Self-improving AI agent loops enable autonomous quality control by having a builder agent create content and a separate judge agent grade it, with the system automatically iterating until the content meets predefined quality standards. This approach eliminates manual quality control by allowing agents to continuously refine their work through multiple rounds of generation and evaluation, ultimately producing high-quality outputs without human intervention. The system can be applied to various tasks including SEO content creation, video production, and other content generation workflows.
Install our extension to search inside any video instantly.
Vibe Coding a NEW Claude Agent OS!
Added:[music] welcome Prestige. Thanks for joining us, mate.
>> [music] >> If you're watching, [music] feel free to post your questions inside the comments.
We're going to build some stuff out inside the agent OS [music] system. Have some new features today.
You fun.
[music] >> [music] [music] >> Heat. Heat.
[music] >> [music] [music] >> Heat. Heat.
[music] >> [music] >> Heat. Hey, heat. Hey, heat. [music] >> [music] [music] [music] >> Heat. Heat.
[music] Heat.
[music] [music] Heat.
Heat up >> [music] [music] [music] [music] >> here.
Who we got here? Lazerof, welcome in.
What's popping? Watch tech. Am I real? I am indeed sir.
Hey, she exactly exactly. I'm going full DJ mode.
MC Goldie on the ones and twos.
Jay, welcome here. Good to see you.
Let's have a little look.
Nice.
So Sheena is or yeah, so Sheena is crushing it with her agent OS. She's inside the air puffer boardroom. As you can see here, she set up a a security update, which is pretty cool.
Client setup as well. This is really interesting. I've not seen this before, but this is I think a great update. And then we got Hermes runtime panel for configuring this. You can actually plug this in with uh the dashboard of Hermes Ruffalo async database demo seed module. Lots of cool stuff looking right there. Let's have a look and pull this up.
Orange hair. That's where it's all going.
It's a future.
Let's take a look at this. Thanks for sharing, Sheena, by the way. It looks fantastic.
Oh, you got Manis plugged in. Whoa.
Hyper Agent Manis perplexity.
Very nice.
Paperclip OS.
Not as good as the agent OS, my friend.
It's okay. It's just super basic.
This looks really, really cool. There's so much to it.
Uh, so you got your Ruffalo team over here and Hermes agents inside there too, right?
What else we got here?
Nice.
Full squadron.
I like the fact that you can uh click down here as well. The UI is super nice.
You can get it inside the air for boardroom link in the comments description or go to the airborne.com.
This is like the good thing about it with the agent OS. You can customize it as much as you want. So if you look at what Sheena's done with hers, it's it's insane.
So, she's taken the structure but made it her own and it looks awesome.
So organized.
And then you can change all the settings for your Hermes here, which is great.
And then you've got it split by categories as well. Right.
Next project is going to be Superclaw.
That's what we're building with Agent OS. Unless that's a new thing that I haven't even heard of.
Got the group chat loop engineering.
That's awesome.
Ruffalo Swarm setup.
Oh, nice. So, you're using the video agents, too.
Wow.
It's amazing to see like you're actually implementing all this and using it as a system. It's so cool.
Shout out to Shino. She crushed it.
Hey, hey, hey.
Get yourself in there. It'd be great to see you in there now. I appreciate that.
But it is a fantastic place. I mean I I don't see anywhere else where you can build that sort of stuff and share it in the same way. Like it a lot of that's I mean you just saw the system that Sheen has. It's just unbelievable.
Sounds good. You can plug that in with Hermes. That's how I use browser automation right now.
All right. All right, we're going to kick this off because we've just had a big update from Hermes already.
[clears throat] So, Hermes have just dropped version 0.17, the reach release. I'm going to guide you through exactly what it means, what's changed, what it means for you, how to use it, etc. We're going to get straight into this. So, you can see the GitHub right here and the new updates.
This literally just dropped 4 hours ago, and this is the new update from Hermes.
Now, if you don't know how to get this update, all you do is you update inside your dashboard. So, to do that, you can go to your Hermes, go to manage, scroll down, click on update Hermes, and it'll begin updating as you can see right there inside your dashboard. You can also go inside the terminal, and then once you've finished updating, it will say 0.17, not 0.16. Or you can go inside your terminal and you can say Hermes update to get the latest version.
So let's talk about the highlights, what's changed, what it means, what it means for you, etc. So some of the highlights here, number one, Hermes reaches iMessage.
So this means essentially that you can use iMessage with Hermes. You can run Hermes phonon login authenticate and then Hermes can send and receive iMes which means like you don't um which means that you can use your Hermes agent basically on your phone you can use it with iMessage is pretty good. So, if you've ever tried to put an AI inside iMessage, it's not that smooth. Now, it is using this new setup. Top of that, they've got something called Raft, right? So, Raft Hermes joins the Raft network as a gateway channel. So, a new bundled Raft platform adapter lets Hermes connect to Raft as an external agent for a wake channel bridge.
So you can see an example what Raph looks like. And you basically add Hermes to it now as well. And this way humans and AI agents can build together according their platform. Again, this is pretty new to me. I've never come across raft before, but that looks pretty cool as well. And then we have improved Hermes desktop app. So, if you've ever used Hermes on the desktop, it's gone a lot better this week.
By the way, some of these releases, they just they bring out like before the announcement of the big release. So, they'll just release them part by part as you can see. And then once you go into Hermes desktop, you can click update now and it'll begin updating and then you can get the latest version of Hermes Desktop. You might be wondering, okay, what happened? What's changed?
etc. So you got native OS notifications, live sub agent watch windows and all sorts of stuff here. Now there's also asynchronous sub agent so you can delegate work and keep going with Hermes. Now that's another big part of the update Now, if you're wondering, okay, what are asynchronous agents? So, you can see an example right here. Basically, what this means is that you can fan out and delegate all your tasks at once. So, you can spawn a team of sub agents all at one time that work together. Now, if you want to see an example of how that looks in practice, basically you can give each you can give Hermes a task and it will delegate it to multiple sub agents when it's a sub agent related task. Right? So basically you have a lead agent like your main Hermes agent profile. Then you give it a task and it will spawn like four to five sub agents and then it will actually gather those sub aents tasks together and you get the results back in your chat. So let's say for example you were building out a website as a use case.
You could say to Hermes, okay, go off, research this topic for this keyword, optimize the page, post it to my website, and also make sure you create a video at the same time, right? And so when it does that, essentially, you would have like multiple sub agents working on different parts of the task that are automatically delegated by your lead Hermes agent. So for example, it could be like one starts working on the keyword research, one starts creating a video, etc. If you want to see an example of that in action, here's an example of how it can play out, right? So, you can say, "Okay, go off and create this video." And then you've got like an editor, you might have a judge, you might have a video writer, and you might have, for example, a script writer, too. And they all work at the same time building out stuff together.
There's also an option for editing images.
So you can now edit images, not just generate them with image to image generate. This is a new skill, but it means you can now edit and transform a source image, not only create one from scratch.
This can work with foul, opening eye, grock, crayer, open eye, codeex or as well. And basically this means you've got image to image editing.
So it's one image tool, but it can do all sorts of things like for example text image or image to image and change the images as well. It's pretty nice. On top of that, we have automation blueprints.
Now you might be wondering okay what are automation blueprints? So basically these are things where you can schedule stuff without relying on the uh on on actually coding the syntax of the task.
So you can get these blueprints from news research in their official portal and then one blueprint can essentially just create a scheduled task for you. So you can see an example of that in action over here and how it works. But essentially what it means is that you can trigger schedule tasks and that could be as some as simple as something like okay go off and research the latest news and then from there it will go off and start implementing it.
So these are automation blueprints from Hermes. Uh this is exactly how it works.
though it's kind of like a piece of code that you can copy and paste from the automation blueprint section on Hermes agent news research but the difference is you don't need to understand it you can just plug it into Hermes and Hermes will guide you through step by step exactly how to implement that now you might say okay what are the use cases so there's all sorts of stuff for example like fixing PRs um coding tasks etc but if you're not really into coding or you're not super technical then essentially you can just do something like a AI use digest. It breaks down the steps. So you can read the three steps here and what it does every seven days and how it structures it. So it works in in a few simple steps. Number one, the task. Number two, the steps. Number three, the actual structure of the output. So what it looks like once you get the report back. And then number four is how to actually word it. So for example, keep it under 600 words. Keep each item to one to two sentences. And then you're good to go on that. And so you can copy and paste stuff like that into Hermes agent and then you've got a full uh automation blueprint ready to go whenever you need it. And it saves a lot of time and also it means you get better outputs without having to be a better coder which is nice. So it's it's a good feature.
There's also Kurs's composer model that has been added through your X subscription which is pretty cool. So you know if you're already subscribed to Twitter for example you you can now use curs's composer model gromp composer 2.5 fast as a model inside Hermes. So if you want to set this up for example we can go to our terminal here. We can type in Hermes model.
Then we can scroll down to XAI GRO Groco off. We just log in from here.
