If data center permits stall after the midterms, frontier models may arrive later and usage limits may tighten, which makes mastering today's AI workflows more valuable, not less.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You run parts of your business or job on Claude Code, Codex or similar agent platforms and want to know whether your subscription value could change.
You follow frontier model releases and want a plain explanation of how data centers, neoclouds and permits connect to model timelines.
You use AI agents on your phone and want a quick tour of this week's Grokbot, Dot and Claude Duet changes.
SKIP IF…
You want a neutral policy analysis; the data center section is one builder's opinion with a single Goldman Sachs figure behind it.
You need a deep tutorial on any one agent platform; each update gets about a minute or two.
TL;DR
The full version, fast.
The host argues Democratic wins in the midterms, with several candidates promising data center moratoriums or pauses, could put a large share of planned compute at risk and delay models like Fable 6 and GPT-7. He walks the data center stack from users down to permits, explains why neoclouds are most exposed, and predicts tighter usage limits and pricier compute. His advice is to get more out of today's models, since a slower frontier raises the value of well-built workflows. The rest covers Grokbot phone calls and pages, Dot's reply and share features, and Claude Duet's early control of local Claude Code.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
States the claim that the midterms could slow the AI boom and previews the week's updates.
00:31 – 02:48
02 · Midterms and the data center pause
Democrats take control of Congress and candidates in Texas, Ohio and Georgia promise data center moratoriums; Opus 4.1 to 5.5 shows how fast things were moving.
02:48 – 06:25
03 · The data center stack
Walks from you down through inference, models, clouds, neoclouds, networking, chips and permits, and why neoclouds are most exposed.
06:25 – 08:28
04 · What a slowdown means for you
Subsidized plans and usage limits could tighten, but a slower frontier makes learning today's models more valuable.
08:29 – 12:55
05 · Grokbot updates
Native phone calls, shopping cards with X Pay, saved cards, pages, and new usage limits.
12:55 – 15:37
06 · Dot app updates
Reply by holding the icon, share to Dot from any app, action button setup, faster notifications and Gmail deep links.
15:37 – 16:28
07 · Dot speed and OpenAI hiring
Calls to Dot pick up faster, you can hang up by saying goodbye, and OpenAI hires Kat Wu from Anthropic.
16:29 – 18:21
08 · Claude Duet controls local Claude Code
Duet can list and start sessions on the host's computer, though the approval flow is still clunky.
Atomic Insights
Lines worth screenshotting.
Ten months took Opus from 4.1, which could build Flappy Bird, to Opus 5.5, which built a 3D multiplayer shooter in one prompt and works unsupervised for days.
Goldman Sachs warned the Democratic wins could put nearly half of planned data center expansion at risk.
Big clouds fund data centers from profits, while neoclouds like CoreWeave fund them with debt, so permit delays hit neoclouds first.
A data center needs power, land and permits before it needs chips, so the policy layer sits underneath the whole AI stack.
A $100 or $200 agent plan can deliver roughly $4,000 of tokens a month, and that subsidy is the first thing tighter compute could end.
If frontier models pause for six months, workflows built on today's models become more valuable instead of obsolete.
The belief that you can skip building workflows because a smarter model will figure it out later only holds while models keep improving fast.
Even a bubble burst would not make AI go away, because the models and tools already exist.
Grokbot can now place phone calls natively, replacing external call APIs like Bland for simple tasks such as reservations or cancellations.
Grokbot added usage limits because, per the host, its serving cost was unsustainably high relative to revenue.
Holding down the Dot icon on an iPhone lets you reply to the agent without opening the app.
A single remote agent platform that controls Claude Code on your own computer is the feature the host wants most, and Claude Duet now shows early glimpses of it.
Takeaway
Why a compute slowdown raises the value of workflows
WHAT TO LEARN
AI progress depends on physical compute that policy can slow, and if it does, the people who already build good workflows on today's models gain the most.
02Midterms and the data center pause
Candidates in Texas, Ohio and Georgia ran on pausing or conditioning data center approvals, which directly targets the buildout AI labs rely on.
Ten months took Opus from simple Flappy Bird builds to days-long unsupervised work, so even a modest delay changes what arrives next.
03The data center stack
Model progress rests on a stack from permits and power up to chips and clouds, so a policy pause at the bottom can delay releases at the top.
Neoclouds fund compute with debt while big clouds use profits, so a permit slowdown puts neoclouds at much higher risk of failure.
Landlords and neoclouds with unusable land or chips are the first to fail, which then discourages further investment up the stack.
