AI agents are coming to Rivian’s cars

Rivian intends to let AI agents built by other companies run inside its electric vehicles, founder and CEO RJ Scaringe said onstage at TechCrunch Disrupt this week. He named Anthropic as an example of a frontier lab whose products could eventually operate in the car, and described that prospect as where the experience "becomes magical." No date was offered for the rollout, and Rivian declined to share further details, directing questions instead to an "AI & Autonomy Day" event it plans to host in December.
The plan builds on an assistant Rivian recently shipped, which lets drivers and passengers control vehicle functions by speaking to them. Underneath it sits an in-house agentic framework that works out which action a person is asking for and routes the request to the appropriate part of the car's software. Typical uses are mundane: lowering the cabin temperature, switching on seat warmers, or having the car summarize the news.
Scaringe described the routing logic with a traffic-light metaphor. When someone speaks to the vehicle, the system decides which model to use. A simple request can be handled by a local model running on the edge, inside the car. A more complex question gets sent up to the cloud, and even there Rivian draws on more than one model depending on the task. That split matters because in-car compute is limited and connectivity is not always available, so keeping basic commands local reduces latency and lets core functions keep working without a network connection. Running a frontier model such as Claude in the vehicle would almost certainly mean a cloud round trip, which introduces dependence on cellular coverage and raises questions about how fast responses feel.
This is not the company's first move toward outside services. As of May of this year, the agent could reach external apps including Google Calendar. Scaringe said the next step is to add products from the frontier AI companies. He was candid that the integration work is substantial, describing a lot of plumbing behind the scenes that Rivian has to build, along with coordination with the labs themselves to make sure they enable that plumbing. That phrasing points at the unglamorous parts of such partnerships: authentication, permissions, data sharing agreements, latency budgets and the APIs a lab must expose before an outside agent can act on a user's behalf.
Safety shapes where these agents are allowed to reach. Because the software sits near safety-critical systems, Rivian has to be extremely careful, Scaringe said. He drew a hard line: vehicle-level controls are not being handed to an agent. In his framing, you would not want an agent deciding it is time to apply the brakes or steer the vehicle. That leaves agents in the cabin-experience layer — climate, seats, media, scheduling, conversation — rather than in the driving stack. For developers, that boundary is the key architectural fact: any third-party agent integration will be sandboxed away from motion control, which likely means a defined set of permitted actions rather than broad access to the car's systems.
Scaringe expects the eventual implementation to look like an extension of how people already use agentic apps. That creates a harder problem than simple command routing: the car would need access to the context those apps accumulate about a user across other devices. His illustration involved a driver asking the Rivian assistant to turn on the ventilated seats, then in the same breath picking up a conversation from the day before about a meal, mentioning guests coming Friday and asking for help planning the menu. For the car to respond usefully, it would have to pull context from a conversation the driver had with the same assistant on another device. Delivering that requires identity and memory to travel between phone, home and vehicle, which is as much a data-governance question as an engineering one.
Rivian would not be the first American automaker to bring AI agents into its software. Earlier this year, Tesla added a feature letting users talk to both SpaceX's Grok chatbot and its Grok Bot assistant. Rivian's differentiator, per Scaringe's remarks, is the ground-up nature of its in-car software: rather than leaning on Apple CarPlay or Android Auto, the company built much of its stack itself, including navigation, while still integrating popular in-car apps such as Apple Music. Owning that layer gives Rivian control over routing and permissions that automakers relying on phone mirroring do not have. A likely trade-off is that the same closed approach makes it harder for outside developers to build for the platform, and that any lab Rivian partners with gains a privileged position inside the cabin.
Why it matters: Automakers are turning the cabin into another surface for third-party AI agents, and Rivian's stated approach — local models for simple requests, cloud models for hard ones, agents blocked from vehicle controls — is an early template for how that will be governed. Developers building agentic products should watch the December event for what integration actually looks like, since context portability between devices, not raw model capability, is likely to be the limiting factor. Product teams at rival automakers will face similar plumbing and safety questions, and the answers will shape which assistants passengers end up using on the road.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
We'll likely learn more about how this will work at the company's upcoming "AI & Autonomy Day" event in December.