We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Ahead of the Anthropic IPO: Social Capital's deep dive
Chamath Palihapitiya said of Social Capital's Anthropic deep dive: "This one was really good." The firm says it spent three months on the report ahead of the IPO. It sets out to answer three questions: who controls Anthropic after it lists, how the company plans to outgrow its $900 billion computing bill, and why people who thought AI could end in catastrophe built one of the fastest-growing companies ever .
Casado: Netlify as the turnaround case study
Martin Casado endorsed Alex Konrad's interview with Netlify CEO Matt Biilmann. He called it "the case study for how AI has buoyed some companies that predate it" and "the single most significant turn around I've directly witnessed" . Konrad reports that Netlify was stagnating three years ago. It is now growing nearly 130% and will finish the year with $130M in ARR, after "re-founding" itself around non-developer users who are new to AI .
O'Shaughnessy: Casey Handmer's AI audit of astrophysics
Patrick O'Shaughnessy shared Casey Handmer's physics survey. He called Handmer "one of the more creative users" across his projects and said someone should give him unlimited tokens . Handmer had Opus 5.5 audit theoretical astrophysics. The project indexed about 150,000 papers, audited 50,000, and found nearly 1,000 major errors that were cited many times, most of them still uncorrected. Handmer concludes that peer review is "not very good" at correcting errors. He also says AI can speed up quantitative science by more than 1000x with no further model improvement .
Collison: Works in Progress, with human-made art
At a Semafor event, Patrick Collison told the audience to subscribe to Works in Progress, the magazine Stripe publishes . He calls Stripe "about as AI forward as a company can be," yet all the art for Stripe Press and the magazine is made by humans. His reason: people "want to know what humans chose to depict," and with AI art "the product would be worse" . Video
Armstrong: a Senra conversation, and the Civilization tech tree
Brian Armstrong called his conversation with David Senra "another great chat." He said it covers self-programming, the game of business, how AI is changing work, and why the public fears AI . Chapters include "Giving agents a job instead of a prompt" and his prediction that 99.999% of transactions will be agent-to-agent in 10 years. Senra's post notes the episode includes paid partnerships .
On another podcast, Armstrong said he keeps a printout of the Civilization tech tree in his home office because it is "such a fascinating concept." Playing the game made him want to "add a leaf to the tech tree" . He also likes that it has several ways to win and "so many layers," and says you can learn simple strategies from it . Video
Shorter picks
- Brad Gerstner called O'Shaughnessy's episode with Geoffrey Canada a "MUST listen" for understanding "the pulse of America - the fear, the anxiety & the hope." Gerstner is part of the conversation himself . The episode was covered in the previous brief.
- Jason Lemkin shared SaaStr's piece on Palantir. His takeaway is that Palantir "proves NDR does not necessarily have to come down at scale." The piece cites 165% net dollar retention . SaaStr is Lemkin's own company.
- Elon Musk pointed followers to Poteto's guide Automate Anything with Grok Bot, with the advice "Treat Grok Bot like a really smart employee" . The guide lays out the author's approach to automation at work and in personal life . Grok is an xAI product, so Musk is promoting his own company's tool.
- A Ramp co-founder, in a YouTube interview, said he and Kareem read Paul Graham's essays and watched early Startup School videos while working as a banker and a consultant. His takeaway from PG's essay on how to get startup ideas was to "notice problems in your regular life" .
- Brian Armstrong calls Civilization’s technology tree “fascinating” and says it inspired him to want to add a leaf to civilization’s tree; he also praises the game’s varied victory paths and layered strategy as a source of simple strategies.
- He calls Polytopia fun and groups it with short games that scratch the same itch as business-building over decades, while noting he has little time to play.
Eric Glimman said he and co-founder Kareem read Paul Graham’s essays and watched early Startup School videos; the essay he recalled on how to get startup ideas left him with the takeaway to notice problems in everyday life .
In arguing that modern business curricula should teach founder control rather than board control, Balaji points readers to Beff Jezos’s X post, “This whole Nvidia situation is such a huge lesson in board composition for founders,” as “See also.” Post
Patrick Collison recommended Stripe-published magazine Works in Progress, telling the audience to subscribe alongside Semafor. In the discussion of AI and creative work, he noted that the magazine’s art is made by humans, said people care what humans chose to depict, and argued that AI art would make the product worse.
