Driving one of the most advanced cars in the world, a Tesla Model Y, got even more exciting with the availability of Grok AI since a couple of weeks ago. I tested talking with Grok while driving for several usecases, and was both amazed of what is possible and disappointed of the limitations. The situation right now in one sentence:

Grok as an excellent aid for short, situational interactions, but has yet to develop its full potential as an all-in-one “intelligent” car assistant.

Situational Constraints

Both the driver and the current version of Grok are very limited in how they can interact with each other in a moving vehicle.

Driving a car at high speeds limits the driver’s attention span and cognitive attention, at least as long as FSD (full self-driving) is not here yet. Having to monitor the street ahead forbids the driver from looking at the screen, making the interaction audio-only. No visual aids, like diagrams, drawings, or film can be used.

Grok is also serverely limited. At the moment, it does not remember previous conversations, limiting the content to just the current session. Also, it seems to be a stripped-down version of the “full” model available on the website.

To summarize: Using an AI in a car is not the same as using an AI on a laptop.

Works Great

Navigation is one of the obvious usecases in a car, and the good news is that Grok shines in this topic. Additional navigational information like “What’s that big building to my left?“ or “Tell me about the history of this city“ works great. These queries are short and contextual, not needing much attention from the driver and not requiring much context.

Working in tech, I am constantly learning or preparing new topics, which is partly supported by Grok. Asking detailed questions about topics already known can be answered even without visual aids or much context. This is an interesting option when thinking about a topic while driving.

This also applies to requests like detailed weather reports. They would be available with the built-in weather feature, but are especially great with Grok because they can be activated via voice accessibility instead of having to use the screen. Also, additional information, like a destination, can be requested more easily.

Works “Kind of”

Another obvious usecase are news reports while driving, with the possibility to ask more detailed questions. These are working kind of okay, but lack depth of information in my tests. Also, you have to be aware of a potential political bias of Grok.

Since AI came around, I’ve used it for coaching myself (personally, professionally, topic-related, learning coach). Doing this with Grok in a Tesla works, but Grok looses context after every session, limiting the usefulness as a coach.

Does Not Work (Yet)

Because of the audio-only interface, the limited capability of Grok or other reasons, some usecases don’t work at all.

As already mentioned, I strive to learn constantly, which of course includes completely new topics. Doing this with an audio-only AI is nearly impossible, as the structure of knowledge is best understood with a graphical representation like a mind map. Without those, only a broad overview can be understood through a conversation.

I assume because of privacy reasons or maybe even security, Grok does not (yet) have access to all car functions like adjusting the temperature, access to the outer cameras, or control abilities in the car. It is currently limited to navigation. While I totally understand the dangers of having a non-deterministic neural network having access to critical components in a moving car, I see a lot of potential currently not used.

One last, less invasive but very interesting feature is about media control. I would totally like to have a podcast search / recommendation! Sadly, this too is not yet supported.

Conclusion

While still in development, Grok in a Tesla today is useful for contextual, low-cognitive-load information while driving. More complex tasks lack the support of visuals and system integration.