Farzad: "Exactly What's Going to Happen" + "The HOW Should ALWAYS Be the Job of the Agent"

Elon says SpaceX will try to catch both Super Heavy and Starship on the upcoming flight. Hours later he said information moving through space will ultimately exceed current Earth traffic by "a trillion trillion times." Farzad says that's "exactly what's going to happen." Then he spent Friday evening posting about agents.

"Catch both"

Elon, Friday at 9:47am CST: SpaceX will attempt to catch both the Super Heavy booster and Starship on the upcoming flight.

If both are successful, it will mark the first time that all parts of an orbital rocket return to Earth intact.

This is the critical breakthrough needed to make life multiplanetary.

The "first time" claim and the "critical breakthrough" framing are Elon's. The post announces an attempt and doesn't say what counts as success for either catch. It names no flight number or date. Yesterday's Digest covered the reported target date.

Elon's post quotes an 8:56am CST SpaceX post promoting "Go For Orbit." Its full text describes "the first orbital Starship launch carrying advanced Starlink satellites bound for space" and invites viewers to "experience the moments behind flying the most ambitious Starship mission yet." This is film promotion, not a flight date or number.

Farzad RTd Elon at 9:52am CST without adding anything. At pull time Elon's post had 80,044 likes and about 3.86 million impressions. Saturday at 6:06am CST Farzad posted a video titled "SpaceX Just Made A Huge Announcement." This pull has only the title, so it doesn't show which announcement the video covers.

Check Elon here. SpaceX here. Farzad's RT here.

"A trillion trillion times"

Elon, Friday at 7:09pm CST, replying to XFreeze: Information traveling through space in our solar system will ultimately exceed current Earth traffic by a trillion trillion times. People have no idea.

A trillion trillion is a 1 followed by 24 zeros. Elon gives no timeline beyond "ultimately" and doesn't say how he'd measure traffic. The XFreeze post he replied to is available in full and discusses a projected Starlink laser network. Its technical figures remain XFreeze's claims, not verified specifications here.

Farzad quoted it at 7:16pm CST:

My favorite thing about this post is that most people will read it and think Elon is literally batshit crazy but the reality is that this is exactly what's going to happen and it's actually extremely weird that it's going to happen 😱

That's Farzad's projection. He doesn't attach a date or a number of his own.

Earlier, at 3:05pm CST, Elon replied to another XFreeze post: When we unveil the details of the Starmind computer next year, it will blow people's minds

That's the whole post. It has no specs and no timing beyond "next year," meaning 2027. The XFreeze post it answers discusses proposed orbital computers; Elon's reply promises details, not a confirmed deployment date. Farzad RTd it at 7:17pm CST, about a minute after his quote post.

At 7:20pm CST, SacctWits replied to Farzad's quote post: One missile can bring down billions worth of data centers. How are they going to secure them ?

Farzad answered at 7:23pm CST: Why do you think America create the Space Force?

Less than a minute later he posted: So Space Force was kind of a prescient move wasn't it.

That's his opinion. He doesn't say what Space Force would protect or how.

Check Elon on traffic here. Farzad's quote post here. Elon on Starmind here. Farzad's RT here. SacctWits here. Farzad's reply here. The Space Force post here.

"The HOW should ALWAYS be the job of the agent"

Seconds after the Space Force post, at 7:24pm CST, Farzad changed subjects:

The craziest thing about agents is that you DO NOT want to be too explicit about HOW you want something.

The HOW should ALWAYS be the job of the agent.

If you don't do this, you will not benefit from the creativity and IQ boost that future models will get, and you'll have to spend a ton of your time optimizing prompts until models get smart enough to tell you that you're being dumb lol

That's his opinion, drawn from his own use. His argument is about time. Spell out the method today and you lock in today's method, so a smarter model later is still following your old instructions.

He liked several replies. clickheer, 7:27pm CST: just like talking to a person, you should tell them WHY to do something, because they may be smarter than you and can figure out the best way HOW to do it without your intervention

Mike Kock pushed back a little at 7:29pm CST: Sometimes they do things in a stupid way too if you know the better way for your scenario and scale.