Go back to the terminal and then you can see curses model which is Grock composer 2.5 fast that you can select inside the list right there. And so it's really easy to switch models and change this. Now, there's also a full profile builder inside the dashboard. Now, if you never seen this before, basically you can have multiple different agent profiles. And this is really powerful. I'll show you an example in a second, but basically, you see here, we've got 23 profiles that we can manage inside our agent operating system for Hermes. And by the way, if you don't have an agent operating system for managing Hermes, definitely recommend it. I'll show you a couple of features in a second. But essentially what this means is like you can assign different profiles to different tasks or even different APIs. So for example, we have GLM 5.2 that just dropped and we've got that assigned as a profile to our agent. So we can switch to that anytime we want just with a click of a button.
We also have for example Javis, we have Kim K2.7 and we can switch between these models super easily. Now to do that with the new update, you just go to manage and then from there you would go to profiles and from there you can start building with this stuff. So if we go inside the chat here, we can switch between these different profiles that we have which is a big update. Number two, each profile can have an individual skill, custom instructions, and it's super easy to manage them. And number uh three is like inside your camb board, you can assign tasks to your AI agent profiles. Let me show you an example of that in action. So if we have a look over here, we've created a video and a blog post using Hermes agent, right? So that's a full blog post deployed on our website and also a full video as you can see right here. Pretty amazing. Now, we did that because we have this new agent profiles update that allows us to create separate profiles. So we have one that's a content judge and basically what that does is look at the content my team of Hermes agents creates and then tells us if it's good enough or not. If it's not good enough then the team have to keep going around until it is good enough.
Here's an example. So this is a video director. So basically what the video director can do is go off and start coming up with ideas for the video it's going to produce which is how we got this end result right here. And so you can have a team of AI agents working together. You can see the judges result here. So it scored them four out of 10 and then they had to keep going around and iterating. You can see we have a content editor and basically this is how you can create a a lot of great content because you've got that quality control in place because you got multiple agent profiles that are creating the content and also quality control on it. And you can do this for any workflow. It's just a case of like creating a loop between your agents so that they can go round and round and improve and improve. And so it's a pretty powerful feature.
Now the skill browser got a complete overhaul as well. So what that means essentially is if you go to the skills inside your AI agents.
So you can run the terminal command Hermes dashboard or you can use our agent operating system link in the comments description or go to the approper one.com to get access to that.
And then once you've done that you can scroll down to the skills section here and you can actually browse all of the skills that are available and you can search by keyword. You can see everything that you've installed. You can toggle them on and off. You can see them by category.
It's much easier to manage all these skills. Now by the way with that Hermes agent profile what we also have here is like we can chat between our different agents. We can talk. We've got Hermes Jarvis, which is actually a profile where we can voice activate a Hermes.
And we can see the full conversation. It has a wake word, so it can just sit in the background ready for you to uh scroll through. We have a briefing button here where basically it can look through all my recent Obsidian notes and then give me a weekly briefing. We can change the voice on this. And then also we have everything that we've built in preview right here as well. So things like the agent profiles, it might seem like a small thing, but actually if you can create a separate profile that does all this amazing stuff. Well, how powerful is that?
The memory, it's all got a major upgrade as well. So basically, you can change the memory better. It has uh some other small stuff as well like a better dashboard login, official WhatsApp, rich text for Telegram, some uh token optimization options, new UI options, but basically the main things I've talked through today are the things that are the most important from this. Uh one final thing is the agent loop uh section here. So basically what this means, you can do loop engineering with Hermes. Now, if you're not sure what that is, basically this allows you to have a loop where you have one agent building stuff and one judge who checks stuff. So, if you look at this feature we've built inside the agent operating system here, you can basically have a goal. So, you can give your Hermes agent a goal, and this could be on a free or a cheap API. We've got free ones plugged in here. Then, we can give it a starting point. Then we have a builder and we can set how many rounds that loop goes around for where we have a judge who looks at the content and says is it good or not? Now if it's not good enough it goes around in a loop and that is Hermes loop engineering because you can basically automate the quality control of whatever you're building until it's actually good. So you don't have to manually check it. You don't have to sit in the middle.
So this is how it works as a system. You set the bar, the builder acts, a judge grades it. It goes round and round until it's fixed and then finally it's done and you get the output. And you don't need to sit in as the quality controller. Your team does that for you.
And when I say team, I mean your team of AI agents. And here's an example of what we got. So this was a real loop where each round the judge scores harder, work higher. And you can see eventually on the pass line it hit 92 and then we got the work done. So the the judge scored the Hermes agent 54, then 71, then 83, and then finally 92. So it just got better and better. And this just saves a lot of time because now you don't need to give the feedback manually.
It just goes round and round as an iterative positive feedback loop until finally you get the good stuff back.
So that's basically it for the Hermes V0.7. Lots of cool stuff right there. A lot of stuff that people won't realize you can do, but if you're watching this and you saw everything that we did today, then you're probably in the top 99% to be honest with you. So, this is all the cool stuff. We've got an agent operating system here where we can actually orchestrate our teams of agents. We can have a a group chat with Hermes directly. We've got a pipeline where we go from idea to build and then we can check everything that we built over here, which is pretty awesome. And then we have a local agent cambban with local versions of Hermes running 24/7.
Top of that we have a chat here where you can talk to your agents. You've got Hermes Jarvis the studio where you can generate images, videos and voice. And you can even connect Hermes directly to Notebook LM. And basically everything you want to automate is all inside this system here. So if you want to get that is inside the AI profit boredom community link in the comments description or go to the AI profitwarm.com inside the classroom. You can ask questions, get help and support in real time. Lots of people using the AI AI agent operating system as we built before. If you want to get that, it's inside the classroom and then just go to the agent OS and you can find the video tutorial on how to install it. You can see the last update date. So, we update it daily with new cool stuff like I've shown you. Uh the new resources with a zip file. And then also if you like stuff like for example Loop Engineering, we've got video tutorials and step-by-step guides on how to use it.
Like even a Hermes shopping assistant, like all sorts of cool stuff as you can see right here that you can get access to and learn from every day. And if you're a complete beginner, you can go from beginner to expert over here. If you want to jump on coaching calls live, share your screen, meet call members, ask questions in real time, you can do that inside the calendar. Inside the map, you can actually meet people in your local area who are building AI agents like you. And this is all inside the AI profit boarding. Link in the comments description or go to the profitboard.com.
And you might say, well, this stuff might be technical or I don't know how to use AI agents. We've got over 184 pages of testimonials from people winning, learning, and growing with AI automation. Many of them have never used AI automation before. So, if you want to get the most out of this stuff, check out the air profit boarding. Thanks for watching. Bye-bye.
Let's see what we got here. We got Randall. Welcome, sir. I am of course good today. How are you, sir?
I've built some good stuff. I need someone to overlook it. Yeah, ask inside the air for him. Lots of uh awesome community members there ready to take a look for you. Usually I take a look too.
Big updates, Randall. Big updates. I have a question. Can we make our own operating system like Windows? Yeah, you can. Uh let me show you an example of what that looks like in action. And you know it's only it would take a long time to actually build like something like Windows. But if you actually want to build like an operating system that looks like Windows, let me show you an example of that. So if we go to it be GLM 5.2.
Yeah, we built GLM OS, right? And basically what this is is an operating system. So you got like, you know, it kind of feels like almost the Mac OS.
You can see the battery, whether we're on Wi-Fi, the date, etc., and the time.
And then it has like a notes app as a calculator over here. We have the terminal that we can use. And we have a music option. And also paint as an app, too. So, can you build your own operating system? Yes. How did we do it?
We did it inside the agent operating system. And then we used the CLI or Kimmy to build out all this cool stuff.
As you can see, I was thinking about integrating foul.
Foul is good, but I just don't like messing around with too many APIs. And also cuz I've got gro and miniax, I don't really need it. Are you showing Hermes agent OS? I changed [laughter] I changed Hermes Jarvis to Jasmine with my own voice. Wow, that's cool. Got loop too.
Ah, interesting. That's good to know.
Thank you.
[music] [music] Heat.
Heat.
[music] [music] Heat.
[music] >> [music] [music] >> baby.
>> [music] >> Heat up here.
[music] >> [music] >> Heat. Heat.
>> [music] [music] >> Heat. Heat.
>> [music] [music] [music] >> Hey, [music] hey, hey.
Hey, [music] [music] [music] hey, hey. Hey, hey, hey.
[music] [music] [music] Hey, hey, [music] Heat up [music] [music] here.
Heat. Heat.
[music] [music] >> [music] >> Heat.
[music] [music] [music] Hey, heat. Hey, heat.
Heat. Heat. [music] >> [music] [music] >> Hey, hey, hey, hey, hey, hey.
>> [music] [music] >> Fable 5 is not back yet.
Honestly, like this week is felt a little bit subpar using Opus 4.8. I'll be honest with you.
Don't know what anyone else has experienced, but hopefully Fable 5 comes back soon.
We can only hope Hey, [music] hey, hey.
GM 5.2.
The best. The best. Let me show you Goldie Bench.
