NEW Claude Agent Swarms are INSANE!

All right. So, today we're going to be learning how to orchestrate Teams of Claude code sessions. Right. So, the new Opus 4. 6 update just came out and they've now created this sort of agent teams feature like agent swarms that you can use and these teammates can work independently each in its own context window and communicate directly with each other. Right? So you can have like multiple terminal windows from what I understand and these agents with multiple terminal windows with claw code running can talk each other right now these are not like sub agents right sub

agents run within a single session and can only report back to the main agent right what you're doing with this is you're having multiple different agents all working on their own tasks in parallel and they're working as a team and they can talk to each other right so agent teams are effective for tasks where paralle parallel exploration adds real value. This is interesting cuz like I really want to understand like what how would you use this and why is it different versus normal stuff. Sper says base 44 uses your content all the time. They know they I don't even know. Nahim says you're doing a lot bro keep pushing. Thank you very much. All right. So let's see what we got here. So you can put this inside the chat here. I think inside Claude and then it can begin to do this right. So we can say like create an agent team to review PR spawn three reviewers one focused on this that and this have them each review and report team findings. Trying to get my head around what this actually means like why would it be useful for me personally. So let's have a look what we got here. So for example ah this is

cool. So you say okay create an agent team to analyze my top 10 performing videos. Spawn three teammates have them debate findings and create a playbook document. Let's see if that actually works. So, if we take that prompt here, I'm going to say analyze my YouTube videos via and then I'll just grab my uh my YouTube channel here. This is going to be interesting. If we have three agents that can analyze my YouTube channel at the same time, I'm going to be using this a lot. So, let's see. We're going to open up a new window here. Open up Claude, right? So, you need to make sure you have Claude code installed. Got loads of training on that. And then we're going to type in this prompt here. See if it actually works. Dr. says, "Given the unique reasoning patterns of Claude 4. 6 and chat GPT 5. 3, how could Open Claw orchestrate them in a single workflow to maximize decision-m efficiency without creating conflicting outputs? " So, one of the ways I would recommend doing that if you really wanted to, I don't think you need to switch between them, but if you really wanted to, then you could actually just have sub agents, right? So you could have one agent via chat GPT codeex 5. 3 that's doing one task for example SEO content and then you've got another agent so for example Claude Opus 4. 6 that could be creating social media content and then you've got these sub agents that are working on different tasks and they can work on them in parallel because you've got multiple AP um you've got multiple agents and APIs plugged in. Flavin says, "Please figure out to make open claw be able to use this new system in any model, not just clawed. " Which new system do you mean? Yeah, short form says sub agents. Yeah, that's what I was saying. Is there any way to make codeex 5. 3 the main brain of openclaw? From what I saw today, um even codeex 5. 3 couldn't figure out how to put codeex 5. 3 inside Claude, whereas Claude opens, but I'm sure this way. I just, you know, I'm not a coder. So

let's say create an agent team to analyze my SEO content, right? And we'll try that. And then I'm going to take some content from my website and just see if it can work with it. So, we'll grab a blog. Take a rand blog that we recently published. Go down here. Copy that. Plug that in. Right to the terminal. Paste that in. Let's see if that works. Agent Swarms, he means uh agent I think it's called agent teams. That's what Claude call it, right? It's not agent swarms. I think it's a Kimmy K25 uh code. is, but we're using teams, right? So, if you look at the documentation from uh claw code, they don't talk about swarms, they talk about teams. Swarms is more for Kimk 2. 5. As you can see here, what's the usage looking like for Opus 4. 6? We'll check that out in a sec. That's a good question. Smashing a protein shake at the same time. There we go. Look at that. Already hit my limit. Already hit the limit. I don't know if the to be honest, like the usage limits have changed. It didn't mention anything about that from what I saw inside the documentation. So, so look at this. All right. So, this is cool. So, it says, I'll launch three specialist agents in parallel to analyze your content from different angles, then synthesize their findings into a playbook, right? And if we have a look here, we're running three different agents at the same time. So, we have this team of SEOs who are working on this, which is pretty really cool. That's a really cool idea. So you got SEO patterns analysis agent, the engagement metrics agent and the SEO semantics agent as well, right? And they're all running as one team that are working in parallel together. Right? You can see the actual layout of them here. So that is really cool. Let me have a look and see. Right? Okay. How what else would we use this for? So for

