just got banned. Claude Fable 5 and Claude Mythos 5 both pulled not by a glitch, but by the United States government. And if that sounds wild, it is
because it happened in the same seven days that China dropped two new AI models that started beating Claude at its own test. The same week a new trick
showed up that lets cheaper AI models work together and match the best models on the planet. And on
the same week that three more tools quietly got upgrades that change how you do search and run tasks whilst you sleep. This is the my AI news roundup.
One week, eight big stories and every single one of them matters far more for your business. Whether you run an agency, whether you sell products
online, whether you're a creator trying to get more done in less time. So let's start with the one that
nobody saw coming, which is the Claude ban. On Friday evening, the US government sent Anthropic a letter. Anthropic makes Claude. The letter told them
to block all access to their most powerful models, Fable 5 and Mythos 5, for any foreign national. That means anyone who isn't a US citizen, whether
they're inside the United States or outside it, even Anthropic's own employees who aren't citizens.
Now, here's the problem. How do you let only US citizens use a model and block everyone else? You can't really, it's too hard to check. So, what did
they do? Anthropic said the only way to follow the order was to shut both models off completely for everyone. The letter actually came from the
Commerce Department citing national security. Anthropic got it at 5:21 p.m. Eastern time. And the letter
didn't even see exactly what the concern was. Anthropic push back hard. They said they're following the order, but they don't agree with it. They
argued that pulling a model used by hundreds of millions of people sets a very scary standard. If that rule got applied to everyone, they said it
would basically freeze all new model releases across the whole industry. So, think about what that means.
The models are now so capable that a government stepped in to control who gets to use them. That has never happened before. Not like this. And here's
what it means for you. The tools you build your business around can change overnight. One day, Fable 5 is the most powerful thing you ever used. The
next day, it's gone and you didn't do anything wrong. You just wake up and it wasn't there. So, the
question becomes, what do you do when the tools you depend on disappears? That's exactly the trap that most people fall into. they pick one AI tool,
they learn from it, they build everything around it, then it gets pulled or worse, something better comes out. Um, or you know, they're just totally
stuck. They're starting over. And that's the whole reason I keep telling people not to just marry one
tool inside the air profitable world. And we actually built something called the agent operating system. So you want to think of it of like a
dashboard in one place. You can build this yourself or you can get my setup inside there. And it's got Claude, Hermes, OpenClaw, all of them into one
place with shared memory. So if a model like Fable 5 gets pulled, which you just did, you don't panic. You
just switch to another model and have the same system that keeps working. So Fable 5 gets pulled, which leads to the next obvious worry. If the best
clawed model is gone, what do you even use right now? And it's funny enough time in that because the answer actually showed up the same week from
China. The first one is GM 5.2. Came out in June the 13th from a company called Jeru, also known as Z AI,
and it's a coding and building model. It's got a million token context window which means it can hold a massive amount of information in its head of
once without forgetting. Now I actually tested GLM 5.2 against Opus 4.8 which is the claude model that's still up. Same prompt, same test and honestly
actually surprised me. So we've got Opus 4.8 over here. We've got GLM 5.2 over here. We asked all of
them to build like a simple game where you know a character runs through a city dodging blocks and GLM 5.1 5.2 sorry actually made a better game.
Right. This is way more fun to play with than what you'll see from Claude and Opus 4.8, which is like super basic and boring, right? So, it's more
fun. It's more polished. Opus 4.8's version actually felt very slow and basic. Now, we did another test
here where we tried to create a landing page, right? A launch page, you know, the kind you'd see from a big tech company. GLM 5.2 actually nailed it.
