The process disappears. The responsibility stays. We tend to think of AI agents as convenient assistants. But an agent is not simply something that carries out instructions — it decides what to show, what to recommend, which services to invoke. The user does less. Someone else decides more. The pattern we kept running into with autonomous driving and ADAS — systems that decide but carry no accountability — has begun appearing inside the cockpit.

By Sang Min Han _  han@autoelectronics.co.kr
한글로보기






On the road to Las Vegas, the experience of driving a Tesla is different from driving another EV. Tesla plans the charging route automatically within its own Supercharger network, pre-conditions the battery before you arrive, and handles payment without a tap. The driver of a comparable EV opens an app, checks an account, and sorts out network compatibility. The gap is not in features. It is in ecosystem integration.
Something similar was on display at Auto China 2026. Xiaomi did not exhibit the SU7 and YU7 as cars. Alongside the vehicles were HyperOS, an AI voice assistant, and smart home integration scenarios. The user was not being asked to learn a new system inside a car — the ecosystem they already lived in had simply extended into one. What the exhibition was really about was not the vehicles themselves but the ecosystem those vehicles would plug into.
Where the person used to move first and the car followed, now the system moves first. That shift is spreading across the industry. And inside it, a much larger question is growing: who, going forward, will own the automotive experience?



Apps Don't Disappear. They Just Become Invisible.

The app era inside the car is not old. Navigation apps, music apps, charging apps, parking apps — users opened a screen whenever they needed something, searched, and launched. The smartphone logic moved into the vehicle. And most of those apps were not the OEM's. The OEM provided the screen. The apps came from outside.
When Apple CarPlay arrived in 2014, followed by Android Auto, OEMs already understood that parts of the experience — navigation, music, messaging — could be determined by the smartphone ecosystem. Most accepted it anyway. Consumers wanted it. That was a judgment made inside the reality of the time.
Now something else is coming into the cockpit. AI agents.
An agent is not an app. An app requires the user to open it. An agent has already made judgments before the user does anything. It reads context — destination, charge level, calendar, weather, previous driving patterns — and calls the services it decides are needed. The user never touches the navigation, charging, or music app. The system fills in.
When agents arrive, apps step back. CarPlay changed what users see. Agents begin deciding what users choose. Competition shifts accordingly — from building better apps to being selected in the moment the agent makes its call.




In the AI era, is the head unit's core challenge to expand functionality — or to design the boundary between the vehicle and its ecosystem?



Those Who Design the Boundary, and Those Who Make It Irrelevant
But who sets the criteria for those choices?

LG Electronics Vehicle Solution put the question directly. The answer, they said, is not expanding functionality — it is designing the boundary. What belongs to the car and must be handled inside it, immediately: climate control. What needs the cloud: music streaming. What has to mesh with the smartphone ecosystem: payments, calendar. Defining where the vehicle's responsibility ends and the external service's begins is the head unit's job. When agentic AI enters the vehicle, the central question is which experiences the car is accountable for. And personalization, they argued, should be built around consent and context rather than continuous learning.
Around the same time, LG was also emphasizing something else: "Through collaboration with Google and major global technology partners, we will continue delivering differentiated experiences."
Designing the boundary while standing on top of a platform. LG demonstrated a solution running on Qualcomm's Snapdragon Cockpit Elite Platform — a single SoC driving multiple displays simultaneously. Navigation for the driver, YouTube for the passenger, different content for each rear seat. All of it on Google's Android Automotive OS.
LG calls itself an orchestrator. But an orchestrator does not write the score.
Within this arrangement, roles divide. In external service areas — maps, POI, charging recommendations — platform players like Google carry significant weight. A structure where multiple LLMs coexist across different layers, rather than a single model, is also taking shape. In-vehicle services and agent decision rules are designed by OEMs. OEMs set the criteria for moments when the agent struggles to judge — when more communication with the user is needed. And whether it is LG, Bosch, or Harman, integrators find their roles within those OEM rules.
The automotive agent is different from the smartphone. It demands embedded and cloud simultaneously. Experience, safety, and policy are all at stake at once. No single player can take everything. Yet in Beijing, the picture looked different. While Tier 1s and OEMs were designing boundaries, something else was making those boundaries beside the point.
Xiaomi exhibited the car as the last node in the HyperOS ecosystem. Not a user learning a new system in a car — an ecosystem the user already lived in, extending to include the car. Huawei went further: powertrain electrification, intelligent driving, cockpit, AI agent, app ecosystem — all in a single booth. A space where no boundary was necessary. When Tier 1s said they would design the boundary, the order above that boundary was already being written somewhere else. The contest was not about who designs the boundary. It was about who becomes the source of authority.
Vehicle safety, brand experience, and data policy are areas OEMs cannot easily give away. In Western markets that demand data sovereignty and privacy, the ability to design that boundary is itself a condition of entry. Efforts to open the vehicle's state and context via API, connecting to external services, are growing. But an open API does not automatically organize the experiential order on top of it. The issue is not connection. It is who, on top of that connection, sets the rules of the experience and decides what gets recommended on what basis.




Xiaomi's booth at Auto China 2026 brought together its AI model (MiMo), operating system (HyperOS), and proprietary SoC (XRING O1) — a single-brand ecosystem that now extends into the vehicle.



The Process Disappears. The Responsibility Stays.

One thing needs to be stated clearly. The automotive agent is different. It must operate as an embedded-cloud hybrid. The car has to keep moving even when connectivity drops. Safety judgments must be made in real time. Experience, safety, and policy are all decision variables for an automotive agent. The weight is different.
Where that weight shows up most sharply is in accountability. A delivery app was recently found to be displaying open restaurants as closed — citing a shortage of riders — and setting delivery zones that did not match actual radius. Users thought they were choosing. The choices had already been designed for them.
What happens when that moves into the cockpit? The AI says: "There's a charging station in 12 minutes. Want to stop?" The user nods. But there is no way to know why that station. Closest? Fastest? Under contract with the OEM? In the process the agent handles on the user's behalf, the user experiences the outcome. The process becomes increasingly invisible.
With autonomous driving, the stakes are different entirely. The moment AI decides the route and the car moves on its own, the transfer of authority is not a question of which service gets called. It is a question of life.
The OEM builds the car. The Tier 1 implements and integrates the system. Big tech provides the platform. When something goes wrong, where does the accountability go? Authority is moving. The rules for where it went are not finished yet.



Whose Experience Is It?

The automotive industry has spent decades developing technology to reduce what the driver's hands do. Now it has begun reducing what the driver chooses. The pattern I kept encountering while covering Auto China 2026 is showing up inside the cockpit too. The system decides. The accountability does not follow.
I once wrote that we no longer look at cars. What we were looking at was not the car, but the time that flows inside it — and the quality of that time was the experience. Xiaomi bringing HyperOS to the show floor, Huawei stacking its entire ecosystem into one booth: both were, in the end, declaring their intention to design that time.
OEMs, Tier 1s, platform companies, and ecosystem companies are all moving toward the same position at once. As their decisions accumulate, two things are happening simultaneously: the experience gets better, and the authority to choose that experience on the user's behalf is shifting. There are no rules yet for the second part.
The authority to choose the automotive experience on your behalf — right now, whose is it?

AEM(오토모티브일렉트로닉스매거진)