[music] All right. So, this is Goldie Bench. And basically what we've done here is, you know, I I test this stuff for hours every single day. And so we've got a leaderboard here where we rank where all the models perform that I've tested out recently. And it will keep updating as we test more and more models as well. So if we look at this, for example, Opus 4.8 still at the top. Um, this was built before Fable 5, well after, sorry, Fable 5 got taken down, but you can see GLM 5.2 right at the top here.
Um, it's pretty impressive, right? Open weights, tested like 21 tasks on it, won a lot of medals on the benchmarks, scored 8.23 out of 10, million token context window.
It is the best. And if you're wondering, okay, what did we build with it? Let me show you some stuff here.
So, here's an example.
Here's another one.
We created this like cool ray caster as you can see.
But yeah, you can build a lot of cool stuff. I mean, this is like a full open world game that we created with GLM 5.2. So, it's pretty nice.
Like, it can can create some nice stuff.
If you're wondering like how did it perform on the benchmarks, etc. So if we pull up the matrix of tasks here, we have the models at the top. So Opus 4.8, GM 5.2, Gro Quinn, Kim K217.
This was a spiral galaxy created from GM 51.2. It's pretty nice.
So yeah, you can compare the alpas here and see which one you think is the best.
They all use the same prompt.
Back on it.
Go get them steps in 20,000 yesterday.
Hey, hey, hey, hey, hey, hey, hey, hey, hey, hey, hey, hey, hey, hey, hey, hey, hey.
Heat. Heat.
Hey, hey, hey.
>> [music] >> Hey, [music] [music] hey, hey.
>> [music] [music] >> Hey. Hey. Hey. [music] >> [music] [music] [music] [music] [music] [music] >> I feel [music] [music] [music] >> [music] [music] [music] >> Hallelujah.
[music] Hallelujah.
[music] >> [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] >> I feel [music] [music] [music] [music] heat.
>> [music] [music] [music] [music] >> Falling. [music] Hallelujah.
>> [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] >> I don't feel [music] [music] [music] [music] >> [music] [music] [music] >> Heat. Heat.
>> [music] >> Hallelujah.
[music] >> [music] [music] [music] [music] [music] [music] [music] >> Hey. [music] [music] >> [music] [music] >> Heat. Heat.
Hey. Hey. Hey.
[music] [music] [music] >> [music] [music] [music] [music] [music] [music] [music] [music] >> Hallelujah. [music] Hallelujah.
[music] Hey. Hey. Hey.
[music] [music] >> [music] [music] [music] [music] [music] [music] >> Heat.
[music] [music] Heat.
[music] [music] >> [music] [music] [music] >> I [music] feel [music] [music] it all.
>> [music] [music] >> Halleluah. [music] [music] Oh. [music] >> [music] [music] [music] [music] [music] >> Hey, hey, hey.
[music] [music] [music] [music] >> [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] >> Heat. Hey, Heat.
[music] [music] [music] >> [music] [music] >> Falling [music] love.
[music] >> [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] >> Falling.
Falling.
[music] >> [music] [music] [music] >> Hey. Hey. Hey.
>> [music] [music] [music] [music] [music] [music] [music] [music] >> Heat. Heat. [music] Hey. Hey. Hey.
[music] [music] [music] I feel [music] [music] [music] [music] >> [music] [music] [music] >> Heat. Hey. Hey. Hey. [music] >> [music] >> Falling around.
>> [music] [music] [music] [music] [music] [music] >> Hey, hey, hey.
[music] [music] [music] [music] >> [music] [music] [music] [music] [music] [music] >> Feel [music] [music] >> [music] [music] [music] [music] [music] >> Falling around.
>> [music] [music] [music] [music] >> Heat. Heat.
[music] >> [music] [music] [music] [music] [music] [music] >> Let me know if you DM me the issues. I can look into them and just make sure the agent OS updates it skill to make sure it's easier to install uh for Linux. So if you if you just direct message me inside there profitable volume, I'll get you that.
Fable 5 is not available yet, my friends, sadly.
Can Jarvis pull up any data I ask it to?
If it's inside your Obsidian? Yeah. Or you can say like, okay, go to this website, go to that website. So you could say, okay, navigate to search console and then it could open up for you.
Let's have a look and see. We'll try that now.
Open up Google Search Console.
Yes. So, you can see it just opens it up like that. That's an example. That was super fast. Actually, I was impressed.
Happy D, man. Happy D.
And then you can see the full history here. Everything that were built with Hermes Jarvis as well. So it's pretty cool.
Uh, netifier.
If you have any questions, by the way, if anyone's watching, you have questions, you want me to look at anything, uh, feel free to post and I'll take a look for you.
[music] on Claude. Yeah, I use the max plan.
[music] [music] [music] >> [music] >> So, Netlify Hey, I'm just I I don't know what plan I'm on actually. Uh let's have a look. Netify plans.
Is there even a max plan? There's no such thing as a max plan. [laughter] Um I think I'm on the the pro plan here, but I'm thinking about switching to Cloudflare just to make it easier.
Yeah, I guess it depends what you want.
And do you have super base? No.
Uh, I don't really build anything that needs super base, but you could build it in if you want. You can easily add it.
Yeah, that's it. So what you can do is you can have the the previews here with anything that you've built or you could customize this and add more widgets inside it as well.
New north model or it's just good for like basic coding tasks locally. So if you want a local coder that's pretty lightweight, North Code is decent and yeah, it's ready to go all the time.
So, so for example, if we go to local engine here and then we type something in like for example uh I don't know build a habit tracker.
And then if we go over here, you can see it's running and building right now.
And then once it's finished, it will have the preview that we can check out.
So you can see the preview here.
So it builds locally inside this section. We ask for it to build a habit tracker. It creates it. We can preview it. And then we've got everything inside our workspace ready to come back to as well.
Yeah, I got loads of training on that inside the air profit boardroom focus as well and avoid overwhelm does agent OS display display cam without modifying it. So it' be the exact same in Hermes dashboard. So from what I've seen the setup inside Hermes dashboard for Cambban is not that nice, right? So this is way nicer to use. So if you go inside Kambban here, we can just plug in a task. it will get it done and it's just much nicer and easier. And then you can have separate camb boards as well.
So if you go over here, you can see the camb board. You can see what it looks like. Uh you can organize everything in one place. And I think that's a lot nicer. Like even just the way it's color coded, you can preview stuff. You can see what people have built. Uh it's it's all set up correctly. And then you can even like preview the stuff that was created as well. So I think that's much nicer than the existing camber.
Yeah, that's a great way to do it. You can just use Claude to orchestrate the agents to to talk to each other. I think that's a great way to do it. So you can do that with Kamban or you can do it with Claude or you can get Claude to actually run the camb for you, which is what I do sometimes if if I don't have much time.
In fact, I would say out of all of those, getting claw to run your camb board way simpler, way easier. Um, yeah, would recommend when I'm building a lot of this stuff or I'm showing new workflows, I'll get Claude to set it up directly and orchestrate the agent OS just because it'll be it's easier, it's faster, it's smoother, and also like you don't have to do everything yourself, right? It's It's way faster and easier. And then you can go off and do something else in the background.
All right, we should have this verified now. Let's check There we go. Oh, nice.
This one we can remove.
This one is getting set up. Nice.
Do you use APIs or uh it depends what we're using but yeah most of the time we use a combination of like free APIs and then the CLI itself. So, for example, like Grock we already have with Twitter and we plug that in the system and then that's ready to go whenever you want.
So, yeah, for me personally, I much prefer to just use a subscriptions like Claude or or GLM C uh 5.2 whatever. It's it's way cheaper than using the API 100%. But at the same time, you could always switch to a different CLI, right?
So, for example, if you couldn't use Claude anymore, I'd probably just use GLM 5.2. do like that's that's the way that I look at it is like you can you can switch to whatever you want.
Um the good thing is if you have the system like you have the agent OS and it doesn't matter what happens you just swap the models in swap the models out no problem.
Let's set up the sign out for this.
GLM will yeah you can plug it into claw code already uh released an easy way to set up GLM 5.2 with claw code. So you can go to GLM 5.2 here and then run it here and then if you have the cloud plan with Olama you can easily plug GM 5.2 to interclaw code and it's pretty chill example here. So you just make sure you've got alarm running and then copy and paste that command.
creating a SAS and it requires users to select a plan. Why would you need to reach out to OpenAI or Anthropic? Right?
Like if if you want to get an API, you don't need to reach out to them. You can just you can just get the API from the website. Uh yeah, that's what I'd recommend.
And if they need to plug in their own plan in there, then you just give like an easy setup so they can easily switch the CLI into it. So for example, with the Agent OS system, like it's pretty easy when people install it to make sure their CLIs are built in.
Well, it depends. It's like there's an easy way and a a simple um a more complicated way. So I would just go if you want if you're worried about time just go with the the simple way and we give you the full instructions.
But yeah, you can see for example inside the a profit board you can see for example like Sheena she's she set this up within like a day uh or within a couple hours. When I set it up personally on my own laptop it took me like an hour or two just to get it all configured and it was good to go.
