📰 Key Summary

Rivian founder and CEO RJ Scaringe said at TechCrunch Disrupt that Rivian’s EVs will eventually be able to run AI agents built by other companies, like the ones from Anthropic, calling it a “magical” step forward. Rivian already has its own AI voice assistant that lets drivers and passengers interact with the car directly through speech, but the company hasn’t announced a specific timeline for this cross-vendor integration yet — more details are expected at its “AI & Autonomy Day” event in December.

Rivian built its in-car software entirely in-house instead of relying on Apple CarPlay or Android Auto, and it developed its own agentic framework to figure out what action a user wants when they speak, then route the command to the right software module inside the car. Scaringe described the system like a “traffic light”: depending on how complex the request is, it picks which model handles it — simple commands (like lowering the cabin temperature or turning on seat heating) get handled by a local model on the vehicle’s edge, while more complex questions get sent to the cloud and routed to different models depending on the task. Since this past May, the agent has also been able to access external apps like Google Calendar.

Scaringe was clear that “vehicle-level controls” tied to driving safety — like braking or steering — will never be handed off to an AI agent’s decision-making, and bringing in products from frontier AI labs will still require a lot of behind-the-scenes engineering and cross-lab coordination. The use case he’s picturing is more of an extension of existing agentic apps — for example, a user could ask to turn on the ventilated seats while also asking the assistant for a suggestion based on context from an earlier conversation (like an upcoming dinner with friends), letting the car’s assistant pull in personalized context the user has already built up in other apps.

💬 JudyAI Lab Commentary

150-word news business writing task, no research or fact-checking needed — just write directly based on the given original summary.

Rivian CEO RJ Scaringe revealed at TechCrunch Disrupt that the company’s EVs will eventually be able to run AI agents from other companies, including products built by Anthropic. He called it a “magical” step forward, with more details coming in December.

This case highlights a layered design philosophy for in-car AI that’s worth a look for any AI builder: Rivian built its own agent framework to interpret what users mean when they speak, then routes requests based on complexity — simple commands go to a local on-device model, complex tasks go to the cloud and switch between different models. That’s essentially the same cost-versus-latency tradeoff a lot of AI applications are already making. At the same time, Scaringe drew a clear line — safety-critical vehicle controls like braking and steering will never be handed to an AI agent — showing that even as companies embrace agentic AI, they still need a clear boundary between “what AI can touch” and “what AI can’t touch.”

Something to think about: when designing your own AI product, it might be worth drawing a similar line up front — which actions can be handed off to an autonomous AI agent, and which ones need to stay gated by a human or a fixed rule.

📅 Source Info

🔗 Further Reading