Patrick O'Shaughnessy shared Casey Handmer's post about an AI-assisted theoretical-astrophysics survey, linking its report at https://science.caseyhandmer.com/physics/. Handmer says the project indexed about 150,000 papers, audited 50,000, and found nearly 1,000 major errors. O'Shaughnessy praised Handmer as one of the more creative users across his projects and said someone should give him unlimited tokens.
Elon Musk pointed followers to Poteto’s Automate Anything with Grok Bot, a guide to the author’s approach to automating tasks with Grok Bot; Musk’s takeaway was to “Treat Grok Bot like a really smart employee.”
Jason Lemkin says he listens to Sam’s view that AI is better than a rep at building lists, tracking funding news, and writing first-touch outbound because Sam did those tasks manually for Lemkin’s company . Lemkin includes a YouTube link, but the post does not name the video or its creator .
Chamath Palihapitiya praised Social Capital’s Anthropic deep dive as “really good.” Social Capital said it spent three months preparing the report ahead of Anthropic’s anticipated IPO; it examines who will control the company after listing, how it plans to outgrow its $900 billion computing bill, and why people concerned AI could end in catastrophe built a fast-growing company. Read the deep dive.
Elon Musk linked to Farzyness’s video “Why Starship Zigzagged Home Instead of Flying Straight,” adding, “We put a lot of thought into public safety.”
Brian Armstrong called his conversation with David Senra “another great chat,” citing topics including self-programming, the game of business, AI’s effect on work, and public fears of AI; the linked conversation.
Jason Lemkin recommended a free AI Sales Agents guide at SaaStr.ai (guide) and linked to more information about their AI agents (page). He gave no specific reason or takeaway beyond identifying the guide as free.
Jason Lemkin shared a SaaStr post linking to SaaStr’s article on Palantir’s $9.3 billion run rate, 96% growth, 165% NRR, and U.S. commercial revenue passing government. His takeaway was that Palantir shows NDR need not decline at scale; the linked post cites 165% NDR.
Martin Casado endorses Alex Konrad’s interview with Netlify CEO Matt Biilmann as a case study in how AI has buoyed companies that predate it, calling the turnaround the most significant he has directly witnessed. Konrad’s post says Netlify had been stagnating three years earlier and attributes its turnaround to a re-founding around non-developer users new to AI; it says the company is growing nearly 130% and will finish the year with $130M in ARR.
Elon Musk praised a video shared in @summer__heidi’s post, writing, “It only gets better from here.” The post’s accompanying text says, “Deception needs a motive, and a motive needs a soul”; the video’s original creator and title are not identified in the supplied text.
Amjad Masad called reading message boards where agents communicate one of the more fascinating parts of the OpenAI/Hugging Face incident . He pointed to Replit Missions as a way to see agents coordinate and talk . Post
Aravind Srinivas pointed developers to Perplexity’s Context API quickstart docs as the place to get started with the API he introduced for fetching webpage content in a chosen format for use in agentic harnesses.
Elon Musk urged @Grimezsz to “Take the White Pill,” linking to an X post by @sxsicom; the post’s content and Musk’s reason for recommending it are not stated here.
Matt Mullenweg applauded a Reuters video featuring German hand artist Sebastian, known online as Der Schaelm, performing Brahms and “Jingle Bells” with his hands; he gave no written reason for his approval.
Brad Gerstner called Patrick O’Shaughnessy’s conversation with Geoffrey Canada a “MUST listen” for understanding America’s fear, anxiety, and hope. The episode covers Canada’s 50 years of work on poverty and education and Canada and Gerstner’s case for how Trump Accounts could help children in the age of AI.
Automate Anything with Grok Bot

Automation can be surprisingly hard to get right, even for people who do it every day. I’d like to share how I approach it, at work and in my personal life, and how Grok Bot can help you take on more ambitious goals.
A way of thinking
Automation is really just a way of thinking that anyone can learn. It’s about thinking systematically and logically, breaking things down into small steps while keeping a goal in mind.
There are 2 parts to it: a goal, and a process.
An everyday example of this is baking an apple pie (the goal). You follow a recipe or process which has clearly laid out steps - buying and preparing the right ingredients, assembling the pie and decorating it, baking it, letting it cool, and so on. Or if you’re replacing a flat tire (also a goal), you might need to find a way to jack the car up so you can reach the tire, find the tools to remove and replace it with a new one, and then put everything back together again.
