printed clearly on their balance sheets.

00:06:41: German OEM revenues have seen massive nominal decline since the first half of twenty sixteen.

00:06:46: BMW is down fifty nine percent.

00:06:48: Fifty-nine

00:06:49: percent?

00:06:49: Yeah,

00:06:50: and look at Volkswagen software Unicariad led by Peter Bosch who's pushing this triple AI approach.

00:06:56: They just reported an eight hundred fifty five million euro operating loss of the first half of the year against Just eight hundred fifteen million in revenue.

00:07:04: wait

00:07:05: Operating at a loss just to build the platform itself.

00:07:07: Yes, and that stems directly from Conway's law right?

00:07:10: Organizations design systems that mirror their own internal communication structure.

00:07:15: So when a legacy automaker with these heavily siloed departments for the chassis The powertrain the cabin When they try to build a unified software stack They just end up with a tangled expensive mess trying to bridge there own internal silos.

00:07:28: Yeah, the friction of legacy manufacturing culture clashing with agile software development is literally costing them a billion euros every six months.

00:07:39: Hardware requires frozen specs years in advance, but software requires constant fluid iteration

00:07:45: which naturally forces the strategic capitulation Part that Goswami noted that VW is essentially throwing in the towel on the pure in-house software build.

00:07:53: The capital burn is just too high.

00:07:55: So what are they doing instead?

00:07:57: They're pivoting to joint ventures launching Karazhan with Horizon Robotics in China, and partnering with Rivian in the West.

00:08:05: Basically they're buying the speed that couldn't build.

00:08:07: Right!

00:08:07: And Speed is ultimately a governance metric not a coding metric?

00:08:11: Yes exactly.

00:08:13: Mohamed Frakruz shared at data point on this that it's just staggering.

00:08:16: A single feature request from China waited eight full months to get an answer for German headquarters.

00:08:21: Eight Months Just For An Answer.

00:08:23: Eight Month of Administrative Ping Pong.

00:08:26: Meanwhile, their Chinese competitors are developing entire vehicles from a blank sheet of paper to full production in twenty-four to thirty six months.

00:08:36: Wow!

00:08:36: You cannot survive if your internal decision loop takes the third time it takes you rival engineer an entire car

00:08:43: Yeah...you're dead on water.

00:08:45: Hey, real quick by the way to you listening.

00:08:47: if want stay ahead of how these massive global shifts and execution gaps are playing out make sure hit subscribe so catch our future deep dives.

00:08:54: We're tracking this transition very closely.

00:08:56: Definitely!

00:08:57: And there's one more execution hurdle here that industry is severely underestimating.

00:09:03: Métay Alves-Bazera calls it The Extend Right Problem.

00:09:07: Okay, The Extended Right Problem what?

00:09:09: Well engineering teams are hyper focused on shift left right Testing software and simulating environments way earlier in the development cycle.

00:09:17: Sure, yeah But they're largely ignoring The right side of the timeline Managing the compute life-cycle Of a vehicle that has to remain updatable for fifteen To twenty years.

00:09:25: Oh wow!

00:09:26: Think about the physical reality For a second.

00:09:29: We are talking About automotive grade Silicon baking In extreme temperatures Like A car parked in Arizona sun Then driven in freezing winter.

00:09:37: Exactly You have NAND flash memory that physically degrades after thousands of read and write cycles.

00:09:43: Now imagine trying to push a heavy generative AI operating system update to a fifteen-year old degraded ship in the year twenty forty.

00:09:52: Yeah, it won't work!

00:09:53: The hardware will completely bottleneck.

00:09:54: this software...the continuous integration pipeline kind of assumes an infinite capacity to update but the physical vehicle is a depreciating static compute environment

00:10:04: which makes the safety implications absolutely terrifying.

00:10:07: if a fifteen years old processor is going struggle just run as simple infotainment updates without thermal throttling.

00:10:13: How on earth do we validate and trust autonomous driving AI systems operating out in the wild, On that same aging hardware?

00:10:20: It is the ultimate high stakes computing environment.

00:10:24: And it's driving a massive philosophical divide In the autonomous vehicle space right now.

00:10:30: End to end AI versus hybrid architecture.

00:10:34: Gerald F. Lackey shared a McKinsey survey showing That only twenty two percent of experts believe end-to-end AI will ultimately dominate.

00:10:43: Just twenty

00:10:43: two percent, I mean i understand the skepticism there.

00:10:46: end-to-end AI where raw camera data goes in one side of the neural network and steering commands come out The other is fundamentally a black box.

00:10:53: you cannot mathematically prove why it made a specific choice In

00:10:57: high stakes environments absolutely demand determinism.

00:11:00: This is exactly why the hybrid approach was favored.

00:11:02: You use AI for perception and path prediction, sure!

00:11:05: But it's supervised by a deterministic rules-based validation layer... Okay

00:11:08: so that has boundaries?

