The handoff loop: delegate in the car, review at the desk

Two weeks after Tesla put Connectors and Grok Bot in the cabin, the most interesting owner reports aren't about coffee orders. They're about handoff. One owner described asking Grok, mid-commute, to move a set of PDFs from a folder on his home Mac to a work drive so they'd be waiting when he arrived, then having an agent draft a near-final presentation from his existing sales materials. Another described capturing a video idea under FSD Supervised, sending it to his agents for research and an outline, and picking the thread back up later on his phone. Neither was reading anything on the center screen. Both arrived to work that was already moving.

That's the pattern this site has called the missing product: a thought started in one place and finished in another without being retyped. It isn't solved end to end, but a usable version now exists for people with the right stack. This playbook is how to run it without breaking Driving-mode rules, and how it changes when you're riding.

The loop in three parts

• Dispatch (in motion, voice only): say what you want done, where the inputs live, where the output goes, and what done looks like. Then stop talking about it.
• Run (while you drive): the agent works in the cloud or on a machine you own. You don't watch it. A progress update should be one spoken line or nothing.
• Review (parked or at a desk): open the result where you can actually read it, fix it, and decide what ships. That's where judgment happens.

Most failures come from collapsing these steps. Reviewing in motion breaks Driving mode. Dispatching vaguely turns the review into a redo. The loop works when each part happens in the place built for it.

What's worth handing off from the car

• Fetch and file: move, rename, or gather documents into a known folder. Easy to verify, easy to undo.
• First drafts from existing material: a deck, a brief, or a summary built from sources you name. You'll edit it anyway, so the draft only has to be a head start.
• Research with a deliverable: compare three vendors, summarize a long thread, outline a talk. Ask for a specific format so the review is fast.
• Prep for the next meeting: pull the last notes, open questions, and the relevant doc into one place before you walk in.

Keep anything that sends, pays, publishes, or deletes on your behalf out of the loop unless there's a review step in front of it. An agent drafting a reply to a client is a handoff. An agent sending it while you merge onto the freeway is a bet.

How to spec a task by voice

You can't scroll back through a voice prompt, so give it a shape you can say in one breath. A good dispatch has four parts: the outcome, the source, the destination, and the finish line.

Draft a six-slide deck on the Q4 pricing change using the pricing folder and last month's board notes. Put it in Drafts in my work drive. Done means speaker notes on every slide. Don't share it with anyone.

Name the folder instead of describing it. Say what not to do. If the agent asks a clarifying question you can answer in a word, answer it. If the answer needs you to look at something, tell it to hold until you're parked.

Driving mode vs Riding mode

• Driving mode, including FSD Supervised: dispatch only. Listen for a one-line confirmation that the task started. No reading drafts, no choosing between versions, no approving anything on the screen. The review waits.
• Riding mode (robotaxi, Cybercab, or the passenger seat): you can close the loop in the car. Dispatch at pickup, review on a laptop mid-ride, and ship before you arrive. This is where the commute becomes a real work block instead of a queue.
• In between: if your ride is a short robotaxi hop with unpredictable pickup times, plan for dispatch and light review, not deep work.

The setup that makes it safe

• Give agents the narrowest access that works. A home computer an agent can reach is a powerful node and a real exposure, so scope it to the folders you actually hand off from.
• Send outputs to a drafts or inbox folder by default, never straight into shared spaces.
• Make the review a habit: the first five minutes at the desk, or one parked minute before you walk in, goes to checking what ran.
• Keep a log. If you can't tell what an agent did while you drove, you won't trust it next time, and you shouldn't.

The car is a good place to start work and a bad place to approve it. Dispatch in motion, review at rest.

This sits next to When Grok Bot can act from the car, which covers which errands are glance-safe, and Assigning work to Grok Bot from the car. Those are about what the cabin can do. This one is about what happens after you stop talking: the work keeps going, and you come back to it with your eyes free. That's the version of in-car productivity worth building toward, and it's one where safety and usefulness point the same way.