example, content analysis, competitor research would be really cool. Like you can analyze five different competitor channels at the same time. And you could create an agent team with five teammates, one per channel, right? So each can analyze in the same way, but they're working on different channels at the same time in parallel. And then you could get jobs done quickly and you've got these agents working in parallel together. It could be like a batch script review. So you have like 20 video scripts and then you've got a team with four reviewers, one's for SEO optimization, one's for hooks, one's for clarity structure, etc. Coding use cases as well. So reviewing um different parts of the code, handling the front end and the back end. That's quite a cool idea. bug investigation, migration and refactoring, uh documentation sprints. So, one reading API documentation, another one, uh writing, setup, and deployment guides, video ideas. This is cool. All right, so you could have like a team of three agents, one generates trendy, viral angle ideas, one generates evergreen educational content ideas, um and one critiques both of those, um sets, right? The only thing that I would warn you of to be careful of here is like if you've got three different agents working at the same time, then you're 3xing your token usage, right? So, it's quite a smart move by Claude because the number of tokens people are going to use or all of a sudden you can use like three or five times as many tokens at the same time, which is not ideal. And then also, we've got script optimization pipeline here too. One thing that I really like, this is

maybe my favorite so far. So we actually we use skill boss inside claw code and that allows us to generate images for thumbnails inside claw code. Now if we had an agent team to design three thumbnail variations that's cool that is a cool idea right so you can have one that designs high contrast text heavy style one that designs minimal face focus and one that designs curiosity gap mystery style right and each explains their psychological reasoning. Insane I love that idea. Love it. So, if you want to get a full guide on how to use these and some use cases for AI agents, cuz I, you know, the first time I heard of that, I was like, I don't know if I'd use it. But actually, when I look at those use cases, I'm like, that's pretty cool. I'm not going to lie. So, what I'm going to do is I'm going to put that inside the profit boardroom. And that's one of the most impressive use cases I've seen uh from Claude. I'll put the link to the documentation in there as well. There we go. And let's see how those agents are doing right now. So, we'll go back to the terminal here. See what we got. and says, "Creating multiple agents only provides the real benefit of honing in individual prompts and using most of the context window, but it also ends up using many tokens quicker. " That's what I was saying, right? The thing is, if you like the thumbnail one is awesome. Create like for me, I hate waiting for thumbnails. So, if I can have three agents that create thumbnails at the same time and I can just move like three times faster, then that's great for productivity, right? That's going to save a lot of time. Same for SEO content. Like I can create three different versions of the same blog in one go in a third of the time and I have to generate them anyway. That's really useful, right? So I can see why this would be really cool. People did mention he runs through tokens like crazy. Yep, I would agree. Teams cost extra. Well, it's more tokens, isn't it?

If we can use the same agent teams inside of open claw for any model, it would be powerful. Oh, this is a cool one. Cool idea from M. If you want self-developing and self-governing agents, you can create a performance coach agent and recruitment agent for hiring more agents. H interesting. I've never heard of that, but I never even thought about that. I don't know if that works. How can I avoid using all my usage tokens up? Any advice? You can switch between APIs. You don't have to use Claude, right? That's probably the best way. Um, I've seen some updates and skills that use less tokens per request as well, but I've not tested them. And also, I'm very wary about stuff like that because you don't, you know, if you don't know what you're doing or you don't know what's inside the skill MD file, then you got to be careful installing it, right? What else we got here? How can orchestrating multiple evolving AR models help people make better decisions in unpredictable scenarios? So, you could just ask Claude that, right? You could say like create three agents that would predict the outcomes from this situation. You have the teams working in parallel. This guy is insane. Short form handsets. So, yeah, that's basically it. Now you can see this takes absolutely ages. So it's still not come back to me. It's still right in the playbook. But if you

want the full setup on how we run Asian teams, you can get that inside the AI profit boarding right here. Link in the comments description. This is my AI automation community and you can get it profitable. com. If you go inside the community, you can post any questions you have. We always have people online. So, you know, there's always someone to help you and learn from and that sort of thing. You can share your wins, share what you're working on. We have a daily accountability group as well if you want to join that. And you can see that it's not just me winning with this stuff, right? So, for example, like Steve Light is crushing his YouTube channel challenge for AI. You can see that Abdullah built his first NAT automation. He's like, "Thanks, A profit boardroom. Appreciate that. First AI automation successfully completed. " You got CJ who installed NAN locally and learned how to do that, right? Loads of people winning and learning and growing with this stuff. Now, inside the calendar, you can jump on live coaching calls, get help, get support on the live call if you need help. And then also inside the classroom, you can jump on this course right here and it shows you how to go from beginner to expert with AI automation in just 5 weeks. Plus, you'll learn how to build your first AI agent in under 5 minutes. On top of that, you'll learn from our playbooks. So, we'll show you, for example, how to automate AI avatar videos based on what's working for me. How to screen share your videos with AI avatars too. How to automate your emails, how I automate my X, etc. Instagram. Everything is inside the AI profit board. Additionally, you can learn how to get more clients with the agency course. You will also learn how to rank number one with AI SEO over here. And also how to grow a YouTube channel based on what's working for me inside the sixe road map here. And if you can't make the coaching calls, no problem. You can watch them back right there. All right. So feel free to get that link in the comments description or go to the AI profitabroom. com.