If we take a look at this, so clean design, working menu, the whole thing. Opus 4.8 8 is actually a lot thinner and more boring. It's planer right
now. Out of about five tests like that, GLM 5.2 actually won most of them, which is why because GLM 5.2
is a cheaper plan. You can plug it into your AI agents and it's beating the model that most people thought was untouchable. Now, one honest note,
there's no official benchmarks, so I can't tell you if it's proven better on paper. I can only tell you what I built with it and what I saw on my own
screen, as you can see right now. And if we have a look at this example, so this was a neon arcade game
like you can see and this one is super basic from Opus 4.8. If we check out the results from GM 5.2, you can see that this game is maybe 10 times more
fun, right? Looks crazy, interesting, fun to play with. Uh, a lot better design. And so that's why these tests tell you things that you should really
follow instead of like just trusting the hype, right? especially if there's no benchmarks out. You
might as well just test it yourself side by side or watch my tests. And here's the part that actually matters for you, right? GM 5.2 can work inside
agent tools like for example claude code and open claw can drop right in. So for example, we've already plugged it in to Hermes agent and we've built
out teams of agents. Plus we've got the GLM 5.2 CLI over here and it's built some amazing things. So
you can use it to power AI agents that run tasks for you. And that's really the difference between, you know, a chatbot that you talk to and an agent
that actually goes off and does the work. Quick example, we got a team of agents with GM 5.2 to create and edit a video as an AI avatar of me. And the
crazy thing about this is that we actually had a judge on the team who judged if it was good or
not. And if it wasn't good, they just kept sending it back and iterating on it. So you can see the video that's fully edited and created. I didn't
touch it. I didn't edit it. Um but you know that replaces a whole team of you know video agents or video marketers. Now the second model from China is
Kimmy K217. It came out from Moonshot just one day before GLM 5.2 and Kimmy is what they call a long
horizon agentic model. So don't let the SC it just means that it can keep working on a task for hours without you babysitting it. So you give it a
goal. It goes off chips away at it and you can see like all the stuff that I've built with it inside this video that I actually created as well. And
so all of this stuff worked. Not like perfect on the first try, but they actually worked. And also had a
team of AI agents use Kimmy to fully edit a video to script it, add an AI avatar. So this is the one from Kimmy and this is the one from GM 5.2. And
they were both pretty good to be fair. So you can have teams of agents just working together with this stuff. Now here's the smart part about how, for
example, models like GLM 5.2 and Kim K2.7 work together. So you can have multiple agent profiles
like you see. So, we've got multiple for GLM there. And then there's a judge agent. So, after the other agents build something, the judge looks at it
and decides if it's good enough. If it's not, it sends it back to get fixed over and over until it eventually passes. So, the quality control happens
by itself. So, this avoids any sort of problems with quality and also keeps your agents on track.
So, now you've got two solid options that fill the gap that the day Fable 5 left. And both of them slot right into an agent operating system, which is
exactly why having them all in one place is super handy. Like you can see now, what if I told you there's a way to get top tier results like fable
five level intelligence without picking a single model at all? That's the fusion story this week, and
it's one of the most interesting things I've seen in a while. So, here's the idea. Instead of using one model, you use a panel. Several models work on
the same question at the same time. Then a judge model reads every answer, pulls out where they agree, where they disagree, and what each one caught
that the others missed. And then it gives you one final answer. And three brains instead of one. And
when good brains work together, you usually get a better answer than any single one alone. Now, the wild part here is that you don't need top tier
models for this to work. in their tests, a panel of cheaper budget models working together landed within about 1% of Fable 5 on the intelligent tests
within 1% of the model the government just pulled using weaker parts. So they found that most of the
boost about 3/4 of it comes from the judge synthesizing the answers together. The rest comes from having different models in the mix. Now, fair
warning here. It is a little bit slower, mainly because the judge has to read everything and fuse it together. And because it runs several models at
once on the same question, it works and walks through more behind the scenes than a single model would. So,
it's not like a shortcut to doing less. It's a way to get a sharper answer when the answer really matters. Now, some people see the word benchmarks
and roll their eyes. I get that. So, don't take them on faith. Just try it yourself on a real test. See what it comes back. And here's a real way you
could use it. So, I set up a fusion panel, act as an SEO content team, like search what currently
ranks for a keyword, find what competitors missing, outline better article. The panel went off, argued out, and the judge handed back one clean plan
with a fully built landing page. And you can also build tools with it as you can see where you just plug in your API and then just ask the SEO content
council like you can see here. So, if you picture this for your business, you got one question,
several AI minds, one clean answer. You know, for example, if you're an e-commerce owner, well, you could ask a panel to write product descriptions
and get a sharper result than any single tool would give you. And you see how all of this connects. So, you got GLM 5.2, Kim, K2.7, Fusion Panels.
They're all the things you'd want plugged into one system so you can switch and combine them without
copying and pasting between five different apps. And that's really the headache that an agent operating system will help with. Now, let's also talk
about getting your time back. So Hermes actually released something called automation blueprints this week as well. And here's what it does. So it
turns a schedule task into like a fill-in-the-blank workflow that you can set up in basically one click.