His fix is to state scaling requirements up front. "Fine line between being explicit enough in requirements while not being explicit in implementation."

Dave Lee, 9:03pm CST: True. Principles over rules.

Todd Gleason, 9:07pm CST, says to treat it "as a high-level manager: assume they know what they're doing, and when they prove otherwise, dig into where they went wrong." He ends with "accept that there will be fewer and fewer times that they fail to outthink you." Farzad replied a minute later: Yes.

At 10:32pm CST Farzad added: Working with AI feels like you are so far away from its true potential, but you keep getting 10x better every 3 months or so.

The 10x figure is his own sense of progress. He doesn't say what he's measuring.

Check Farzad's agent post here. clickheer here. Mike Kock here. Dave Lee here. Todd Gleason here. Farzad's reply here. The 10x post here.

"PLEASE read this"

At 9:11pm CST Farzad quoted a reply to his agent post: If you use agents a lot - PLEASE read this.

The reply came from pyginapithon at 8:38pm CST. The full note_tweet text is available, including the discussion of saved corrections, GitHub work and weekly reviews.

pyginapithon starts with "I manage agents like competent people." Then: "They forget almost everything that isn't deliberately saved." Grok Bot and Build "both keep a 'priors' list," and in pyginapithon's words, "if I say 'review your priors,' they know what I mean. That's where recurring corrections have to go."

Then comes a specific case. Their bots had been defaulting to Cursor cloud agents for GitHub work, so pyginapithon has to remind them to use their GitHub token. "The instruction only sticks if it's written into their priors or skills." They don't use Cursor cloud agents themselves. They call them "cold (blank) agents" that start from a fresh clone of the repo with no priors about the project.

pyginapithon also admits they "still clutter their priors with extra commentary," which takes a weekly prior review to clean up.

Next to Farzad's post, the two fit together. Farzad says to leave the method to the agent. pyginapithon says to write down what the agent keeps getting wrong. At pull time, Farzad's quote post had 103 bookmarks and 116 likes, while pyginapithon's post had 166 bookmarks and 79 likes. Only the latter had more bookmarks than likes.

Check pyginapithon here. Farzad's quote post here.

Grok Bot, Hermes and shopping agents

Mike P (mikepat711), Friday at 9:10pm CST: I've been messing with Hermes all day, and it's just making me love Grok Bot more. I never messed with OpenClaw or Hermes before Grok Bot. I was a 100% Claude user (Cowork, dispatch, code, etc).

This all feels like the past. Like Grok Bot is the next evolution of what this is.

It's obviously very cool and has the benefit of letting you control the models, or even run models yourself. But for the majority of users, it just doesn't hold a candle to Grok Bot.

That's one user's comparison after a day with Hermes. Farzad liked it.

Alex Finn, 7:47pm CST, says that after "using Grok Bot nonstop for the past couple months," it's "the best AI agent ever." He says it "has automated basically 100% of work I don't want to do." His examples include watching AI companies' X accounts every 15 minutes, testing two apps he's building and writing PRs for his review, and running a site he says makes revenue while he puts in "0 effort." He also concedes: "There's literally nothing Grok Bot can do that the others can't." He credits the user experience. Farzad liked it and RTd it at 8:33pm CST.

Both are user opinions. And Farzad isn't neutral here. He builds his own Grok Bots, Solo and Shorty, and links them under his videos.

Earlier, at 7:56am CST, Alexis (@alexisohanian) posted: My agent is buying a LOT more from Shopify-powered stores. Not investment advice, but if I were running an e-commerce business that didn't support agentic commerce, I'd feel like I was making a big mistake.

That's his opinion, based on his own buying. He doesn't say how much more or over what period. Farzad RTd it at 9:54am CST.

Check Mike P here. Alex Finn here. Farzad's RT here. Alexis here. Farzad's RT here.