So yeah, it's pretty quick. It's designed to be fast as possible to set up and we're improving it daily, too.
Put all your SEO reports in the OS might be something I do in the future, but right now I don't do that.
I think potentially, but I would recommend that instead of doing that, you just use a video agent inside the agent OS cuz that'll be far more effective. Bear in mind like it's really a custom workflow.
So I wouldn't go to and and also yeah I just wouldn't go to M Jarvis to fix that.
I would use the video agent instead.
One sec. Back in a sec.
So yeah, I would use a video agent here instead.
And the same for like images, I would use this section instead.
Well, I mean, if you want to use free API keys, then it's free. You can make it as expensive or as cheap as you want to depending on what you want to set up.
So, for example, if you already have like a Twitter subscription, you could just use a free CLI with Grock.
Or you could use, for example, like curses 2.5 composer model that just dropped with a war on Grock. So you know even like for example anti-gravity CLA is free to use. So you can just plug in your existing stuff and then you know there's no extra requirements.
But yeah, the the absolute minimum would be free.
Honestly, I don't recommend that anyway.
Like either way, uh when I've tested it, it's not very good. Wouldn't recommend that. I just feel like that's what Twitter is against right now.
And so it might work for like a week or so when I've tested it, but then you know the account doesn't last. So I wouldn't recommend doing that.
That's why we haven't done it.
Otherwise, we would have built that in.
>> [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] >> I love you.
[music] >> [music] [music] [music] [music] >> So for this, let's have a look here.
Yeah, for this we we have a team that will post for us because I think if you look at these platforms like they don't they don't like to uh to send it straight there but everything else we can do. It's just like the actual posting we don't. But um I do know some ways around that. So, if you DM me inside the air prof, I'll send you some examples of what we've done in the background that I probably couldn't talk about here. [laughter] Uh, let's have a look. Yeah, additional question. An agent OS. Yes, it can. So, I've linked it to Comy UI before. Honestly, I didn't use Comy UI that much, but we've linked it to Comfy UI before. You can also link Hermes agent to Comy UI. They have an MCP for that. And then when it comes to stuff like NAS as well, they have an MCP and also Claude has a connector for that so you can link them all together.
Yeah, Sheena says if you wire it up correctly, Claude can do that 100%. How is your staff using AI? So content creation, schema, reports, etc. That's the sort of stuff we're using with this.
You would just give it the you would whatever you use to set it up, you would give it the details of the 11 Labs voice you want to use with the API and then you can just change it from there.
Let me add that as well.
Wow, the claw just stopped.
That is not useful.
Wow, look at this.
This happens like every day now where Claude just stops working.
Yeah, it could if you have a good enough setup.
You need a really good setup for that.
Oh, look at this. Claude has stopped.
All right, let's move on to something else.
>> [music] [music] >> Falling.
[music] [music] [music] >> [music] [music] [music] [music] >> Let's see what questions you got here.
Wow.
So, okay. Number one, thank you so much.
Number two, let's see how we can help you. So, been building an app in Claude Code for over 3 months now. Learning has greatly evolved. Fantastic. Well done.
Um, my earlier work is full of code that I don't want to use.
How do you change stuff without breaking it all? It's a good question. So, the way that I would look at this, right, is number one, keep a backup of what you have. Number two, have a staging environment.
Number three, test it before you go live on anything.
And number four, ask Claude to remove any efficiencies or unnecessary code step by step.
Not everything, but focus on the 8020.
Eg. What are the easiest changes to make without breaking everything?
So those are the the five steps I would make right keep a backup of what you have number one have a staging environment test it before you go live on anything and then ask to remove any efficiencies or unnecessary code step by step not everything right um just focus on like the 8020 okay what these changes to make without breaking anything what are the 20% of changes the remove 80% of the code that doesn't work and just keep a change log as well of what was updated.
So those are the six things I work on.
Now having said that, am I like a an expert coder? No. Like but that's just my advice based on what I've learned about coding with AI. So that's the way that I would approach it just to be safe. And then you got like six simple steps you can follow step by step just to make sure that you you know you remove the inefficient code but you also keep it useful as well.
What is that? Let's have a look.
Ah, nice.
No, I haven't seen that, but it's a cool idea.
People are saying it will be back by July the second. That's when people are most confident.
Interesting. Great website.
Thanks for sharing, Mark. Good to see you.
Just a note that my my screen is randomly uh it stops sharing occasionally. So if that happens again, I'll be back in like 10 minutes. Uh I just probably need to restart or something like that.
So, we have a brand new update from Notebook LM where there's three big changes to notebookm itself, which is a powerful free research engine that we can use, for example, as an AI agent for creating all sorts of research reports.
It can create videos, podcasts, etc. I'm going to show you the new setup here that's just come out recently. So, first of all, you get a better chat experience. So, you can actually uh plug in sources as you can see and you can add your own sources as well. And it's got over 100 cured software skills. So, you can get better, deeper research, more complex analysis. So, that's the first big update. And then number two is that you can visualize data, PDFs, images, and Excel sheets as well. And then finally, there's an Aentic research companion, which means you can now start a notebook by entering your loose ideas and questions into the chat. Then notebook will guide you through it. Now, what we've actually set up with our agentic operating system and if you don't have something like this set up, I definitely recommend it because you can basically manage notebook in a more powerful way where you can segment everything. You can see your library of all the notebooks you've got set up. You can do more research inside this section and ask it questions. You can chat with your notebooks as well. So if we go to the chat here, we can actually speak to our notebook directly. And then also we have a studio here where we can generate anything from that particular notebook.
So that could be like a presentation, audio view, video, my map, infographic, flashcards, etc. Now we've used an MCP to connect these. And the thing that I would say here is like if you saw Fable 5 recently get taken down by Claude, then you know that it's much better to have a system where you can plug models in and out and it doesn't really matter what happens next. So you have much more control over the system. So this is why we've set up this agent here and then what we can do from here is we can take the assets that we've generated with AI.
So we've got for example the research reports, we have the slide decks, we have the videos and the podcasts and everything else and even infographics.
And bear in mind like you can generate all of this for free using this setup.
So let me guide you through it. This is something I call the instant research engine with notebook because basically you can have like a a really wellressearched report or podcast or video whatever you want and you can just feed it a pile of links right um so this is powerful stuff. Now the engine that powers it is notebookm but as you can see like notebookm is super messy itself. So if we go directly into notebookm you can see that we've got notebooks all over the place. We don't know which ones have videos generated.
We can't see everything that was generated. There's no real library there of everything that we've created. That's why you want something like this where you can see everything you've built in oneclick glance and then you can get access to it straight away. Right? It saves a lot of time. It's a lot more efficient. And then if you need to go through some of your notebooks here, it's much easier to see it this way than it is to look at it like with a a big grid and then you have to look through the whole list, you could do the list setup here as well. But again, like you don't really know what you've created or what setup or you know what each one of these is about. It's much easier to organize it like this as you can see. So we can ask it one question and then it can answer from every source with citations. And this is a really powerful research engine that we've set up. So for example, if we say, okay, you know, what's the the one thing that a builder should know about the June 2026 release, then it comes up with answers and sources like you can see. So it comes up with, for example, the new coher north mini code update, the GLM 5.2 update, the new research from Hermes agent, the new release. And so like in one single question, you can get the latest news and it's really really recent. Whereas for example if you ask that same question is I applaud or something like that it's usually quite outdated and it will find stuff from like two weeks ago or 3 weeks ago this is all stuff within the last few days. So the way this works is like you can plug in your sources the engine answers and then you get cited answers which makes it a much more powerful setup.
So this is how it works. You can feed it like different ideas, different sources, etc. It can read through those.
It connects and then it can perform, right? And by perform we mean like it can generate different assets. It could generate videos or it can generate slide decks or whatever you want. And then you can actually ask it and get cited answers as well. Now, how do you feed it? So, you would just go inside the library here and you can type in new notebook and go from there. How do you do the research? So, you can select a notebook that you've created. go into the research section here and ask questions. Now you can switch between fast and deep. Deep is the research agent where it can really look deeply agentically and find relevant information in the news. Now inside the chat as well, you can speak to it. So this is kind of like a custom agent that's fully trained on the sources you've plugged in. So for example, this is about the agent operating system and the June releases. And then we can ask it questions and go from there, right? And you can see how it replies.
