## Intro [00:00] So, I spent the last two weeks building hundreds of different AI agent workflows, mostly using OpenClaw, but I also use Manisclaw Code and even Perplexity Computer, which just came out. And my biggest realization through this process, companies are going to have very narrow AI agents that operate in a team. And my current plan is to build 15 highquality AI agents that run our entire growth division here at vibco.dev. And so I want to take some time in this video to explain why I believe that narrow agents are the future and we'll also kind of talk about why I'll be using OpenClaw for this project. And so let's just dive into the video. Over the past two weeks, we tested many different agents and the main four that we tested were OpenClaw, ## Perplexity Computer and Manus [00:48] we tested Manis, we tested Clawed Code, and we tested Perplexity Computer. And so perplexity computer is actually really interesting. You see here you can actually switch from search to computer. And the way perplexity computer works I can say please make an app. This right here is a single task. I can give an AI agent and just like chatbt it will start to work except this AI agent will get a sandbox which is just a computer that's running in the cloud and it can actually create files. You can see here that whatever it creates can open up in this side panel right here. So it's like chat GPT with a computer which is exactly like Manis and Manis has been around a lot longer than Perplexity Computer and it operates the same way. Manis was the first kind of general agent tool that was released that had every single task that you put in has access to a computer. And as you can see here, you can see view Manis's computer. And so we can view this over here. And so you can see that this AI agent comes with a computer. It can create files. It can edit files. And it can do many different things that you would do on a computer. And so that's how Manis and Perplexity Computer work. You enter a task, it spins up a computer and depending on how you prompt that task, different things will happen on that computer. It can create different things. It can do a whole host of things. And so if you run five tasks, each one comes with its own little computer. And so this can be seen of more as like a command center, right? This is a command center for agents that have access to a computer. And this is cool for certain things, but it's actually not what we want. I don't believe this is going to be the most useful form of AI agents. I think the ## OpenClaw [02:43] most useful type of AI agent will be something exactly like OpenClaw. Openclaw is an AI agent that runs on one computer. And you can see that uh Mac minis are literally sold out right now. It's really hard to get a Mac Mini or a Mac Studio because so many people are running an AI agent, OpenClaw, on these computers. And so basically what OpenClaw did is they basically put an AI agent on a computer and then they gave it really good memory. They gave it really structured skills that you could very easily add. And then they also added a gateway. And this gateway allowed you to chat with OpenClaw from different applications. You could do it on Telegram. You could do it on WhatsApp. You could message it on Discord, on Slack. And this is why OpenClaw went really viral. It gave an AI agent a computer and then made it accessible in all of the tools that you already communicate with other people. And so the first openclaw AI agent that I created had many skills. So I added ## Too many Skills to my first AI Agent [03:52] and so this is an overview of the most useful skills that I gave my first AI agent. Um my favorite skill was this social media transcript analyzer using an API called Supera Data. If you guys want to look it up and use it, it allows you to turn any YouTube link, uh, Twitter link, Instagram link, or uh, Tik Tok link into a a transcript. So, you could very easily analyze social media. I added this. And then I added the ability for it to control my notion. And then I added the ability to control all of my Google workspace. So, my calendar, my email, Google Docs, Google Sheets, etc. And then I gave it access to our linear so I could take a look and see where are we at with the product. Uh what's launching soon, things like that. I gave it access to Figma. It could literally control Figma on my computer. It could generate any type of media using FAL. It could even edit videos, which was my previous video. And what I realized over time is that the more skills that I added, right, as the amount of skills increased, the dependability of the AI agent decreased. And so that's when I realized, okay, well, you can't really add unlimited skills. It stops being super useful. It doesn't use the skills at the right time. Uh the context gets super clouded and it doesn't use the right integrations and the personalities ended up getting jumbled. And so that's the conclusion that I came to, right? We need to create a team of AI agents that have, I would say, 7 to 10 skills each instead of building out AI agents that have 30 skills. This is the sweet spot. As you go above this, the AI agent stops performing super well. And so that's one of the main reasons why I think Manis has a lot of potential. It's just not my tool of choice. Because when you hit use skills, you can go to the manage skills and you can kind of see all of their official skills. And so they have all of these different types of skills that you can add, but you're not adding it to a specific agent. Rather, you're adding it you're adding it to your command center, which means everything is proactive. You have to go to your command center and then ask it to do it. People simply want ## People want an employee [06:09] an employee that gets things done. And if you think about a really good employee, you think like the employee will actually just like do things that surprise you. A good employee will make suggestions