04What a slowdown means for you
Agent subscriptions are heavily subsidized, so tighter compute could mean stricter usage limits and higher prices in the short to medium term.
If frontier models pause, setting up workflows on current models pays off longer, because they will not be replaced in a few weeks.
Even a bubble burst would not remove AI, so waiting for a smarter model before building workflows is the riskier bet.
05Grokbot updates
Native phone calls let an agent handle reservations or cancellations without a separate calling API.
Shopping cards, X Pay and saved cards move the agent from chat into real purchases.
Grokbot's new usage limits show how serving costs force even fast-moving platforms to ration heavy automation.
06Dot app updates
Replying from the home screen icon and sharing links into the agent cut the friction of delegating small tasks.
Setting the iPhone action button to open the agent makes voice delegation a one-press habit.
08Claude Duet controls local Claude Code
Remote control of local coding sessions from a mobile agent app is the next unlock, but approval prompts still make it slower than a direct local session.
A single agent platform that can start branches and sessions on your own computer is the stated goal, and early versions already work.
Glossary
Terms worth knowing.
Inference
The compute used to run a trained model each time you send it a request, as opposed to the compute used to train it.
Neocloud
A newer cloud company, such as CoreWeave, that borrows money to build GPU capacity and rents it to AI labs by the hour.
Data center moratorium
A government ban or pause on approving new data center construction.
Grokbot
An agent app from xAI and SpaceX that can search, shop, build pages and now place phone calls on your behalf.
Dot
OpenAI's personal agent app that the host talks to by voice, text and iOS share sheet.
Claude Duet
Anthropic's agent platform that can send tasks to Claude Code sessions running on your own computer.
Quotables
Lines you could clip.
07:42
“As AI slows, learning how to effectively set up workflows with AI, the value actually increases.”
A single counterintuitive claim with no setup needed.→ TikTok hook↗ Tweet quote
00:01
“This week's midterms could put the brakes on the AI boom.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
I hate to say it, but this week's midterms could put the brakes on the AI boom. Democrats just swept back in control of Congress. And that could make it much harder to build the data centers these AI companies are counting on.
Which means we may not see Fable 6 or GPT -7 as soon as we thought. And this has huge implications for using AI in your job, your business, and your everyday life. And today we're going to break...
it all down and what all of this means but we have plenty of new updates to cover as well Like the fact that Anthropix new Duet platform now has better control over Claude code. OpenAI has some huge updates to their iOS app and regarding Grokbot, we have some good news and some bad news.
So the first update that we're going to talk about today involves the outcome of the midterms and how it could drastically slow down AI and slow down the acceleration that we've been seeing over the past three years. Long story short, we are shifting to the left. Democrats are now in control of both the House and the Senate.
And these three states right here, Texas, Ohio, and Georgia, all ran on slowing down data centers. Gina Hinojosa. Said, when I'm governor, I will issue a moratorium, a ban on all data center approvals.
Ohio. Amy Acton said, Ohio is open for business, but we are not up for sale. If you do not meet those conditions, we do not want you in Ohio.
Keisha Lance Bottoms of Georgia. My position is that we need to pause on new construction of data centers. So we may be ending the era of acceleration.
And just to give you a perspective on. on how fast we're accelerating. There was a 10 month period from November of 2025, we had Opus 4 .1.
Opus 4 .1, my benchmark for how good this model was is like, in how many prompts can I create Flappy Bird? It's a very simple game. Now, 10 months later, we got Opus 5 .5.
This is a 10 month period. I created a full 3D multiplayer. Call of Duty game in a single prompt.
And Opus 4 .1 could only work for like 20 to 30 minutes effectively, unsupervised, where Opus 5 .5 can work for days. So 10 months, we had this rapid acceleration. And so what I'm concerned about now is it could take much longer before we get something like Fable 6 or GPT -7.
The amount of compute and data centers required to get these new models is pretty drastic. So today, Goldman Sachs warned that the Democratic wins could put nearly 50 % of data center expansion at risk. And so this has drastic effects over the economy.
And so what I want to cover right now are the main players involved and the types of players that exist that could lead to a massive slowdown in the world of AI. And there's different parties or different types of people that sit at different levels of the data center stack. And guess who's at the top?
It's you. You are running. apps like Codex, like Cloud Code.
You have your subscriptions and you're using AI in your business. At least you should be. Anyone who is sensible is doing so.
But in order to run your agent platforms like Cloud Code and Codex, you need to run the models, which is called inference. And then you need to train the models like GBD 6 .1 Sol and Cloud Opus 5 .5. It requires compute for the models and then running the models, which is called inference.