The same for the research. So if we go inside this one and then we go to research, we can ask it and then we can get visuals as well. So you can see here for example, we said are you there? And this is custom trained on my agentic operating system and then it has all the information about what we've done recently, what we've plugged into it, what we've used, etc. which is super powerful. So the library itself is great for getting research. The research section is good for going very very deep on that research and getting cited sources. The chat is great for just asking it general questions like for example if you need to learn something or if you need to like get maybe like you generate some content something like that. Then finally you have the studio as well. So based on this different stuff that we've plugged in here we can for example say okay create an audio overview of the agentic operating system. And what that will actually do is generate a podcast about that particular topic. So if we've trained it to know everything about the agent operating system that we've built, which is this system we're using right now, then it can generate podcasts, it can generate videos, it can generate infographics on the spot using that particular example. So you can see in progress for the podcast that we've just generated, in progress for the infographic, and you can also generate multiple things at once. So you could have parallel agents working to create podcasts and infographics and videos and mind maps and everything else at one time. So you can see two examples right here that it's pulling in. Now you can view those at any time or you can click the the pull option here and actually pull it into your um library as you can see here. So if we have a look at this one for example, this is a research report we've generated with notebookm that's custom trained on my agent operating system. knows everything about my agent operating system and then I can train my team. I can create content with it. I can generate social media posts. I can create videos and podcasts about the agent OS. I can even get the report and open up the research report. And this is just a beautifully organized research report, a full PDF with seven pages of useful information about everything that we've plugged into it.
Right? So it talks about the ecosystem, what we've used, how we set it up, the loops that we've plugged in, the models that we've used recently, right? So you can see here it's like, okay, here's three big breakthroughs in June. So for example, Hermes agent v 0.17 that just came out, GLM 5.2 and code here north mini code, which is a new local model.
And then we can see the actual benchmarks on this particular model. So this is about GLM 5.2. It talks about how big it is, um, how it performs on SWE, how the architecture works, how the system is set up for safety, and the same for North Mini code, right? And it's generated all of the research, all of the content, all of the diagrams, all of the charts, etc. directly inside this setup, which is pretty powerful stuff.
So, it's really cool as you can see. And then you can go back inside the chat.
You can ask it questions. You could say, "Okay, what did we release in June and then it will start thinking and come back to us in a second." And then whilst that's thinking, we could go off and do something else inside the agent operating system. So whilst this is working, we could go off into anti-gravity and start building something else out. We could go into claude, we could go into paperclip, etc. Right? So these agents also work in the background in parallel whilst you're doing something else which is pretty amazing in itself.
And then we've got the research reports.
If we've generated for example that podcast we can see where we're up to. So that's still in progress but we have the infographic here.
So this is the infographic and you can see that's now plugged into our system.
Right. So we've got that infographic ready to go.
So it's pretty amazing because you know you you're essentially creating like a custom agent creating done all your sources. It can generate videos, infographics. It actually works very smooth to use. It can generate amazing research reports. It's just a full agent. Now bear in mind as well, Notebook LM itself is free to use. The MCP is free to use. You could create your own agent operating system like this and build it for free with something like Claude Hermes as well and free APIs and then you're good to go as well, right? So you can make the whole ecosystem free if you want to as well.
If you want to get our system, it's inside the AR profitable border, the agent OS, but if you want to build your own, you could do that too. And so the way it works is like you feed it with your sources. So for example, we fed it with information about our agent OS system, then it reads the sources, it connects to notebookm, it turns research into stuff you actually use. So for example, it could be like a podcast, could be slide decks, could be a flashcard guide if you're trying to learn something, a study guide, etc. And then you can ask it questions inside the chat as you saw before. And so you plug in your sources or information like we plugged in information about our agent operating system that goes into the instant research engine. And then we get a podcast, we get a briefing document, a mind map, flashcards, etc. All inside one beautiful system, which is super powerful. And you can see the quality of this stuff. Like it looks super nice.
This is an amazing infographic we generated. And the reason they look so nice is because it's using Google and it's using Nano Banana 2 to generate the outputs and they look absolutely amazing.
Same for this slide deck. So you can see the full slide deck right here. It looks great.
And the great thing about this is like before, you know, you'd have all your articles that you were researching in one in different places, notes everywhere, very hard to find and build everything together. Whereas with this system, you can add the sources, walk away, come back to a podcast, come back to an infographic, come back to videos, etc. And it's a really, really powerful system. You might also say, okay, this is too technical to set up. But you can see AR profit boarding members here are getting amazing results. So, you know, if they can do it and I can do it, then you can do it, too.
So, the first thing you're going to see is the dashboard. Then, we've got the studio for generating stuff.
And you've got the agentic agent as well for research, too.
So, you get a research engine, podcast whenever you need it, answers you actually trust cuz they're cited. You can generate social media content or content in minutes, not afternoons. And it's all inside one dashboard because it's organized inside the agent operating system. So if you want to get our setup, the agent operating system is inside the AI profit boardroom. It turns the nom research agent into part of one system. So that this can also plug into your obsidian memory. You've got the full agent operating system zip file to install the setup walkthrough, four weekly coaching calls, daily tutorials, a 30-day road map, and 3,600 members who with a member map and a 24/7 community.
Right? And again, if you're thinking this sounds technical or it's hard to set up, etc., it's pretty simple. Like, you can see how many reviews and testimonials and wins we've got from people using stuff like this and particularly building their own agent operating system. So, I'm not a coder, I'm not technical, they're not technical, but we can all build this together. And the great thing is when you have a community, you all learn and you grow together instead of like kind of, you know, coding alone, which can get a bit boring. And uh you need people on the journey. That's the thing that I would say. That's the best thing about this. So feel free to join us inside here. Inside the community, we answer your questions. You can get help with support in real time. Inside the classroom, you get access to all my best trainings. If you want to get the agent operating system, we've got it over here. If you want to get a full training on how we use notebook inside the agent OS, you can see that we've got a full tutorial guide on it right there. Inside the calendar, you can jump a weekly coaching calls inside the map. We've got people you can connect with in your local area and city who are building with stuff like this as you can see. And it's all inside the AR profit boardroom.
Link in the comments description or go to the aiprofit.com. Thanks for watching.
Let's see what questions we got here.
Keep it up. Thank you very much, sir.
So Randall says, "I use notebookm a lot.
Your agent OS is really useful. Thank you so much. Do you have a video on just instant research?" Yeah, you can find it inside the classroom. So if you go to the classroom here, then you go over to new daily updates. You'll see the notebook research section here. You can also if you just type in notebookm inside the community, you can find all the cool stuff here as well.
That's cool. I like that.
Yes. Yes, it can. No, you don't. You can just use MCP. You don't need the API for No.
Yeah, if you feed it that information, it could morning. Look at all this. Thank you very much. Yeah, I mean like we've built so many amazing things with this.
Yeah, it's all included inside the agent operating system, the nom setup and there's a training. Yeah, like Sheena says, there's a training and resources in the boardroom on that topic.
>> [music] [music] [music] [music] [music] [music] [music] >> Hey. Hey. Hey.
[music] [music] [music] I feel [music] [music] [music] >> [music] [music] [music] [music] >> Oh, [music] hey.
[music] This is a good question. So got a question here which is obsidian is tech not technically a memory layer. What are you using for agent memory? So I do use obsidian as a memory layer. So everything that we do it gets logged inside our memory galaxy and then we can use it whenever we want. Now, if you go inside like Hermy setup, they have honcho and stuff like that. But honestly, for me, like Obsidian does the job. It's it's really good for giving context. Still my agents understand it.
Like for example, when we were using Nom before, it actually plugs that research and the chats that we have with Notebookm inside our memory galaxy. And so the great thing about that is like every time we create a piece of content or we're using Notebookm or whatever we do inside the HON OS, you can see this is updated like every hour. The great thing about that is that this all plugs into our memory system. So, whatever we use with Hermes agent, um, you could make it as complicated as you want to, but I think obsidian is good to go. Like, it's it's as good as you need it to be. You can organize it differently and you could add honcho and stuff like that, but I genuinely think like obsidian is is enough to plug in as a memory for most people.
Let me check. You mean on Claude?
Mine resets in an hour.
Today I'm going to show you how to use loop engineering with AI agents for AI SEO. And this is something I've been experimenting recently where basically your agents self-improve themselves.
They quality control their own work which is very important for SEO. And then for example, you can have multiple agents working together to build the content, but then also to grade and iterate the content if it's not good enough. And you can see, for example, our rankings for this website right here. The trajectory is awesome. It's gone from basically nothing to getting, let's have a look, 222 clicks per day and that's growing all the time. So this is a really powerful system for just making sure one you create SEO content that actually ranks and number two selferate and improve based on this system. And it's just like a one-click system as well. You can make it very easy. And I'll explain exactly how that works in a second. So this is something I call the self-improving SEO engine, which helps your AI rank itself. So build a writes of content. A separate judge grades out of 100 and lists what's wrong. It loops until it passes, then it publishes. And there are two ways to run it with AI SEO based on what I've tested. And you can see a real blog that's actually written right here. So this was uh a blog published with these AI SEO agents that created the whole system and then actually talked about the system inside this blog they created together. So again this was fully published. I can't even log into this website.
My agents completely control it and they quality control everything. You can see it's actually written nicer than most people can actually write a blog. It's formatted beautifully and the traffic as well is actually there working to back it up as well.
Now if we actually see okay does this rank not just inside Google but also inside AI. So you can see us if we type in for example this keyword here best AI community you can see us ranking number one inside Google AI overviews here for this keyword and also here and then also you'll see that we rank as well directly By the way, my screen keeps stop stopping sharing. So, I'm just going to restart and then I'll come back in a second in about 5 minutes.