that are useful. And I believe that in order to do these three things like get things done, do things that surprise you, and make suggestions that are useful, you need to have specific goals or you need to give AI agents intent. And I actually got this from on Twitter. EMTT Shear. He was the interim CEO of OpenAI when Sam Alman almost got fired, but he tweeted, "Prompts are so late 2025. We are giving models intents now." And I think I I would say like we are giving AI agents with computers intents. And the definition of intent is intention or purpose, right? We are giving these agents purpose. And I believe that if you're going to go through this paradigm, it's really hard to give these agents purpose because they're so general. They have so many different skills. It's super proactive. Uh and so that's why I don't really like perplexity computer and manis. And so that's why I want to create a team of narrow openclaw agents with very specific goals and skills. And I'll explain a little bit more about why I ## Testing Narrow AI Agents [07:35] want to do this. And so when we were testing these AI agents and I have a bunch of these AI agents running. This is my journal bot. And uh I have these agents running in Telegram right now. And after testing them for two weeks, I realized that this is the way this whole space is going. A focused agent with a specific personality with specific tasks and a a specific heartbeat. And what I realized is that all of these things are better when they're focused, right? when there when you have a team of AI agents or a team of agents running on a computer that are confined right to a more narrow focus everything performs better which allows you to focus your skills and integrations on what will be useful to reach a specific goal let me give you an example so my favorite agent that I'm using right now that I message in Telegram is this content bot which is specific for creating YouTube videos. ## Narrow Agent Example (YouTube Agent) [08:33] This is my YouTube agent. So, this focused agent is my YouTube AI agent. And the only thing that this agent is focused on is creating YouTube videos. And it has three goals in its files. It knows exactly what goals it's optimizing for, which are subs, views, and conversions. And I'm not to say that everything that I do on YouTube will uh is optimized for these three things. But whenever I ask it to create a script, for example, it knows that I want to increase the amount of subs, increase the views, and increase conversions. The main reason really narrow goals are super useful is that it allows you to create hyperspecific skills that can be verified whether you should add them. If your skill does not have anything to do with your with your goals of your AI agents, you shouldn't add them. So, it makes it super easy to add skills. For example, the main skill that I use for this YouTube AI agent is YouTube research. And then this allows me to think about, okay, what integrations are useful for this YouTube research skill? And so, I use two for this. I use the SER API skill uh integration and I also use the super data API. This one allows you to scrape transcripts. This one allows you to like search through YouTube. It's really useful. Skill number two is thumbnail generator, right? I generate thumbnails and I even have my AI agent every single morning scrape my competitor's thumbnails and then it comes up with ideas uh and kind of modifications of their videos with my face. And so for this we need access to Nano Banana. And for this skill specifically, it it has some relevant context that it needs, which is uh photos of Riley, right? It needs my photos in order to create an image of me. So that's just some useful context that we need to give it. And number three is it needs to be able to uh control my notion, which is where I keep all of my scripts for my YouTube videos. And um for this, we actually just need the notion integration. And so you can see here that it's kind of this direct path, right? Your YouTube agent is in charge of optimizing for YouTube subs. Uh it wants to increase the amount of views you get and it increases the amount of conversions you get from your video. That is what my agent is optimizing for. And now when I go to my agent and say, "What skills do you need?" It knows exactly where we're going. Right? This is a path. And if you've spent any time hiring people, the most annoying people to hire are people with vague skills. They don't have specific goals. They're good at talking, but they're they're not good at driving towards a specific goal. They're good at distracting from the goal. The best employees to hire are people who are like, "Yep, I'm really good at certain things, and I can help your company reach these goals, which will ultimately help your company." It's very simple. ## A Team of Narrow Agents... Why? [11:32] And this YouTube agent is just one of the agents that I use. And so the question now becomes why narrow agents. The first reason why is when we find a super useful agent, it's very easy to duplicate, right? You can remix it for something else. It could be relatively simple to turn a YouTube agent into a Tik Tok agent, right? We could create a Tik Tok agent that's only focused on Tik Tok. We could create one that's only focused on Substack, for example. And when you create these smaller agents, it's easier to duplicate. And you know, when you try and create a massive agent that has like 50 skills, it's just hard to extract just the the portion uh that you want to duplicate out of it, right? And so I can very easily duplicate my my single um narrow focus agents. Additionally, um this makes it super easy to share with your team. So, for example, today I built a journal agent. My journal agent is lives in Telegram. And this agent is a little bit more hands-on. And so, basically what it does is it reaches out to me every 30 minutes. Sometimes it doesn't reach out if nothing needs to be done. And