Now, here's where everything starts to get a little bit dicey. Clouds and Neo clouds. And the main distinction between the two is like how they are funding the compute.
So clouds like Microsoft, Amazon and Google have these massive data centers. And the way they've created these data centers is they're profitable companies, right? Microsoft is a very profitable company.
Google is a very profitable company. So is Amazon. And they are able to take these profits and build out these data centers.
Yes, they do some other funding, but at large, they are funding these massive data centers through profits. And they are less at risk because they have profits in other parts of their business. NeoClouds are a brand new type of company.
The main one is CoreWeave. I'm pretty sure that's the biggest NeoCloud. And what they're doing is they're raising money, mostly through debt.
to build out compute and then they rent it to OpenAI and Anthropic, right? They make money by renting compute to OpenAI and Anthropic, which is needed to train models, do inference, and then a bunch of other things involved in the AI stack. And so that's what's getting really dicey because, you know, NeoClouds can't just...
You don't just buy compute and then turn it on. There's a lot of things that you need beneath it. And the things you need beneath it are the networking.
So you need to like link thousands of these chips together, right? That's where you get all these data centers. And then you also need chips.
And so that's when like NVIDIA GPUs or AMD or Google's TPUs come in. And beneath the chips is you have the building and cooling. So this is the actual data center itself.
But you can't just like build a data center anywhere. Right. You need to make sure that the building has adequate power and you need the right land.
And part of the land is does it have the correct permits in order to actually build a data center itself? And so that's where these new officials come in. New officials set the rules for power permits and taxes.
And this is kind of where. a pause would start. And if you do not have the correct permits and then the rules get more confusing, this stops the rest of the stack from actually investing.
The landlords, right? The landlords, the people who are basically renting out the space, basically have a lot more vacancies. So a lot of those companies could go under because no one's renting out the space because they don't know if they're actually allowed to build the data center there.
And then neoclouds are the ones that are most in trouble because they've already borrowed money and they. bought GPUs from NVIDIA and they might not be able to actually build in the property that they've already purchased or agreed to buy. And this could easily cause many of them to go under.
And so let's imagine for a second that policy prevents the investment in more compute. I've come up with kind of six ways that this could actually deeply affect your business. The first one is that subsidies could end or slow down.
So today's plans, like if you have Claude Code and Codex, you get tons of credits. You basically get to use like $4 ,000 worth of tokens in a month if you pay for the $100 per month plan or the $200 per month plan. It feels pretty unlimited in terms of the amount of Opus I can use.
And we could get even tighter usage limits on the agent platforms, right? as we'll get to in just a second, added some usage limits last week that people are upset about. These could get even worse as compute constricts, right?
If compute constricts, that means there's not enough compute to do inference and it will just become more expensive to get access to compute. So at least in the short to medium term, we could see more usage limits. If there's one takeaway is that if everything slows down, Right now, and we don't see Fable 6 or GPT -7 as soon as we thought, learning how to use AI will get more valuable.
There's a certain sense. right now in the business world that ai is getting so much better so fast that you don't even need to learn how to do these workflows because by the time you get the workflow set up ai's already progressed past it and then now at that point you can just ask claude fable 6 to just do it and it will figure it out but if we go through this slow period where there's six months without any advancements to the frontier of ai models then it will become even more valuable to use the current models of the day to create workflows and so i i think that if the world slows down and everything.
And it feels like AI won't progress. That just means the value of like getting the most out of the current AI models will go up. So you should a hundred percent focus on that, right?
Don't think for a second that if the bubble bursts, AI is going to go away. That's not in the realm of like possibility, unless something crazy happens where like all the data centers are burned down. So I wouldn't main takeaway of this section is as AI slows.
learning how to effectively set up workflows with AI, the value actually increases. And so speaking of the value of learning how to use AI effectively increasing, let's talk about some other updates from Grokbot, DOTS, and then also from Anthropic. And then a few other ones we'll get to in a little bit.
We're going to go through these quickly. The first one is a brand new feature inside Grokbot, which is call with Grokbot. So SpaceX just announced that you can now have Grokbot make calls for you.
Hi, Dan. I'm calling to cancel a customer's internet plan. Okay.
How about 10 % off for a year if you stay?
I appreciate the offer, but they got a much better deal with Starlink. great you're all set then have a good one so before you would need to sign up for something like bland which is an external api that allows your agents to make calls well now it is just baked into crockbot You can ask it to call and make a dinner reservation.
