Watch
Short notes on FSD, Cybercab, cabin AI, and connectivity. We track them for one reason: what they change about doing real work in a vehicle.
Voice-off-the-glass controls are the headline: adjust climate, media, and navigation by talking to Grok. Entertainment still leads the demos, which is fine. The thing to watch is whether work surfaces follow — can it pull up a doc, start a meeting, or hand off to your laptop? Until then, the voice layer is a nicer remote, not a productivity layer.
Public robotaxi rides are scaling in Austin and a few other jurisdictions, which makes Riding mode real for more people. The caveat is the patchwork: what is legal in one city is not legal twenty miles away, and insurance rules are still catching up. Treat the map as local and temporary.
The first Cybercab interiors suggest the cabin is being designed around entertainment more than work. That is a choice, not an oversight. If the productivity pitch is real, we need to see a usable work surface, a screen that extends a laptop, and seating that supports typing. Until then, the robotaxi leg is a consumption slot, not a deep-work block.
Recent Tesla interface leaks show Grok getting persistent context inside the car: it can see prior conversations, resume threads, and surface reminders tied to location. That is the beginning of continuity memory, not just a better voice command. The test is whether it can hand off cleanly to the desk and phone, or whether it becomes another silo.
The two-seat robotaxi is being sold partly on what you do instead of driving. Watch three details rather than the headline: whether the cabin gets a usable work surface, whether the screen supports anything other than entertainment, and whether the seating geometry works for a laptop. If those land, a robotaxi leg becomes a genuine deep-work block. If not, it is a theater on wheels.
The main reason in-car assistants fail is not intelligence, it is coverage. Every rural stretch that drops a session breaks the habit. Direct-to-vehicle satellite links change the calculation for cloud-dependent tools, and would make riding-mode video calls viable on routes where they are currently a joke.
Riding mode is not a technology question, it is a jurisdiction question. Where unsupervised operation is approved, the whole trip becomes work time. Where it is not, you are in driving mode no matter what the car can do. Expect the map to stay patchy and the rules to keep changing — check locally, every time.
Nobody has solved continuity. A thought started at the desk, developed in the car, and finished on a phone still passes through three separate assistants with no shared memory. That handoff is the largest open opportunity in this category, and it is a software problem, not an automotive one.