00:11:09: Yes

00:11:10: this layer has hard coded physics constraints.

00:11:13: It only jobs to just say no and override the A.I..

00:11:16: It's digital equivalent of driving instructor with passenger side brake

00:11:19: pedal.

00:11:20: That's great analogy.

00:11:22: The AI

00:11:22: might suggest veering slightly into the oncoming lane to avoid a pothole, but the deterministic layer checks the radar, sees an oncoming truck and physically prevents the steering rack from executing that command.

00:11:34: It imposes hard limits.

00:11:35: Exactly!

00:11:36: But...the AI camp is attempting bridge this deterministic trust gap.

00:11:42: Joe Oakwist & Shamir Anjum highlighted NVIDIA's newly open-sourced Alpamao II Supermodel.

00:11:48: Okay what does it do?

00:11:49: It's a thirty-four billion parameter reasoning model, but its real breakthrough is generating these chain of causation traces.

00:11:58: It doesn't just issue a breaking command.

00:11:59: it links its decision to specific vectors in the visual scene!

00:12:08: In

00:12:09: text-based, large language models AI hallucinations are a massive totally unsolved problem.

00:12:14: The model's just confidently invent facts.

00:12:17: so if this NVIDIA model is generating language to explain its driving behavior how do we know the AI isn't just hallucinating?

00:12:23: A highly convincing post rationalized excuse for why it Just caused an accident.

00:12:27: that

00:12:27: has very fair point but It's a critical distinction To make here...the chain of causation Isn't just a text generator making up a story to ground its language output in the actual geometric and semantic sensor data.

00:12:43: Okay, so it's tied to the physical

00:12:44: world?

00:12:45: Right It highlights the exact bounding box of pedestrian that triggered breaking threshold.

00:12:51: It is attempting to make blackbox transparent but basically showing its mathematical work

00:12:56: I see, But even if math is completely accurate Christian Eckert raises a fascinating user experience dilemma here.

00:13:03: Oh

00:13:03: yeah!

00:13:03: The UX side of this is wild.

00:13:05: Right If the car can essentially explain it's internal reasoning in real time Should passenger actually see on screen?

00:13:12: If i'm sitting back seat Does a prompt saying, I swerved left because the probability of the cyclist falling was eighty-two percent?

00:13:19: does that actually build my trust or does it absolutely terrify me by revealing exactly how much independent life for death judgment this machine is exercising.

00:13:27: Yeah

00:13:27: do you really want to know?

00:13:36: We implicitly trust human drivers, even though they are distracted and flawed mostly because we don't see their internal probability calculations.

00:13:45: Seeing a machine calculate the odds of fatality in real time might induce constant panic.

00:13:50: It definitely would for me, but while we're debating the psychological UX of a robo-taxi Jason Corso brings up much more sobering reality about scale.

00:14:00: Building a brilliant bespoke autonomous system and testing it on four thousand highly maintained robo taxes in a geofence city is an engineering marvel.

00:14:08: sure absolutely But it completely ignores the two hundred seven million legacy human driven vehicles already On The Road today that are actually causing real world accidents.

00:14:16: Yeah!

00:14:17: The macro safety crisis caused by the Legacy fleet.

00:14:19: If autonomous safety software cannot scale down and integrate into the broader chaotic ecosystem of older vehicles, well it isn't solving the global fatality rate.

00:14:28: It is just creating a niche hyper-expensive luxury service for specific zip

00:14:32: codes.".

00:14:33: And when you attempt to scale that software across millions connected vehicles globally... You immediately trigger the regulatory immune system!

00:14:42: Scaling connectivity isn't just an engineering problem.

00:14:45: It makes the vehicle a target for cyber threats, which brings the massive hammer of compliance down on the automakers

00:14:52: Speaking of which Dr.

00:14:54: Jeanine Johnson points out a massive bottleneck arriving very soon.

00:14:58: The EU Cyber Resilience Act takes effect September

00:15:02: eleven.

00:15:02: Oh that's right around corner.

00:15:03: it demands a twenty four hour early warning system for vulnerabilities and A fourteen day final report once a fix is available.

00:15:11: A fourteen-day timeline to deploy a fix to a fleet of millions of cars.

00:15:15: That is practically science fiction for legacy automakers, you aren't just pushing the patch into your smartphone app here and over.

00:15:21: the air update requires rigorous hardware in loop testing.

00:15:24: If you patch vulnerability on the infotainment module You have prove it doesn't cause memory leak that crashes the braking system.

00:15:31: That homologation and firmware signing process takes months.

00:15:35: Compressing that to fourteen days turns basic software maintenance into constant business continuity crisis.

00:15:40: the compliance overhead alone is going to break smaller suppliers.

00:15:44: Right, and shifting from the EU to the US The regulatory focus moves from resilience To outright supply chain decoupling.