You can have it cancel plans or you could have it do cold calling, I'm assuming. And so that's pretty insane. And in order to get this, you actually need to update your Grokbot app.
So I'm going to go ahead and do that now. Hey, Grokbot, is the calling feature set up? Am I able to make calls now?
Can you look up my favorite restaurant, Fairfax, in... West Village and try and give them a call and see if there is any availability for 9 p .m. on Saturday.
Looking up Fairfax in the West Village, then I'll queue a call about Saturday at 9 p .m. Okay, it found it on Google Maps. How many people?
Okay, it's asking me. Let's say three. Okay, so this is interesting.
Now it says Grokbot wants to make phone calls for you. I accept.
Call Wow, it's calling call card is up tap call So we got no answer but it can make phone calls and so you can see it right here You can approve the call and I think it went straight to voicemail. I'll probably make a dedicated video on this Testing it and coming up with some better use cases, but grok bot can now make phone calls Okay, and now let's move on to some other updates that they released in grok bot I will admit they are releasing so many new features really fast Like I would say like one or two decent updates every single day are coming out to grok bot.
Let's go through them real quick So the first update they added are these little cards? So if you ask for the fact that you want to like buy some flip -flops it'll like render these UI cards and you can click on them directly and You can actually buy them and they're trying to make it really easy to purchase.
And Grokbot is actually hooked up to XPay. So on Twitter, I have a decent size, like 300 ,000 or 250 ,000 followers on Twitter. I make money every month from posting on X and I'm able to just spend it directly inside Grokbot because it's hooked up to XPay because that's how they pay you.
Right. If you go to Twitter. Wait, I don't have it on my phone, but they're making it really easy to pay directly on the platform.
And last week, they also released a their own version of one password kind of where you can actually upload cards that can be saved. So you can if you have a credit card, you can upload it in the app and your agent can use it securely. And then finally, they've made a ton of updates to their pages feature.
So if you think of Claude artifacts or GPT sites. The version for Grokbot is pages. So I can say, please add these to a pages doc and make it like a mood board.
And now it's saying making a flip flop mood board page now. And then here it shows up as this little card. You can either click on it here or you can click on it here.
And this is like a page. And I'm going to say, please add more detail to these, maybe generate it and then make a table comparing them. And now, as of a few days ago, Grokbot can make multiple changes at the same time.
And here you go. It made a ton of updates and I highly recommend trying out pages. And then finally, one thing to be wary of, there are more aggressive usage limits when you use Grokbot.
And yeah, they added usage limits. Some people were complaining about this because they have a ton of automation set up and now they're hitting their limits much sooner. And Elon Musk.
stated publicly that their serving cost for running the Grok bot was unsustainably high relative to revenue and they had to put on some usage limits and some people were upset about this but just be wary of it. Okay so now we're going to go through the updates to DOTS which is OpenAI's agent platform. And I'll talk about in just a second why I think this platform is going to get exponentially better over the next month or two.
So the first update to Dots is if you hold down Dots on your home screen, you can reply directly to it. Here I just got a notification. Film the first mini chat video.
I can just hold it down and I can very easily just say, hey, can you please add this to my calendar tomorrow at 11 a .m.?
And you see that without even opening my phone, I was able to hold it down and respond to my agent. And look, all you do is hold it down. So I'm just going to hold down my screen here.
And now I can respond directly like this. And it just said, I'll put the first many chat video on your calendar for tomorrow at 11 a .m. Hold it down.
Thanks. You can respond. Very cool.
The next thing that you can do is you can actually share things directly with your dot So if I press this little share button right here click share and I can hit share post via Look at this your dot shows up so I can share this with my dot You can do this with any type of link any image any video? Super easy to share things and I can just send it to my dot and this is really nice because you might just want to just tell your dot to remind you of something later.
Or I could say, hey, can you please use this in a video as B -roll? A lot of cool use cases for this. Another thing that you can do with your dot is you can set it up so that your action button, which is the button right here on your phone, when I press this button, it takes me to my dot.
This is super useful because I'm constantly using the voice chat with my dot. I call my dot all the time. I highly recommend setting up the action button.
If you want to know how to do this, just go to your dot and say, how do I. set the action button to open up my dot. Super easy to do it.
If you want to stay up to date on all the updates regarding dot, follow this guy, underscore Samir ISM. He is going to be releasing all of these. Some other things that he mentioned is that there's faster loading when you tap from a notification and then also Gmail links, deep link to the exact email inside the Gmail app.