>> [music] [music] [music] >> I love you feel.
[music] Heat. Heat.
[music] [music] [music] >> [music] [music] >> Oh, I love you love. [music] Hallelujah.
[music] [music] [music] >> [music] [music] [music] [music] >> Heat. Heat.
>> [music] [music] [music] [music] [music] [music] [music] [music] [music] [music] >> I feel I feel [music] [music] [music] which you can see over here inside the agent operating system. You can use the loop section here and inside the loop section you type in your definition of done. So for example, a high quality SEO optimized humanized piece of content for this particular keyword. That could be the definition of done. Beautifully formatted, reads like a human for example. Now from there you can give it a starting point if you already have a blog that you want to improve. Otherwise you can just start from scratch and leave this empty. It's optional. And from here you can select which API you want to use. So we could use free APIs like for example N2 step 3.7 flash or we could use So, for example, round one probably not that good. Then it's going to improve step by step on every single round until it finally passes. It has to score 90% or higher to actually pass the judge. So, this is how it works and you can have a self-improving loop. Now, this doesn't just have to be for AISO. You could apply it to anything. But the system here is all about self-improvement and also having your agents run autonomously without you, right? If your agents can build without you, create without you, improve without you, well, then you're almost at a point where you've got AGI that's self-improving because it can just go off and do its own thing without you. And so, that's how these loops run and how powerful they are.
you cut out for Yeah, I know. I know. I did already mentioned that earlier.
I mean, if you train it to be It depends what you're trying to do there. But if you mean for example for just like becoming a version of you, yeah, you could train up to be like that.
Do you have a skill tool to connect to Hermes agent desktop?
Well, I mean it should just sync automatically, but that's why. Oh, so for that, if you want to search with this stuff, then you just you would plug in firecrawl. Firecrawl is a free API you can plug in for web search. Also, if you're using Olama models, they have web search built in by default. So, that's another option, too.
Now, another really interesting use case for this is that you can actually have video agents. You know, you can have videos that are generated with AI and self-improved as well. So, here's an example of my AI avatar talking about loop engineering and how it works. And [snorts] it's basically breaking down the whole process. But the way that we generated that video itself, is by having a team of agents working together, one for script writing, one for creating the video, etc. And they work together as a team, and the judge checks a video. If it's not good, it has to loop around again, selfiterate, and self-improve until finally it's actually good. Right?
So, you get the point. like it can create amazing things because you've got that quality control place and system in place and and that's really what makes a good SEO agent or an SEO agent in general.
So with this as well, you stop being the loop, the machine becomes the loop. So an example of how this works in reality is like you know in reality most people using AI for SEO they ask for a draft they read it it's not good enough you type what's wrong you try again you read it again round after round you give feedback and you're the judge you're the notetaker and you're the one in the middle every single time which is quite energy draining when you use this system that I'm showing you right now the self-improving SEO engine takes you out of that chair because you write what done looks like a builder model drafts it a separate judge grades out of 100 and lists every floor.
The notes go back to the builder and it loops round by itself until the score finally passes your bar. So the builder never grades its own homework. A cold adversarial judge does. And that gap is exactly why we got really good outputs with the video that I just showed you and also the blog post itself. And this is great for for automating almost anything with this whole system.
Now you can run a self-improving loop inside the agent OS. Here's how it works and this is the first method. There's two different methods for this. So inside the agent operating system inside the AR profit link in the comments description or go to the profit.com.
It's a text box not code. You just type what done means pick a builder and judge and set the maximum number of rounds and hit run. And then you close it up. So an example of this run it right here. Now can it grade videos? Yes, it can. Can it grade SEO content? Yes, it can.
Anything, any sort of project that a judge could check.
This method works.
So, for example, we gave it one job, write a publish ready SEO blog targeting Hermes agent self-improvement loop. A cheap builder drafted it. A separate judge, which was GLM 5.2, graded it adversarially out of 100. He eventually scored 92 and passed. And that draft is a live blog post you saw just a minute ago. Right? So if you're thinking AI grading AI is like kind of rubber stamping itself only if the same model writes and grades. The judge is a different model and it's told to be adversarial and find problems with the code. So whatever you're creating here it's designed to critique and review it.
Now there's another way to do this as well which is Hermes cambban boards. Let me show you an example of that. So method number one was the loop engineering system we've got over here.
Method number two is that you could have a camb board like this. So for example, if we go over to the content system, you can see we have a blog post and we have a video generated over here. And this was with a content judge, a video director, a content editor, and a team of separate Hermes agent profiles that could look at the content, selfiterate it on a loop until it was finally good.
Now, how does this work? So essentially this scales the same idea but to a team.
So the camb board runs many Hermes profiles at once. You have a researcher, a video producer, a judge profile that grades the work before it moves to done.
So these are all separate tabs. You got triage to-do, ready, running, blocked, and done. And that's how it basically works. So you can drop a goal into triage. The orchestrator breaks it into tasks and assigns them across profiles.
The same loop principle applies. So nothing reach is done until the judge passes it. And this is the board, the research, the keywords, drafted the SEO posts, produce the explained videos you've seen today. Right? So number one, method number one is one loop for one piece. Number two is a loop running across a whole content team. It's the same engine, but it's a different scale.
So you have the orchestrator that goes to the researcher, the writer, the video producer. The judge looks at the content quality. If it's not good enough, it loops around. If it is good enough, well, then you get the video and the blog post shipped as you can see right here inside this system.
Now, there's basically five parts to this in terms of a self-improving SEO engine.
So, you define what's done. You write what a great result looks like plainly. You pick the model. So, a builder to write a different judge to grade. That could be a cheap or a free builder or a sharp judge. You just never want the same model for both. You can walk away. So you run this, you walk away, you set the maximum number of rounds, you hit go, you come back to it later, and then you get the graded results and logs, and then the winner finally ships. So it publishes. That could be to your blog, to your funnel, etc. You might say this sounds technical, but loads of people inside the profitable volume are building stuff like this. So I know that if they can do it and I can do it, you can do it, too. And so if you look at this system, the old way is like you ask AI for a draft, you read the whole thing, you swap what's weak, you type out every fix, you paste it again, you read it again, you repeat that five times, you lose focus and energy every single round, you settle for good enough because you're pretty tired, and then you have no record of why it got better, so you start cold the next time. With this loop, you turn something that used to take an hour of quality control into five minutes. And so you write what done means once you pick a builder and a separate judge. You hit run. The judge grades out of 100 and the builder fixes it on its own. And you come back to a past result. And every round is saved to your Obsidian Vault memory system. So what we have over here inside the memory here is that you can see that we have all of our logs with our agents plugged into this system so that we can come back to it later and see what we've created. And that's super useful because then we can easily find okay what's working, what's not working and also our agents can read from that and learn from that. And so the old way would take like an hour, the new way would take like 5 minutes. And you might say, well, that must, you know, require a lot of tokens, but not actually because you can run a free or a cheap builder with a free judge and then you're just using free APIs for everything. Now, if you want both loops ready to run, the loop section and the camb boards are part of the agent operating system inside the air profit boardroom. We've got one dashboard where Claude, openclaw, Hermes, you all share one memory. So, every loop already knows your business, your client has your voice, right? and the full agent operating system with the loop section, the camb and every model wired in is inside there. You get pre-built setups. You get full coaching calls a week and daily tutorials as new models drop. And you can have a 30-day road map plus everything else you need to win with this stuff. Now, let's talk about beliefs that might be holding you back. You know, some people say, "Well, I just need to learn to prompt better and then I can do it myself." But a better prompt still puts you in the chair for every round. So, the win isn't a better prompt. It's not being in the loop at all. That's how you save time.
Other people say, "Well, my work is too nuanced for a machine to judge." The machine doesn't decide what's good. You do when you write the bar. The judge just checks the work against your standard every round.
So, you learned how to stop reading drafts when it comes to SEO. You learned how to stop settling with mediocre work from your agents. You learned two different ways to run the judge. So you can have one loop which is a loop section or you can have a team with a cam bound and judge. You saw how it can publish itself. So we actually gave it the netifi API and then it can publish directly to our website. So you don't need to log in. It's free to do if you use free APIs for free models and I've given you examples of that today. Like for example N2 is one. North mini code is another one on open router. You could use step 3.7 flash with Hermes as well.
And then you stop being in the loop. You send the bar. The machine earns the pass. So that's basically how the whole system works. You can make your content grade itself and every piece you publish from now on can be drafted, graded and fixed by a loop before it ever reaches you. The agent operating system inside there has both ways ready to run. So you get the full agent OS zip file, the loop section, the camb, every model that already plugged in there, the setup walkthrough done with you step by step, four weekly coaching calls, daily tutorials, a 30-day road map, 3,600 members inside here. It's just an awesome community to learn and win and grow from with AI automation. Also, the agent OS can do like so much more. But that's just an example of this. So, you get it inside the profitable link in the comments description or go to the proferborn.com. Inside the community, you get help and support inside there. I personally answer the questions inside there. You get access to all of my best trainings inside the classroom inside the calendar and you jump a weekly coaching course. If you want to get the agent OS system, you can get it over here. If you want to learn how loop engineer works, we have a full tutorial and guide on it here. here if you want to learn more detail about that. And you can also connect with people in your local area who are building with stuff like this as you can see. And that's all inside here. And this just a great community for connecting with great people and learning and growing on a journey together.
