#OpenSource contributor and technology entrepreneur | @founding | https://xitter.cf/t.co/jVd9McmctD | https://xitter.cf/t.co/CDkdPePwdq | prior Telestax, Mobicents, JBoss/ @RedHat
Austin, TX
Joined May 2009
- Tweets7.4K
- Following2.2K
- Followers1.7K
- Likes30.8K
FSD v14.3.12 release notes: one unified model now runs Actually Smart Summon, FSD and Robotaxi.
That's a big step. If it's the same model and Robotaxi runs unsupervised, regular Teslas are effectively one step from unsupervised.
What's left: Robotaxi runs in geofenced areas Tesla controls, backed by a remote assistance network. My Cybertruck has the same neural net and the same cameras, but not that safety net. Tesla still has to approve each vehicle class and location before dropping supervision.
Same brain, different body, different rules.
@Tesla @Tesla_AI
Source, from the v14.3.12 release notes: "Unified the model between Actually Smart Summon, FSD, and Robotaxi for more capable and reliable behavior."
notateslaapp.com/software-up…
DOGE for bot plz
Token waste compared to Grok App and Grok Build is an order of magnitude worse.
I’ve never hit past 30% on Super Grok Heavy including Grok Build.
Grok bot is awesome but week after week it keeps silently eating up all Ultra weekly budget in one or two overnight background tasks without any visibility into its detailed usage logs. Only vague guesses what might have caused the token burn.
Goes from 25k line items in usage logs to 100k to 500k without clarity which bot performing what task for what good reason.
@poteto @grok bot @SpaceXSI
Earlier this month I started moving the heavy coding to machines I own. The bot budget still ran out. So I pulled last week's usage export.
Under 1% of tokens were the bot writing anything. Most of the bill is bots rereading their own history on every step.
37% of calls carried over 100k tokens of context. Those calls took about 60% of the spend.
The export can't say which bot did it. Still my ask.
Rolling out now: each job starts fresh and gets a pointer to its input, not a copy. Scorecard Nov 2.
What's the biggest context you've caught a bot dragging around?
Email used to be an app you lived in. Now it's a connector my bots read through.
Phone bots are next. Most voice platforms still want your call flows, your contacts and every call record living on their side.
If Grok Bot is where you actually work, that's the friction.
Would you rather your phone bot be an app you log into, or a plug-in your bot just uses?
The confusing thing with Bland and some of their competitors is that they want voice flows, system or record connectors and call decision traces residing on their platform. Users who choose grok bot as their main AI workspace, have to figure out how to navigate the tension and bridge the gap.
As analogy, Email clients used to be a monolithic app market. Now it’s just a connector to email protocols and storage.
VSCode and Cursor used to be monolithic apps with their own GUI and many plugins. Now cloud agents or local coding harnesses are highly optimized thin coding specialists bridge Grok Bot to a file version control backend.
We will have to see how GUI heavy platforms navigate the shift to added value plugins for agentic workspaces.
That’s one way to show progress with cutting gov waste. Trump admin sending apology checks for the healthcare.gov clusterf* .
Long long way to go with opaque healthcare insurance premiums , pricing and billing . Most of the $$ still goes to middlemen focused on paper not care.
@TeslaCharging consumer support is not at the level of @Tesla vehicle support.
Installing was the original yellow flag. Maintenance experience confirms it.
Raised this Sep 18 in the app. 6A draw vs 48A. After 8 hours the car had gained about 10%. Took the truck in Sep 28 at Tesla Service Austin-West. Vehicle service did not own it. Redirected me to Tesla Energy. User friction point 1.
- Trying to contacting @Tesla Energy support via the app resulted in a bounce to (877) 798-3752. User friction point 2.
- That desk was not available on the weekend, so Sunday was gone. Friction 3.
- @tesla App has some troubleshooting for Wall Charger but instructions are invalid or outdated, so dead end. Friction 4.
- @Grok has no good answers either. Friction 5.
- The official support line is (877) 798-3752. It answers with a stone age IVR menu as if grok voice is on another planet. Friction 6.
- Once I finally connected to a person, the call ran half an hour. He spent it telling me not to read what the Tesla app recommends and to ignore what Grok recommends. Friction 7&8.
- Then asked for photos and serial number information over email. Sent while on the call. Not good enough. Need more photos from different angles. That had to wait until I was home, which is outside their hours. Friction 9.
- Emailed the next batch Monday evening, as requested. Still no confirmation by Tuesday 11am. Friction 10.