So if you were to ask your agent, like, Hey, any interesting emails, it may send you a hyperlink. If you click on that hyperlink, it'll take you to the exact. Gmail email, which is really useful.
Additionally, it's going to get significantly faster to call your dot. Right now it rings. When they first released it, it took like 20 seconds for the agent to pick up.
Now it's five seconds and soon it will be immediately. Also, if you're calling your dot, you can hang up now just by saying like goodbye to your dot and it will just end the call, which is really nice, especially if you use them with AirPods like I do. And then finally, OpenAI is actually getting a ton of new talent.
They just got Kat Wu. He and Boris were one of the original people to work on Claude Code at Anthropic. She worked on Claude Cowork.
She's leaving Anthropic to join OpenAI and work on their core products. And this is one of the main reasons why I think their agent products are going to get so much better. Okay, the final update of the day comes from Claude Duet, which is Anthropic's agent platform.
And they are starting to show glimpses of the feature I want the most. I want to be able to control Claude Code running on my computer or Riley's MacBook Pro from Claude Duet. So like I said, there are some glimpses of it.
So I can say, hey, what sessions can... you work on in cloud code on my computer? And it says just one.
Can you please make a new branch that adds light mode called light mode? A new branch, don't push. And so what I want to do is I want to have one single AI agent platform that can control cloud code on my computer.
That's like my dream. We're starting to see it.
See, this is the annoying part of it, though. It's like making me do these pass keys. And this is not what I want.
So I just had to sign in. And so now it says sent to app V2 development, make a light mode branch. And so we're just on the web app.
You can do this from the phone or the web app. But if we open up Claude, you can see that this was sent to my local Claude code session. So this is one of the main reasons I still use Claude projects because it allows you to very easily spin up local sessions.
You don't have that whole approval thing. So right now we're starting to see glimpses of it getting easier to control Claude code from Claude duet, which is, I think is a big deal. It's just still not quite easy enough.
to spin up local sessions but i'll keep you guys updated on all the updates with claw duet i'm starting to like it a little bit more all the agent platforms are great right now and i'll be covering them in great detail so i hope you guys learned a lot from this video and i will see you here in the next one
The Hook
The bait, then the rug-pull.
The title promises that the next frontier models may never arrive, and the opening line ties it to the midterms. The first seven minutes make that case through the data center stack, then the video turns into a fast tour of this week's agent updates.
Frameworks
Named ideas worth stealing.
02:52model
The data center stack
You
Inference
Models
Clouds and neoclouds
Networking
Chips
Building and cooling
Power, land and permits
Each layer depends on the one below it, so a permit pause at the bottom stops investment all the way up.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
Seven attention problems from social media and AI agents, and the plain fixes one builder found over 60 days: an agent-run notification filter, custom feeds, paper, walks, slop filters, and a daily journal.
A creator with 2 million followers maps his whole agent stack onto one grid: Claude for making things, Codex for running the business, Composio as the shared toolbelt, and GrokBot as the Twitter specialist.
A 20-item weekly rundown of AI agent news: Claude Dashboards and Motion, GrokBot routing to Opus 5.5, ChatGPT's inline mini-apps, Nano Banana 2.1, and Adobe's open-source clones.
Riley Brown spends 24 minutes inside the new Claude Projects: one orchestrator chat spinning up parallel threads that share memory, artifacts, and routines, plus token costs, the new in-app design mode, and the phone workflow.
One structured prompt into Claude Code's new Fable 5.1 turns into a Trello clone that the creator, his team, and three different AI agents can all update on the same real-time board.
A week where xAI undercuts Anthropic and OpenAI on price, and GrokBot's named-agent structure hints at how every AI platform is about to fix the chat-versus-work problem.
00:20
00:31
00:43
01:15
01:29
01:43
01:52
02:13
02:17
02:38
02:52
03:05
03:19
03:33
03:47
04:00
04:14
04:42
04:55
05:09
05:23
05:37
05:50
06:04
06:18
06:32
06:59
07:13
07:27
07:41
07:54
08:08
08:22
08:37
08:49
09:20
09:31
09:44
09:58
10:12
10:26
10:34
10:49
11:07
11:21
11:35
11:45
12:02
12:16
12:30
12:43
12:57
13:21
13:38
13:52
14:06
14:20
14:33
14:47
15:01
15:12
15:28
15:56
16:10
16:22
16:37
16:51
17:19
17:32
17:46
18:00
18:14
24:13
Video of the Day31:28
20:50
24:06
25:02
25:21