Let's see what we got. Hermes uses obsidian as a storage for its memory vault. So when agents log decisions, context or conversation history, that data lands as structured markdown files.
That's great. Do you prefer LM plans? I mean for me my favorite is of course Claude. Claude is my absolute favorite out of everything.
If you create a website, another agent needs to view it. No, you can just use a HTML file, right? So it's just code which LL models can do that. Uh so for example in this situation we use GM 5.2 but you could use claude as well.
Again when you're when you're creating a video even that's a HTML right so your agents can read code they're very good at code and so like they can look at the the quality of the outputs and see if it's good or not.
[music] >> [music] [music] [music] [music] [music] >> Oh, [music] hey. [music] Heat. Heat.
[music] [music] [music] >> [music] [music] [music] [music] >> Yeah. So, basically this is using something called hyperframes. There's two skills you can use. For me personally, I like to use hyperframes.
You can also use reotion as well. And basically, you can create videos using HTML with your AI agents. And that means your AI coding agents can understand them, they can build them, they can create them pretty quickly. So, it's a free open source skill you can give to your AI agents and then they can create videos for you. So, that's you know, uh, how the agent system over here works with the video agent. So if we look at this example, this whole video was written um with hyperframes and and you know edited together and it looks pretty awesome.
And then for the B-roll we used Grock and this whole system puts it all together.
just add a new update to the agent OS system a bit.
>> [music] [music] [music] [music] [music] >> Today We're going to be running through the latest questions I've had about our agent operating system. If you're not sure what that is, basically this is a powerful system for having all your agents working together. So you can see for example, we have paperclip, we have the AI agent mastermind group chat, we have apply pillow where we can go from idea to implement this ASAP, agent camb, we even have for example more advanced stuff like a local engine for AI agents, a loop engineering system here for self-improving AI agent loops. We have a SEO section and a video agent here that can create amazing stuff. So this whole system is about making sure that we have all of our agents in one place, which is more important than ever. As for example, more models get released, more updates come out. Doesn't really matter about the models that come out or the agents that come out. What matters is that you have a great system to manage it all, which is what we've built with the agent operating system. And you can build your own as well, right? And you know if you look at the old way versus the new way it's the same person same goal but two totally different workflows. So for example before you'd be juggling loads of tabs copying and pasting between chat to claude and all your other models. And then for example it gets messy because you got all these different subscriptions in one place. You got random stuff that you create but you can't find it anymore. There's no memory system and you start from scratch every day, right? And you feel like AI is making you faster but the day actually ends up with the same to-do list. That's the old way. The new way is like you have one tab, everything built in, all your agents ready to go. You don't need to go to terminal or anything like that.
All the workflows you actually build out dayto-day, you can save in one place. As you can see here, we've even got a music agent, a game studio, etc. We got notebook 11 and a research agent here that can generate research reports and infographics and podcast in one place as well as videos. We have a full memory galaxy here where like every few minutes we're getting new memories built into the system with our AI agents. So, we've got this amazing memory and context our agents can draw from and they can all plug into the same system so they're all synced up and we never have to reexplain ourselves again. Literally, I've I've not had that problem for months now since I built this. So, this is a powerful system and we're going to answer some of the latest questions inside the community. This is the profit born community where we share our agent OS setup and basically we answer questions inside here because I know that if people inside this community have those questions, you probably have similar sort of questions as well. So the first question we have from Victor which is you know I'm interested in everyone's setup what are you running how are you running it etc. So we have uh you know his setup is like open core on a Mac M3 with 64 gig RAM using local LM setup of quen 3.6 which is pretty awesome a MacBook Air with 8 gig RAM and Hermes and then a Surface Pro 6 with co-pilot using it for like IT work development work and some marketing.
What am I using personally? So, I actually have a Mac Studio set up with Apple M4 Max 36 gig of memory. And the one thing I'll say here is like it's actually not that great for running local models from what I've tested before, but it is amazing for having this operating system, right? Cuz this is just running 24/7. And you can see here that we have claw, we have open core, Hermes, and all these other CLI plugged in. Now, for me personally, I don't think that you need all of these different CLI plugged in. You can just pick and choose whichever ones work for you or whichever ones you're using for.
And I would say, you know, the best systems are typically the simplest systems. So, even if you just have like Claude and Hermes plugged into this, it will be very powerful. And then you can orchestrate all your agents together using something like Paperclip, which you've got over here. You've got the AI agent mastermind group chat as you can see. And then any ideas that come from the group chat when all my agents are taken together, anything that's actually good gets plugged into our pipeline and then we can start building it as you can see in one single click. Pretty cool.
And then we approve or deny the plan.
And once we've done that, it will actually start building like so. So you can see that's just gone from human approval to implementation. So we can get ideas created very quickly running in the background. We have a local AI agent cambbound as well ready to go. And that's the system that I have.
The thing that I would say as well about this system is like because we give it to you, you don't need to worry about the latest updates or what comes out.
Like we build it in and we test it every single day. I actually have something called Goldie Bench where I benchmark all of my live demos. We see which one performs the best, which models are the best recently, and then we actually give them medals based on the tasks they complete and what they complete. And if you want to compare like, okay, what the outputs for each model. So you can see the stuff that were built side by side for each model over here, which is pretty cool.
And then also you can see the individual notes on what we did and what we built with each model. So I test all this stuff for you. So you never need to worry about, okay, like what you know, what are the latest updates, what's come out, is it actually good or not. You can just see it all here and test out for yourself. If you actually click on some of these examples, you can actually uh you can test out and play the stuff yourself as well. So, you can test out everything I've built and see if it's any good and see if you actually like it or not.
So Jose was talking about a new app called Dictate and basically uh so this is called Handy. Sorry. It allows you to like basically dictate text into any app using AI. So you just hold a button.
That sounds pretty cool. The one that I actually personally use is Whisper Flow. That's pretty good as well, but I think there's so many of these coming out that you can just pick and choose whichever one you want. I will say it's a little bit slow sometimes, but most of the time it's very reliable.
This is pretty interesting. And so Joe was posting about like his experiences with Hermes, what he did to fix it, all the problems that he had.
I think this is super interesting. You can see for example here he's talked about all the different problems that he had step by step from Hermes. And this is one of the things about open source projects.
So if you're not familiar, for example, with Hermes, Hermes is a great system.
It's a it's a really powerful agent, but it's open source and it's it's not as smooth as maybe some people would like it to be. I think if you want something really smooth and easy to use, you can just use something like Claude. This is how we actually build the agent OS. So, we use Claude directly and then that builds out everything into the agent OS system. And I will say that's a lot smoother versus something like Hermes, but Hermes is great to play with and also it's great for running like schedule tasks and also having more customizations. So for example, the customizations we can have with Hermes are really cool in terms of like we can have a live chat with it. We can have Hermes Jarvis which is a voice activated AI agent that can control my computer.
We can preview everything that we built with it. We can see all our previous conversations. Also have a studio here where we can generate videos, images and voice with Gro or miniax on the coding plans plugged in. And then we can see everything that we've built as you can see right here which is pretty cool in itself as well. So everything in one place where you can just view it and you know come back to it anytime. I think that's fantastic. Is it as smooth as something like Claude?
Absolutely not. And I think that's a trade-off with open source projects.
I think it's great that you also document this process because if you learn from your mistakes and you document how you got over each one, well then you're improving yourself and your skills along the way.
And you build up a great system in terms of making mistakes, but then also learning from them later.
Tron just joined as well. He said, "Hey guys, new here. Got to say, taking the six week beginner course. Did it in five to six hours today, which is pretty insane. Uh, finally understands how the industry works and been trying to figure out for a while now. He was so overwhelmed with all the new stuff coming out, but it's simple. understand the basics and then add the new stuff where it actually belongs. And I think that's a fantastic philosophy.
So what he's talking about here is the sixe course that we had for learning AI automation and it's so true like you can over complicate AI but most of the new models and the new updates that come out you actually don't need. They're just fun to learn about. But if you really want to stay focused, I think one of the best ways to approach it is you just focus on one thing at a time and you look at, okay, where am I spending my time and then how can I implement and automate that. So great to see you here, Tron.
Thanks for sharing.
Now, this is a great question. So Michael actually set up the agent OS with paperclipip and it's pretty straightforward which he liked which is great to hear. Super happy it was simple to install. So inside the paperclip here he's asking like okay how do you reduce the amount of tokens you use. So there's a few ways to approach this. Number one you can use free models like for example Hermes which is free and open source and then you can plug in step 3.7 flash into it which is a free model. You can access via Hermes via news research portal.
Option number two is that you can use free APIs. So for example, N2 on open router is free until the 22nd of June and there's always new free APIs coming out on open router that you can use.