"Must watch"

Alejandro Sahuquillo, Friday at 1:30pm CST: After 7 years of research and 6 months making this documentary, I'm handing you what I believe are some of the most valuable 51 minutes you'll ever spend.

I interviewed former @SpaceX, @Tesla, @neuralink and @PayPal employees. Read 40 books. Took 9,200 notes.

Those numbers are Sahuquillo's. The pull doesn't include the documentary's title.

At 2:25pm CST he thanked Farzad: Thanks, @farzyness, for letting me use your proprietary clip. And for being part of the documentary!

So according to the filmmaker, Farzad is in it. Farzad liked that post and replied at 2:30pm CST: "Excited to see it do well!" Moments later he quoted the original: Must watch.

Keep that in mind when you weigh the recommendation. At pull time the post had 1,513 bookmarks, more than its 920 likes.

Check Alejandro Sahuquillo here. His thanks here. Farzad's reply here. Farzad's quote post here.

Robotaxi app: one rider locked out

Austen Allred, Friday at 10:46am CST: Dear Cybercab/Robotaxi team:

I was on an early Testflight version of the Robotaxi app, and now my account will not let me book rides no matter what I try.

Can someone please help me fix?

Farzad RTd it at 11:12am CST. Four seconds later he replied with just "@Austen @robotaxi" to pull in the account.

This is one rider's report. The pull doesn't show a response from @robotaxi, a fix, or whether anyone else hit the same problem.

Check Austen Allred here. Farzad's RT here. Farzad's reply here.

Quick hits

Sawyer Merritt, Friday at 3:30pm CST: SpaceX has just announced that after over a decade of service, Don Harrison has stepped down from SpaceX's Board of Directors, effective November 4, 2026.

The full post quotes SpaceX's thanks and says the company's 8-K reports no disagreement over its operations, policies or practices. SpaceX's November 6 announcement independently confirms the departure and says Harrison joined the board in February 2015. The 8-K confirms the November 4 resignation, effective immediately, and the no-disagreement statement. Elon replied at 7:46pm CST: "Don has been a great board member. Very much appreciate his contributions." He added: "At this point, there are simply too many conflicts of interest with Alphabet, so he decided to step down." That explanation comes from Elon, not the filing. Farzad liked his reply.

At 12:28pm CST Sawyer quoted Elon: "I insist on doing some of the work myself manually. I've pulled power cables the size of an anaconda. I've done the fiber optic cabling; I try to do at least a little bit of everything so I'm not numb to the pain of the people who are actually doing it. It's not as though I'm sitting in an office and someone has to come to me. I'm walking the floor asking them questions actively." Sawyer frames this as an account of work at the Memphis data center, from the 33rd Annual Baron Investment Conference in New York. This is Elon's account of his work, not an independently observed work log. Elon quoted Sawyer at 1:23pm CST: Pulling massive power cables is a great workout and far more productive than lifting weights 🏋️‍♀️

Farzad liked it.

Check Sawyer on Harrison here. Elon's reply here. Sawyer on Memphis here. Elon's quote post here.

Farzad's videos

Farzad posted four videos between Friday at 12:33pm CST and Saturday at 6:06am CST. This pull has only their titles. Under "Cybertruck Is Making Room for Cybercab" he replied: "Build your own short form content with my Grok Bot Shorty." Under "SpaceX Just Made A Huge Announcement" he replied: "Made with my Grok Bot Solo."

Farzad says the narration in the SpaceX video uses a clone of his voice. At 7:22am CST Saturday, TwoWeeksLOL asked under that video: "Did you clone your voice for it too?" Farzad liked the question and replied at 7:26am CST: Yes using ElevenLabs v4 model.

Check Cybertruck Is Making Room for Cybercab, Friday at 12:33pm CST, here.

Check Google Just Gave Robots A New Way To Learn, Friday at 3:04pm CST, here.

Check Elon Musk Stuns Billionaire, Friday at 8:25pm CST, here.

Check SpaceX Just Made A Huge Announcement, Saturday at 6:06am CST, here.

Check Farzad's voice reply here.