Now before this, setting up an automation inside Hermes meant like fiddling with triggers and figuring out which connections you needed and sometimes
it would just not work. Now there are readymade templates. You can copy one, paste into Hermes, and it's good to go. The one that I actually set up
this week was a research agent using the blueprints from Hermes directly. So every week you can go
out, find the latest news, write it in a clean report and save it into my memory system automatically with Obsidian. So instead of spending like an
hour every morning digging for news, I can set it once and now it just runs. I call this the Goldie Hermes automation engine because that's a shift
like 30 minutes to set up this time and then you've got a research agent working for you forever. And
there are blueprints for all kinds of things, right? For example, watching your competitors, tracking mentions, monitoring rankings. An agency owner
could set up a blueprint to watch a client's competitors and get a fresh report every Monday without lifting a finger. And the best part is you don't
have to be technical. You don't write any code. You tell Hermes what you want in plain English, and
it builds the job. Now, next up, we have Notebook M. And this is Google's research tool, and it just got a big upgrade. So, Google gave it for aentic
powers running on their Gemini 3.5 model. Now, what does that mean for you in plain terms? Well, you can now type in a loose question, no sources, no
setup, just the thing you want to know. And notebook will actually go out and hunt for the best
sources on the web for you and pull them in. And not like just the first 10 links, like the strongest ones. And then it can turn all that research
into all kinds of things. So, for example, we turned it into a full agent architect blueprint inside a beautiful PDF. We've created videos, podcasts,
slide decks, infographics, charts, full PDF reports with real data and sources cited. Now, every fat
comes with where it actually came from originally. So, it can even write and run code now. And it can build spreadsheets and charts to show the
results. I actually built a framework around the new version. I call it the Goldie deep dive. And in five steps, you ask a loose question. It hunts
for sources. It gathers them, titles them, summarizes them. You chat with the whole thing to deeper. And
then you turn it into a visual. That could be a mind map. It could be a chart. It could be a podcast. Knowledge you can see and share, not just read.
And here's the honest bit like notebook on it own is fine, right? But most people make one report and then just never touch notebook again. The magic
is when you plug it into a bigger system so the research actually goes somewhere into content into
your SEO work into your next product. A coach for example could feed a stack of their notes into notebook alm and then ask it questions and walk away
with a finished guide. And the last one is Google's managed agents. And this one's a little bit different. So Google released a new way to run AI
agents in the cloud through their Gemini API. Now here's what that means in plain English. Normally,
when you run an AI agent on your own computer, it uses your machine and it ties up with managed agents, the agent runs out in the cloud instead. So,
on a computer that isn't yours, it's working away whilst your own laptop does nothing. And it's powered by Gemini 3.5 and Google's coding agent. So,
you can give it a task. It works on it for around 30 minutes at a time, goes off and gets stuff done,
and then it comes back later with the finished work. I actually had it go deep on a research report. It went out, found everything, pulled it
together, and put it into a clean presentation with dates and animations, which is way nicer than I would normally get. So, basically, you have a
sandbox computer in the cloud doing the heavy jobs. Nothing touches your own setup. It browses the live web, it
can run code, it can fix itself, and the finished files just come back to workspace. Now, was it perfect the first time round? No. Expect some back
and forth. The first report didn't look that great. If we actually have a look at the new report that it generated, this is the presentation looks
super nice. is clean and looks way better. Right now, imagine doing this research yourself. You'd have
like 40 tabs open. You're trying to babysit the whole thing. The new way is you just give it one task, you walk away, and your machine doesn't have to
do anything. An agency, for example, could send it off to research a whole new industry overnight and have a full brief waiting in the morning. So,
let's pull this whole week together because it tells one clear story. The most powerful clawed models
got pulled by the government in a way that's never happened before. The same week, two new models out of China stepped up and started winning tests
against Claude. A trick showed up that lets cheaper models team up and nearly match the best. And three tools, Hermes, Noble Clem, and Google Cloud
agents all got upgrades that hand you back hours of your week. Now, here's the takeaway. The tools are
changing faster than anyone can keep up with on their own. The model you love today might be gone tomorrow. The best new thing might come from a
company you've never heard of, dropped on a random Saturday with no warning at all. The people who win with AI automation aren't the ones who pick the
perfect tool. They're the ones who can switch fast, plug in whatever's new, and never get stuck when one
thing disappears. And that's a whole game now. And that's exactly why I built this whole agent operating system. It's one system where Claude, Hermes,
OpenClaw, GLM, Kimmy, Fusion Panels, Notebookm, Google's Cloud Agents, all of it plugs into one dashboard with shared memory. Now, you saw that this
whole week proves a point. When Fable got uh pulled, anyone running a setup like this just switched
models and kept going. Inside the air profitable boardroom, you get the agent operating system, the zip file to install it, a 30-day road map to set
it all up, and four coaching calls every week where we wire in the newest tools together the moment they drop. There's over 3,600 business owners in
there building right now, plus a member map so you can connect with people near you and daily
tutorials on every tool I covered today. Links in the comments description or go to the airruffborn.com. Thanks for watching. I'll see you on the next one. Cheers. So,