Rivian Plans to Run Third-Party AI Agents in Its EVs
Rivian already lets people talk to its electric vehicles through its own AI assistant. Founder and CEO RJ Scaringe now says the company wants to go further: running AI agents built by outside companies such as Anthropic inside its cars.
Scaringe described the plan onstage at TechCrunch Disrupt this week. "That's where it becomes magical," he said. He gave no timeline. Rivian declined to share more details and pointed to its "AI & Autonomy Day," an event the company is hosting in December.
Why Rivian can attempt this
Most carmakers rely on Apple CarPlay or Android Auto for much of the in-car experience. Rivian took a different path. It built its in-car software from scratch and wrote many of its own applications, including navigation. It has still integrated popular apps such as Apple Music.
Owning the whole software stack gives Rivian room to plug in outside assistants. TechCrunch suggested candidates such as Anthropic's Claude, OpenAI's Dots, or Meta's Muse assistant, though Rivian has not confirmed any partner by name beyond Scaringe's mention of Anthropic as an example.
How the current assistant works
Rivian says it built its own agentic framework. When someone speaks to the vehicle, the system works out what action the person wants and sends the request to the right part of the car's software. In practice, that usually means routine tasks: lowering the temperature, switching on the seat warmers, or summarizing the news.
Scaringe compared the system to a traffic light. "If you ask the vehicle something, it decides what model it's going to use," he said. Simple requests can go to a local model running on the vehicle itself. More complex questions can be sent to the cloud, where Rivian uses several different models.
Since May, the assistant has been able to reach external apps such as Google Calendar. Adding products from frontier AI labs would be the next step.
The hard part: plumbing and safety
Scaringe was open about the difficulty. "There's a lot of plumbing behind the scenes we have to do to make sure that that works, and it's not just within Rivian," he said. It will also take close coordination with the AI labs so they enable that plumbing on their side.
He also stressed that this software would sit close to safety-critical systems, so Rivian has to be extremely careful. The company draws a clear line here. "You wouldn't want an agent to determine, 'hey, it's time to apply the brakes, or time to steer the vehicle,'" he said. Vehicle-level controls will not be handed to an agent.
What it could look like in the car
Scaringe expects third-party agents in the car to work as an extension of how people already use these apps. That means Rivian would need a way to connect the vehicle to the context that assistants like Muse gather about their users.
His example: a driver asks the Rivian assistant to turn on the ventilated seats. In the same breath, they say, "We were talking yesterday about this meal. I'm planning to have guests over on Friday. Can you help me think through what we're going to make?" The car would then pull in context from a conversation the driver had with the assistant on another device.
Rivian would not be first. Earlier this year, Tesla added a feature that lets users talk to both SpaceX's Grok chatbot and its Grok Bot assistant.
Our Take
The most useful detail here is not the "magical" pitch but the architecture. Rivian keeps the agent away from braking and steering, routes requests between on-device and cloud models, and treats outside assistants as an extra layer rather than the core. That matches a wider pattern in agent design: the safest controls sit around the model, not inside it.
The pitch also depends on personal context following the user across devices, an idea other companies are betting on with device context. That raises questions about data sharing between carmakers and AI labs, and about who is responsible when an agent gets something wrong - a topic insurers are already preparing for.
It is worth watching whether Rivian names partners or a timeline at its December event, and how strictly it limits what outside agents can touch.