- Called 877. IVR menus all over again. Different support agent. Could not find the case by my phone , name and even by the refid I had from the email thread. On hold. 10 minutes later call hang up. But guess what, only the agent was cut off, while I was asked to remain on the line for a survey... which was also broken :). Friction 11.
- One more 877 call. IVR. Miraculously the third rep instantly found my case and ordered an in-warranty replacement for the defective heat sensor. 7-10 business days estimated time to arrive with self replacement instructions. No Supercharger credits available while the Wall Connector is out. I wanted to leave good feedback to this agent in the survey, but the survey broke again :)
- Case is still developing...
I have the receipts. @teslaenergy
@elonmusk, the bucket is leaky. Pouring greatness from the top is sabotaged by a wobbly back office.
- New Wall Connector stopped charging the Cybertruck around 60% last night. Same unit still charges the Model Y fine.
Called Tesla Energy this morning. They won’t own it and sent me back to Tesla Service — the same Service that previously sent me to Energy. Classic loop.
Insisted on escalation. Someone is supposed to call back. We’ll see.
Alerts on the truck: HVBATT_a185 (Unable to charge) + PCS2_a062 (External charging equipment not providing power). Solid blue / red lights on the connector depending on the moment.
Foundation Series owner. Doesn’t seem to change anything. @cybertruck @Tesla @TeslaCharging @teslaenergy @elonmusk
Have your people talk to my people is back.
it won't be long before "have your bot talk to my bot" becomes common!
some fun use cases of multiplayer grok bot over email could be:
• debugging (you can send my bot bug reports!)
• customer support
• booking meetings and events
• group buys
• buying and selling stuff off craigslist
what have you used bot mail for?
How are you allocating resources with your Grok Bot now that the budget is unified with Grok Build ?
No F1 team lets an outside shop own the car. Plenty of founders do exactly that with their product.
Pit Crew Edition: Hire a CTO Bot, Own Your Roadmap. Bring your repo, app or backlog.
Tue Oct 20, 3-5 PM CT, Austin + online.
RSVP: luma.com/ovk9lgue
Even on the Stripe interview with Zuck, Alex keeps referencing his X feed as signal for Muse feedback. Not his Instagram, Facebook or LinkedIn feed.
There must be something about signal vs noise ratio on X. Early adopters , candid critics ? @grok ?
Muse Wonder: Zuckerberg and Wang's $100B AI youtu.be/XfdIhd1g6EI?is=bFRa… via @YouTube
Interesting that @alexandr_wang referenced his X posts about Muse and people reacting to his tweets a number of times during this awesome interview on the birth of Muse.
I don’t think I heard him talk about Instagram or Facebook posts even once.
youtu.be/S6l3aRsecuE?is=6Cpu…
What clicked for me: he said Meta has distribution to 3.5B people and that still wasn't enough to break through. So he wrote every tweet himself and retweeted dozens of real user wins. Reach is rented. Founder voice + customer proof is owned.
What finally fixed my AI chief-of-staff updates: one card, not a stream of pings.
Tue–Fri: a phone-sized daily card. Bottom line, numbered decisions (D1, D2) each with a recommendation and a default, what got done, today's calendar.
Monday: a traffic-light weekly card that absorbs all the weekly scans.
"Catch me up" anytime = the same card, fresh, minus anything already decided. I can say "D1. Elaborate" and decide in one tap.
Run by @grok Bot.
Real examples from today's cards:
🟢 "Mobile org chart is live on your X reply." (After I said the first one looked terrible on a phone. It remade it tall with big type.)
🟡 "Gmail filter is still trashing mail. Rescue pulled back 3 more real emails this hour." It caught that on its own.
D1 "Ship the waitlist page?" Me: "Elaborate." Five plain steps, a rec, a default. Tapped yes. Shipping.
D2 "Let the code-review bot comment on PRs?" Yes. 10 seconds.
The best part: it flags when it changes its own recommendation, and why.
My Grok Bot caught something I'd missed for days: a bad Gmail rule was quietly trashing my real email. It pulled 232 threads back out of Trash, ruled itself out as the cause, and set up an hourly rescue until I fix it. I never asked. Proactive agents are underrated. @bot
That’s nice. @spacexai credits users who publish cool apps. This was just a quick birthday party rsvp form with @grok build.
Less hassle than wrestling with existing SaaS apps to include a parent waiver for kids.
It would be nice to unify grok build CLI , grok build in app and grok @bot so they can all collaborate seamlessly on the same repo without multiple handoff steps.
