Rivian has introduced an in-car voice assistant and now plans to allow third-party AI agents to run on its vehicles, CEO RJ Scaringe said at TechCrunch Disrupt. Scaringe named companies like Anthropic as potential partners and described support for external agents as a future step without providing a timeline.
Built-in software and an agentic framework
Unlike systems that rely primarily on Apple CarPlay or Android Auto, Rivian developed much of its in-car software internally, including navigation and a custom agentic framework. The company says that framework interprets user requests and routes them to the appropriate software component, using a mix of local models on the vehicle and cloud models for more complex queries.
Scaringe described the system as a kind of “traffic light”: simple commands can be handled by edge models, while complex requests can go to cloud models. As of May of this year, the agent could access external apps such as Google Calendar, Rivian said.
Third‑party agents and safety limits
Scaringe suggested agents from frontier AI companies — for example Claude, OpenAI’s Dots, or Meta’s Muse — could be integrated, but warned the work requires significant “plumbing” and coordination with the labs that build those models. He also emphasized safety boundaries: “You wouldn’t want an agent to determine, ‘hey, it’s time to apply the brakes, or time to steer the vehicle.'” According to Scaringe, vehicle-level controls will not be handed over to agents.
He outlined a user scenario in which the vehicle accesses context collected across devices: a driver might ask the Rivian assistant to ventilate seats and also reference a prior conversation about planning a meal, with the car pulling relevant context from another device’s interaction with an agent.
Rivian pointed attendees to its upcoming AI & Autonomy Day in December for further details. Earlier this year, Tesla added access to SpaceX’s Grok chatbot and its Grok Bot assistant, illustrating a broader industry move toward embedding agentic AI in cars.
Original source: TechCrunch AI