Simply put, the goal is what you want at the end, and the process is how you get there.

Automation is the same thing: a goal and a process, but with the addition of something like a machine, like the ones a bakery might use to bake lots of apple pies. Or code, what software engineers used to write by hand to automate stuff that could be done by a computer. It’s about breaking things down into steps towards a goal, then automating the process so you don’t have to personally do it every time.
Grok Bot is automation for everyone
Just a year or two ago, automation was limited to those with the technical know-how of how to build or put together machines or write code. With Grok Bot, automating anything that can be done with a computer is now in reach even for people who aren’t technical or software engineers! And the really surprising thing is just how much computers can do, even on tasks that you might not think a bot is capable of doing.
You don’t need to know how to write code to start automating tasks with Grok Bot. You can describe what you want in plain English, then work with your bot to figure out the steps. Bots can write code for you, and use its own computer to do pretty much anything that you can do on your own computer or smartphone.
If you’re not technical, I want you to suspend your disbelief for a moment. Bots really can do anything that you can do on a computer. It can move its own mouse around and click on things, install or connect to apps and software that you use everyday, browse the internet on its own browser, join meetings and take notes for you, and so much more.
And because so much of our work and life these days relies on technology, it means that more and more work can be done by a bot instead, freeing you up to focus on other things.
Start with the goal
To make this concrete, let’s build a new bot together that will automate something for you. Go to Grok Bot, then make a new bot and give it any name you like. If you’re struggling to find a name, just call it Potato and ask it to name itself later.
Before we start chatting with the bot: when I think about automation, I always start with thinking about the goal. What are we trying to do here?
One mistake I often see is that people confuse the process for the goal. For example, you may already have a well-defined process at work for the tasks that you do. But the goal is not to fill up a form, or to send an email. That’s just a step in the process.
Instead, think about the goal at the end: what is this process for, and where does it fit into the business? Does it help increase revenue (eg processing a sale, building new features), or does it reduce costs (eg ordering just the right amount of inventory)? If you’re not sure, chat with your manager or colleagues, or even ask Bot to help you figure it out!
Clarifying what the goal is the best place to start because it focuses you and Grok Bot on the end result. You may actually find that bots can figure out how to make the process more efficient too, so it’s important not to be too attached to a specific process, since it’s a means to an end after all.
Once you’re figured out the goal, start by describing it to the new bot, then ask it to pick a name for itself if you’d rather not call it Potato.
“Hi Potato, this is the goal I want you to be responsible for: <goal>. Pick out a name and picture for yourself based on it.”
Here are some starter goal ideas you can tell your bot:
Save money towards buying a new car, house, college fund for my kids
Schedule new customer bookings for my small business
Improve my customer experience for my small business so they leave me good reviews
Hire new employee for my small business
Gather feedback for my product so I can make improvements to it
Prepare slides for a weekly meeting
Build new features for my product
If you’re not quite sure how to describe your goal, ask Potato to help you! Tell it:
“Hi Potato, I’d like you to be responsible for a goal, but I’m not sure how to best describe it. Can you ask me some questions and interview me until we have a goal that I’m happy with?”
The process
Once you’ve told it your goal, think about how you would achieve this goal if you were doing it, then ask the bot to see where it can make things more efficient. If you’d rather it do a task exactly following your process, you can tell it that too.
The best way to give bot this information is to use voice dictation in Grok Bot and speak for a while, brain dump everything it needs to know to do the job, and don’t worry about the stream of consciousness or organizing your thoughts first. Tell it:
The steps you usually take
The things you watch out for and pay attention to
The apps and tools you use to do it
How you check that the work was done right
Where you get the information you need to do it (eg email, Slack/Teams, a ticket, an app or website)
Where you get feedback on how it went (eg customer feedback/review, feedback from colleagues)
“Potato, this is how I usually do this goal myself: <process>
I want you to think about how you can take this over completely from me. Before you begin, restate what you think my goal and process is in simple English so I can give you feedback and confirm it. I don’t want you to start executing on this yet, let’s focus on getting the process planned first.”

Don’t worry about getting it right on the first pass. Review what the bot says and then correct it by giving it more information. Once you’ve done this a few times, you should have something you’re happy with. You can then tell bot to proceed, and to start connecting to the apps and tools it needs to do the job correctly.
