Grok Bot has just officially launched in @Tesla vehicles. I got early access a couple weeks ago. You can order stuff on online (such as Amazon), manage your email, calendar, finances, or other tasks through Grok @Bot using just your voice in your Tesla. There is no separate app, everything just lives within the existing Grok Voice interface in your Tesla (as shown in my Model Y below). One good use case for Grok Bot in your Tesla: You get in, tell Grok Bot to order coffee at a location of your choice, and while your Tesla drives you to the coffee shop using FSD (Supervised), Grok places the order so it’s ready when you arrive. Grok Bot in Tesla vehicles is currently only available to SuperGrok Heavy users, but will expand to more owners later. Connectors are available to all users.

A few Grot Bot @Bot enthusiasts with @Tesla vehicles were invited to a small super secret beta testing group last week. We were asked to keep this under wraps until the service was ready to go live. That time has come! @Grok in your Tesla vehicle can now communicate to your Grok Bot team. Look to @SpaceXAI for the rollout plan. I know this may initially seem awkward, but wait until you try it. The fun and productivity you will find will amaze you. If you own a Tesla and haven't set Grok Bot up yet, now might be a good time to get a basic setup started, even if just to explore what it can do for you. We were not paid to do any of this testing or promotion, just enthusiasts enthusiastic about what is possible when you start connecting the dots between difference technologies. The possibilities are endless.

Grok @bot is now integrated into Tesla. Thanks to all the teams working hard on this. It can handle: - DoorDash / OpenTable / Uber Eats for pickup and restaurants on your way home or to work, while discussing what to order with family - Discussing your work and even spinning up a PR - Prepping documents on the way to meet clients And more! Tesla + Grok @bot placed the coffee order when a Cybercab drove me to work.

Over the last few weeks, Grok Build teams have optimized the harness for 4.7: - Better presentation, less slop, and better prose - Better self-validation - Improved thinking efficiency - Improved long-horizon monitoring tasks - Improved VLM and multimodal tasks Many FSD and Cybercab features I personally shipped were done with my overnight agents. Give it a try.

13 years ago, my team at NVIDIA was helping Audi develop vision systems for self parking at that time. The video itself still looked impressive, even today, especially considering the compute was tiny at that time. Lots of CUDA graph optimization went into it to make it run — like separable convolutions that serve as bedrocks of inference engine today. It sounds trivial today but backend we had to rearchitect how GPU reads camera frames isochronously without causing back pressure and jitters. Yet, after the demo we stopped working on it. It didn’t go anything except making video and playing it at all-hands. Back then, I had a code repo for “cool-demo-but-never-shipped”. It’s a graveyard. That prompted me to decide joining Google to at least ship a thing or two since I felt bored at writing papers. Shipping is hard, but at least my family can use it.

I have some sympathy for people who work on LLMs. A lot of the job has become the pursuit of bar charts. For years, Full Self-Driving only had to be good where it actually matters: the world. Ourselves, friends, families, customers, and unscripted miles. Real weather, real attention, and real consequences. If we had been forced to chase those benchmarks, I don’t think it would have become a good product. You can climb a chart without ever having to be present. Presence is the whole job.

In retrospect, game engines were not that different from inference engines — both need to support multiple users, batching requests, and agentic workflows (farmers and bots). Gaming studios used to publish many papers on scalable engines that moved the frontier.