it analyzes everything that I do, every meeting, every video that I make, everything. And if it wants more context, it'll ask me every single day. It'll write multiple journal entries logging everything that's useful. um and everything that it needs to know about me just so I can create this like running log of all of the important information that's important to business and content. And the purpose of this agent is that it informs all of the other agents, right? So this journal agent has access to notion. Every single agent that we have has access to notion and all of these other agents are aware of the journal that the journal agent creates. So, my email newsletter every day is just going to read my journal agents journal and it's going to come up with ideas for email newsletters. So, in my journal agent, it'll know when there's product updates. And then my email newsletter agent will be like, we'll just draft up an email newsletter that needs to go out to our email list, which is 300,000 people. And in this email newsletter agent, uh it has very specific goals, right? This newsletter agent has very specific goals which is optimize the amount of conversions from our email newsletter, right? And and to maximize click-through rate, uh to op uh open rate, things like that. Um and it doesn't have to be clouded by any of the journal agents goals. It just has access to the journal that the journal agent creates. And so anyway, this is a really useful agent that I created and I want to help an my co-founder create more content. And so now because I made a really narrow journal agent, I actually just shared with him my entire open claw agent and he was able to duplicate it in about five minutes. And so it was really sharable. So when you create a really narrow agent that's super useful, it's very easy to duplicate it and share it with other people. And it also makes it just more understandable. Right? This openclaw agent runs on a computer. In the computer, it has access to files, right? like this computer. Every time you use OpenClaw, you are basically editing files, right? All of these skills are just markdown files stored in the OpenClaw folder. If your focused AI agent only has a few skills and and a handful of integrations, it's a lot easier to understand when you send them to other people. And then finally, right, this one's pretty intuitive, right? If you have a narrow set of of goals, right, goals, right? You want to hit specific KPIs. In the case of the newsletter, it would be like open rate, uh, subscriptions, right? You want people to subscribe, and ultimately like click-through rate, and like if you just had like a newsletter business, it would be like, you know, you could have revenue. When you have very narrow goals, it's very re reviewable. You can look at that agent and be like, "Yep, you did a good job." Or, "No, you did a bad job." You know exactly where it needs to change. You It's It's It's pass fail. When your AI agents are pass fail, it's a lot easier to just cut them, right? the a lot of AI agents that you create over the next few years are not going to be worthwhile. And the more narrow they are, uh, the easier it is to say, "Yep, you did good," or, "No, you did bad." So, get rid of it. And then the final two reasons you want to do a narrow focus is you can create easier loops, which allows them to be more autonomous. I have multiple narrow agents that are very simple loops, right? It only has a set of three tasks that it does every single day. It knows what it's optimizing for and it can just go in those simple loops over and over and over again because tasks are just are called cron jobs which are triggered at specific times during the day. And the more narrow your agent, the easier it is to get in a predictable loop and you can just let it run. If you have a super mega agent that's super massive, it's harder to do this. So, I think you understand my objective here. I want to ## In Summary [16:27] create very narrow AI agents with very specific goals and I want them to operate in a team. Right? This could be my team. It's hard to say exactly which agents I'll be adding. Right? As I do more workflows, I'll notice where we need to create a new agent. But the one thing that I think that Perplexity and Manis got right is they're actually using a computer in the cloud. They spin up a computer per task. Right? when you type in a task, they spin up a computer and that agent can use the computer. I actually don't think that's going to be the paradigm. I think it's going to be OpenClaw running in a computer in the cloud. And so it's up to us as a company to figure out how do we efficiently run these in the cloud and two years from now and each one of them has, you know, 20 agents, that's 200 AI agents. How do we efficiently run all of these AI agents in the cloud? Also, how do we share these AI agents with other people on the team? That's one thing that we really need to figure out. And then how do we get these AI agents to be able to communicate one another or at least share memory? And in future videos, I'll be talking about how to do this. How to get your AI agents to actually share memory so that like as one AI agent does something. I know that all the files are actually contained to that AI agent, but there's actually ways that you can actually communicate useful information to your other AI agents, very similar to how you operate in a team, right? the engineering team needs to communicate to me, the marketing team, on how to actually market the product. And there's actually ways that we can get AI agents to do this. So, that's the next few questions I'm going to be answering. Uh, but that's kind of what I wanted to share in this video. Narrow agents that run in the cloud, I believe, are going to win. I think that's what people are going to find the most use from, and that's what I'll be talking about over the next few months to run our company. I'll see you here for the next