In-cabin AI is moving past "what's the weather" toward actually getting things done: booking a table, drafting a calendar, ordering flowers. The bigger story isn't the capability. It's who ends up controlling it once the request leaves the car.
AI agents in the car are moving beyond better voice assistants toward actually completing tasks for drivers, which could make the cabin dramatically more useful. ABI Research estimates AI-driven in-cabin experiences will grow from about 5 million vehicles today to 70 million by 2035.
What This Actually Looks Like
The examples aren't exotic: booking a restaurant, ordering flowers, handling a productivity task, all through a natural voice request instead of a list of nearby options to scroll through. Paul's own experience outside the car made the case better than any industry projection could.
My six-year-old's flag football schedule came through as a text. I dragged it into Claude and said make it easy to add these to my calendar, and within 10 seconds it spit out a file I just double-clicked to add.
That's the exact shape of task this technology is headed toward handling natively while driving: a notification comes in, the driver says what they want done, and it's done, no scrolling, no typing, no pulling over to sort it out later.
The Risk Automakers Are Actually Taking On
Cubic3's Steven Cochrane named the tension plainly: no single company can own the whole stack, what's in the vehicle, what's in the network, and what's in the cloud. The moment a request leaves the dashboard, the automaker is dependent on connectivity, latency, and whichever AI provider is actually doing the work.
Why This Fight Already Has a Precedent
Paul and Steve both landed on the same comparison: CarPlay and Android Auto. Automakers spent years trying to build native infotainment systems worth using instead of a phone. GM held out the longest and eventually gave in anyway, because building something good enough to compete with what's already on a driver's phone turned out to be harder than just plugging into it.
I just don't think you're going to want to duplicate that effort. You're going to set up one connector and say, here you go, this is what I use now.
Steve's read: this is a two-provider race right now, and any automaker not building around one of them is building a system drivers will bypass the same way they bypassed native infotainment. Tesla's early integration of Grok is the current example of what a vertically-built version of this looks like in practice.
Want the full conversation?
🎙️ Listen to today's Automotive State of the Union episode for the complete discussion.
The lesson isn't really about AI. It's that this industry keeps underestimating how fast a mediocre proprietary experience loses to the one already in someone's pocket.
Automakers have a real choice to make here, and it's the same choice they already made once with CarPlay: spend years building a worse in-house version, or partner early and keep some say in how it works. The dealerships watching this unfold aren't the ones building the technology, but they'll be the ones fielding questions about it the moment customers expect their car to do what their phone already does.