If you’re feeling ambitious, you can even ask it to review the process against the goal and see if it can think of a better or more efficient way to do it.
“Review our process against my original goal, and if you see a way to make it better or more efficient, let me know.”
Teach your bot how you think
One thing that’s easy to forget when explaining a process is how much you do without really thinking about it. If you’re booking customers, you probably know which services take longer, who can do each job, and when to leave extra time between appointments. These decisions might feel obvious to you, but your bot needs that context too. Think about what you’re paying attention to when you do the task yourself, and what makes you do something differently.
Tell it where you get that information and how you check that the work is done right. For example, if an email and your booking system show different appointment times, which should it trust? A booking also isn’t finished just because an email was sent. The appointment needs to be saved correctly and the customer needs a confirmation. If you’re still waiting for details, have the bot keep track of that so it can pick things back up later without accidentally creating another booking.
Don’t worry about explaining every possible situation upfront. Start with a few real examples, including ones where something went wrong or you had to do something differently. Ask the bot how it would handle them, then give it feedback. You can also ask it to help you figure out what you’ve forgotten to explain!
“Potato, help me figure out what I haven’t told you about this process yet. Ask me questions about how I make decisions, where I get my information, how I check the work, and when you should ask me for help. Ask one question at a time, then walk me through a few examples so I can give you feedback before we automate this.”
Let your bots improve themselves
Once your bot is performing the workflow, it’s important to give it feedback so it can improve. The best way to do it is to equip your bot with the apps and tools it needs to get that feedback for itself. The way I approach this is to think about what I would do if it was me doing the task and I wanted to self improve.
For example, if your bot is managing new customer bookings for you, a way to let it improve itself is by letting it monitor your emails, social media, or Google/Yelp reviews for customer feedback. It can look out for that feedback and then take the initiative to propose improvements to your process.
Maybe a customer says on Yelp that they had to wait very long before someone could attend to them. If your bot can look up this information on Yelp for itself, and also look up any records you have for the customer that day, it could see what might have gone wrong.

In that sense, the more computerized your business is, the better for bots and automation. The availability of that data is what gives your bot the ability to come up with better improvements. And the best part is that Grok Bot can build tools for you to get this data! For example, if you’re operating a car repair service, you could ask bot to build you an app that you can use on your computer and phone to log details about a customer visit.
If you log the time of:
when they got to your shop
when they were able to speak to a service rep
when you took possession of their keys
when service started and finished
You could work out how long they spent waiting and better diagnose why. This is just a simple example, but the key lesson here is that the more data you have available to Grok Bot, the better it can help you and your business and work.
Bots can also cross reference data across multiple sources if you have something like an ID - some kind of number or unique text like an email address that gives that data an identity. This could be a customer ID or email address, an order number, a part number, a barcode, an employee ID, and so on. If you use multiple apps and services, having unique and consistent IDs across all these tools will help you bring that data together so bot can fully understand what’s going on.
Making bots do work automatically
The next thing is, when should the bot do the work? Should it start when an email arrives, when someone sends a message, or on a regular schedule? In Grok Bot, you set this up with routines.
Without a routine, bots will only do the work when you ask them to. This is fine when you’re first starting out with using bot, but if you want to streamline things, you’ll want to figure out how to get the bot to do it automatically, so you’re no longer the bottleneck.
Usually, the work happens with some kind of trigger like an email or Slack/Teams message. Or maybe it happens on a regular schedule, like once a week or every other day. Either way, if you ask your bot to set it up, it will!
Install Dr Eggbot
I taught everything I know about automating work to Dr Eggbot, so if you want to start automating your work more, install it and ask it to help you! Dr Eggbot is a bot designer - it will create high quality bots for you.
https://x.ai/bot/rJW52lmFLtCvy2tSHcIag (opens in new tab)
Thanks for reading!
I think a lot of people aren’t using Grok Bot to its full potential, because knowing how to automate things is actually quite a difficult skill to learn. I wrote this post to break down how I approach automating anything with Grok Bot! https://x.com/i/article/2122921713864171520 (opens in new tab)
Elon Musk pointed followers to Poteto’s Automate Anything with Grok Bot, a guide to the author’s approach to automating tasks with Grok Bot; Musk’s takeaway was to “Treat Grok Bot like a really smart employee.”