Option number three is that you can actually get local models and run local models depending on your setup. So those are three different options that you can use to reduce the amount of tokens and resources you need to build out stuff like paperclip.
or just a combination of all three.
Now this is a great question from Wes and Wes is asking like you know how do you what are the use cases for an agent operating system like why would you use it how do people use it dayto-day etc. So the way that I look at this is you want to analyze okay where are you spending your time and then what do you need to automate with the existing setup you have. Now the obvious use case of course is like having everything in one place where you've got Claude and Hermes and your memory layer all plugged into one system so that you don't need to switch between tabs. That's an instant times saver.
But then also you can build out new workflows based on what you spend your time on. So for example, inside the agent OS system, we actually have sections for a video agent and an SEO agent because that's what I used to spend a lot of time on. And so if you can automate the stuff that you spend time on and then put it into a workflow inside your agent OS, that's a huge win as well.
The final thing that I would say is like you don't need everything inside there.
You just need the stuff that you're actually going to use. And when you set it up, you could actually ask your agent to remove certain sections if you don't use them and you want to simplify it.
And that way you keep everything simple, but you customize it exactly how you want it.
Alex is asking about Mac Mini versus Mac Studio for local AI agents. What's the best stack here? And for me personally, I've not found local models that great on a Mac Studio or on a Mac Mini. The things that blew me away in terms of local models have been on a RTX or on a Nvidia Spark. So for example, there's a member inside the profit volume actually created a AI avatar video using a local setup.
And you can see that example right here.
So he actually created, this is Daniel who's an absolute legend with his local setup using the RTX5090 with 32 GB of uh VRAM. He generated this avatar video completely locally, which is mind-blowing. Like if you look at the quality of it, it looks like Hey Gen, but this was generated locally.
And again, just to recap, he used the RTX5090, 32 GB of VRAM with 120 GB 128 GB of RAM and 4 TB of SSD.
That's probably a much better setup than using a Mac Mini or a Mac Studio.
So, we got a question here from Farah about, you know, overwhelm and how to stay focused, especially when there's so many new things coming out and there's so many different options. So the first thing that we actually teach inside the the classroom here is how to avoid and never worry about overwhelm again because it is the number one thing that I've seen every entrepreneur struggle with particularly in the AI space.
So if you're worried about overwhelm or you're not sure which direction to go, I would recommend checking out the focus protocol we have below.
The main thing to note is like you don't have to do everything and whatever you implement you're already winning. So just focus on one thing, identify what you need to work on to start building that one thing out and then go from there.
The more you simplify, the more focused you can be and the more you achieve.
This is pretty amazing. So, Jose has already built out his big agent OS. So, he said, "One day here and already have a big operation system. I'm happy to announce that finally I have an Aentic operating system that's ready to go.
Great content, great community." So, shout out to Jose. That is amazing. And if you look at his setup here, he's basically taken the system that we have, but then fully customized it and also he's uh converted it to Spanish as well, which is pretty cool. So you can see here, for example, he's he's basically got the system from us, but then customized it with his own setup. So you can see his version here versus our version, right? Quite different in terms of the UI, but that's because he's customized it. So number one, well done to you, Josie. That's super inspiring to see and well done for setting up so quickly. Like within one day to have a full agent operating system ready to go is absolutely amazing. Especially when I've seen it so customized here. I really appreciate the positivity as well.
Also, some people look at this stuff and be like, "Ah, you know, agent operating system, that sounds technical."
Actually, it's pretty chill to set up, right? Because we've optimized the process to be as easy as it possibly can be set stuff like this up. So, You know, I've seen so many people who are non-technical build amazing stuff like you can see right here. Loads of people setting up their own Asian operating system setting up amazing stuff. So, you know, if they can do it and we can all do it together, then we can all learn and grow together inside a community like this. So, it's fantastic to see. Really inspiring.
Alex was asking about how to automate Substack with AI. This is something we're experimenting with right now.
Honestly, I wouldn't say it's like mind-blowingly good, but we have grown to to this amount as you can see.
Um, we started from zero, so we've got some nice growth on that. Now, if you want the full setup on how to do that, and we actually have a computer agent that does this.
And it runs on a daily schedule.
So if you want to get that set up, we'll give you the skill MD file there and then you can just customize it to your setup and your situation.
Another question we got was about custom agents and building them on a VPS. So, how do you do that? So, if you just want like a basic sort of Hermes agent on a VPS, you can use something like Hostinger. Now, if you want to build a full agent operating system, John actually did this here as you can see.
So, he built the agent operating system on a VPS. It has mobile and desktop access. And so, you can see an example what I built. It's got a full tutorial on how he did it. But, basically used Hostinger with Cloudflare and now it can work on any device, which is pretty insane in itself. So, we'll just post a tutorial right there for you.
I personally don't use VPS. I just run everything locally, as you can see.
So, we got a question here from Daniel, which is like how to set up a voice agent with Hermes. How do you do that?
So, there's a couple of options here.
Number one, you can use Grock and Grock plugs in as oorthth into Hermes agent and then you can use that as voice agent. So for example, if we go to Hermes here, we can have a live conversation with Hermes Jarvis and also the talk section here and then we can just have a kind a voice agent with Hermes Asian directly. Now if you want to set that up, we've got a full tutorial on it as you can see and we've got like prompts and a full setup guide as well with all the layers right there.
Basically, you know, to to simplify as much as we can, what you do is you just go to Hermes and you say, "Hey, I want to use Grock or Miniax. I've already plugged in. Can you set up the voice with that?" And then you can go from there.
Option number two is you can actually use 11 Labs. And with 11 Labs, you can actually call your agents directly. Now, for me, I set this up for fun and then I shut it down because obviously I don't want people calling my Hermes agent. Um, that wouldn't be very safe. But if you want to set this up or if you want to learn how to do it, we've got a full guide with step-by-step instructions for being able to speak on a live call through your phone with Hermes agent right here.
And that uses 11 labs.
You can actually just give the documentation from 11 Labs which you can see here and then that will plug into Hermes agent once you give it the API.
And the final setup is that you could actually just use the agent OS because that has instructions on how to set up the voice directly inside the zip file.
So you can just use that directly. So there's three options there for voice agents with Hermes. Number one is Grock and you ask Hermes set up. Number two is 11 Labs and you ask Hermes set up. We've got a full guide on that. And then number three is you can use an agent operating system and just speak to Hermes directly inside the talk section here.
Now, Elizabeth is asking, okay, you know, if you're getting a new laptop, what could you use with an agent operating system? So, for the actual laptop setup, it's pretty lightweight for running agent operating system if you're not using local models.
So, for example, you know, we have a a MacBook 3 Pro and that runs the Agent OS with no problems.
To set up WhatsApp or Telegram with agent OS, you would use Tail Scale. I don't use that personally because I just want everything kind of sandbox locally, but you could set that up with tail scale.
and they have the full documentation on the website.
All right, so that's all of the questions answered today. Some fantastic questions. If you want me to answer your questions like this in a live video tutorial, then you can join the air profit boardroom and answer questions like this every single day inside the community.
Inside the classroom, you get access to all my best trainings, learnings, lessons, etc. We actually have a full agent operating system you get here with this is updated daily. So, you can see the last update date here. You get the video tutorial guide to set it up and then also the resources that you can install. So, it's just a zip file. And then when we add new daily updates, we add video tutorials with everything you can use step by step inside the system. So, whatever you're looking for, you know, if you're like, right, I want to use GM 5.2 inside the Agent OS, here's exactly how you can do it. We have a full guide plus a video tutorial on how to set up.
Inside our calendar, we have four week coaching calls where you can wire this stuff together, ask questions, share your screen, meet other people running agent operating systems. inside the map.
You can meet people locally who are building with AI agents like you see right here. And this is all inside the AI profitable broom link in the comments description or go to the arprof.com to get access. Thanks for watching. Cheers.
Bye-bye.
All right. So, I think we've answered every single question, every single comment here. Thanks so much everyone for being a part of this.
I will see you on the next one. Cheese.
Related Videos
AI Agent Mastery Certification Course: Lab 4 – Tools & MCP
arizeai
350 views•2026-06-16
Real-time Voice cloning, Kimi K2.7 CODE, GLM 5.2 and 3D reconstruction | AI News
kaiexplainsYT
111 views•2026-06-16
He Believes AI Could Replace Humanity Faster Than Anyone Expects
LondonRealTV
815 views•2026-06-15
General Session by Rami Rahim-The next generation of networking: From vision to self-driving reality
HPE
108 views•2026-06-17
[PLDI 2026] Flatirons 3 - LCTES (Jun 16th)
acmsigplan
191 views•2026-06-16
Google DeepMind’s AI Halves UK Housing Planning Time
60secondsignals
467 views•2026-06-17
The Creators of Claude Code and OpenClaw don't Prompt Their Agents Anymore?!
ColeMedin
569 views•2026-06-18
Why prompt injection is AI's biggest fail
usemultiplier
1K views•2026-06-17