00:15:52: right.

00:15:52: Jennifer E tisdale shared the developments around the us connected vehicle security act.

00:15:58: And just be super clear to anyone listening right now We are strictly imparting the structural and supply-chain implications of this legislation as found in source material without endorsing any political viewpoint whatsoever.

00:16:10: just looking at the facts.

00:16:11: Exactly, this act which unanimously passed The Senate Commerce Committee phases in strict software restrictions starting in twenty-twenty seven and hardware restrictions for connected vehicle technology linked to China Russia Iran or North Korea.

00:16:28: And the mechanics of actually executing that mandate are just staggering.

00:16:32: Untangling a global hardware supply chain in three to six years means finding and replacing deep-tier microcontrollers, sensors, and communication modules that have been baked into vehicle architectures for a decade.

00:16:43: It's

00:16:43: massive!

00:16:44: it formally designates The Connected Car as a vital national security domain...

00:16:48: ...and isn't just federal regulators throwing up roadblocks either—the friction is intensely local.

00:16:54: Matthew Raifman highlighted a massive gap in California's AB-Seventeen Seventy Seven.

00:17:00: The law successfully created an emergency geofence option, allowing police to restrict autonomous vehicles during active

00:17:07: crises.".

00:17:08: Did

00:17:08: you make sense?

00:17:09: It does but cities still completely lack the legal authority to restrict AVs during planned large events.

00:17:15: Oh

00:17:16: right and the practical result of that loophole was fully exposed during the Fourth of July gridlock in San Francisco.

00:17:21: Oh the mess.

00:17:22: Waymo Vehicles operating strictly by their programming, struggled to navigate the unpredictable congestion and erratic crowd behavior.

00:17:31: And the city couldn't legally mandate them out of the dense areas ahead time.

00:17:35: The local infrastructure in technology were completely at a sink

00:17:38: Which brings up brilliant analogy from Todd Smith about how industries protect themselves.

00:17:43: He describes the reaction to autonomous vehicles as an industry's immune system.

00:17:48: When a foreign body enters a biological organism, the immune system attacks it under the guise of protection.

00:17:54: Yeah Todd points out that trial lawyers are actively lobbying against autonomous vehicles citing deep concerns over safety and accountability.

00:18:02: I mean The argument being if the software is the driver who is legally liable in a crash, right?

00:18:08: Is it the OEM?

00:18:08: The software developer.

00:18:09: The hardware provider...

00:18:10: Right!

00:18:10: It's completely valid legal question.

00:18:13: but Todd argues that lobbying is actually an immune response protecting massive revenue stream.

00:18:18: They are protecting the hundred and eighty billion to two-hundred twenty billion dollar annual auto liability payout industry.

00:18:25: Wow

00:18:25: Two hundred billion.

00:18:26: Yeah,

00:18:27: if AVs drastically reduce collisions they drastically reduced litigation.

00:18:31: fewer crashes mean fewer court cases which means billions and lost V's.

00:18:36: The legacy system is actively fighting to protect its own economic survival cloaked in the language of consumer safety.

00:18:42: That as a stark reminder that deploying disruptive technology Is rarely just a battle of engineering it?

00:18:48: It's a battle against entrenched economic preservation

00:18:51: totally which leaves us with a final provocative thought to mull over based on a post by Roberto Diesel.

00:18:58: We have spent this whole deep dive unpacking architectural shifts, software execution failures the AI trust gap and regulatory immune systems.

00:19:08: but what if all of this friction is happening because we are still looking at the vehicle as the final product?

00:19:13: Rather

00:19:14: than just a node in a broader network?

00:19:16: Right What If The Car Is Just One Component In A Mobility Life Cycle?

00:19:21: Roberto suggests that the ultimate financial unlock for a software-defined vehicle might not actually happen while wheels are moving.

00:19:28: A personal vehicle sits parked and idle, for roughly ninety five percent of its life.

00:19:32: It's just taking up space?

00:19:34: Exactly!

00:19:35: Through autonomous fleet integration you could turn a rapidly depreciating asset into an active revenue source.

00:19:41: Your vehicle can autonomously enter a ride hailing network or deliver packages when asleep.

00:19:46: It completely inverts the fundamental model of vehicle ownership from a sunk cost to an active yield generator.

00:19:52: The mechanical watch is dead, we are attempting engineer living breathing continuously evolving ecosystem and like any rapid evolution it's messy heavily contested requires entirely new survival skills.

00:20:08: if you enjoyed this episode new episodes drop every two weeks.

00:20:11: Also, check out our other editions on charging and battery tech commercial fleets.

00:20:15: And autonomous mobility markets.

00:20:17: Thank you for joining us as we work through the realities of this transition today.

00:20:21: We really appreciate your taking time to learn alongside us.

00:20:24: Keep questioning systems building future of mobility.

00:20:27: Remember to subscribe.