I Can Actually Watch My AI Agents Work Now

All right, so here I am in Cloud Code in the terminal right over here. What you'll notice is that this agent is actually my Cloud Code agent. So when I sent off this message where I'm asking it to spin up two researcher agents, one to research Google, one to research OpenAI, and then create a visual diagram, you'll see that this agent actually gets that task. And now you can see that we have three agents actually running right now in parallel. This one's researching Google AI strengths, and this one is researching OpenAI strengths. And this is the main one that delegated the tasks to to those two researcher agents, which you can also see on this right-hand side. That's exactly what's being ran right now. And if I wanted to open up another terminal, so I add another agent, it plops in another agent right here in which I can have a different conversation with. Give me a quick analysis of my YouTube comment section from the past couple videos I've made. So now I shoot that message off to this AI agent in Claude Code, and it's going to pick up the task and start working on other things. As you can see now, this one is searching files and finding out what it needs to do. And you just heard that little notification noise, which means that this spun up a subagent and you could see that it was analyzing comments, but now it needs approval. So basically what we have here is a visual way to see all of the different terminals and subagents that we have working in Claude Code. We get notifications and we get visual things like this that tell us what they're doing. So I can see that those subagents on this left-hand side are disappearing, so I can hop back over to that first terminal and I'll be able to see that this first one is already done. and now the OpenAI one is still being worked on, which is this agent right over here. Okay, I don't know where she's off to, but you can see now that both of these agents are done. They're idle, they're done working. So in the first terminal we can see here's the visual. It came out clean. We have stuff about Google's edge, Google's gaps, OpenAI's edge, OpenAI's gaps. So here's that actual visualization that it made. We have strengths, weaknesses for both Google and OpenAI. Sometimes AI generation can be notoriously bad on text, but this one looks pretty decent. Only a few errors I can see. And then coming over to the second terminal, we can see the comment analysis over the last 2 weeks. So we've got some pain points about Cloud API costs, edit into Cloud Code, getting first client with zero social proof. We've got most requested videos, we've got notable patterns, and then we have recent video performance snapshot. But I'm sure you guys could tell that the point of this quick demo wasn't to actually show you the outputs. It was to show you what we were able to do here and how we could see these 2 different terminal agents spin up subagents and we could keep tabs on them visually. So unfortunately, I'm going to say goodbye to these agents so we can get on with the video, and I'm going to kill these terminal sessions right here so I can actually explain to you guys how this works and what I think about it. Okay, so the way that we are doing this is we are using Visual Studio Code, and that's where I've been using Claude Code as an extension. If you've been following my channel for a while, we're using an extension in VS Code called Pixel Agents, which basically turns your AI coding agents into animated pixel art characters in a virtual office that you can fully customize. So I'll have a link to this page right here, and I will have a link to the GitHub repository in the description, but I'll also show you guys like exactly how you set this up. So before we actually jump into that real quick, just wanted to go over like what are we doing here and what is Cloud Code? Because that's the little agents that we were actually spinning up. Well, Cloud Code is an AI in our terminal. It can write and edit code. It can understand our code base and it can run commands on our behalf. And what's going on when you're using a coding assistant like Cloud Code is you basically are just looking at terminal or you're looking at what the AI model is thinking and what it's doing. but what's actually happening behind the scenes is files are getting read, plans are being created, code is being executed, and tests are being ran. So there's a lot of stuff going on that we don't really visually get to see, especially when we start to spin up tons of parallel agents and have different terminals running. So basically what happens is Cloud Code is working, it creates an activity log that already happened by default, and Pixel Agents basically just picks up that activity log and then displays it in some sort of animation. It's honestly a very similar flow if you watch the way I set up my OpenClaw personal assistant, and I could see what he was working on and if he spun up any subagents. And like I said, it's pretty cool to be able to monitor our different agents at a glance rather than having to flip through different terminals, especially because I love to multitask when I'm using Claude Code. So one monitor I'll be coding with stuff and the other monitor I'll be working on other things and I can just kind of easily glance back and forth. So if you want to actually try this out for yourself, it's super easy. You just have to go download Visual Studio Code if you don't have it already. Unfortunately, right now there's only support for Windows. So if you're on a Mac or a different operating system, I don't think you can use this yet. This literally came out like a few days ago. But anyways, once you are in a fresh instance of VS Code, this is what it will look like. And what you'll do is on this left-hand side, you'll come over to extensions and you'll literally just type in Pixel Agents. And if you type in Pixel Agents, you'll be able to see right here, Pixel Art Office, where your Cloud Code Agents come to life. This is the one that I installed and I've just showed you as a demo. It's fairly new, but I am going to go over some security stuff near the end of the video. But you could read through this to understand what exactly this does. You basically have one agent per character. You have live activity tracking. You can design design the office with the floors, the furniture, everything like that. You can see speech bubbles, you get sound notifications, sub-agent visualization, and there are 6 different characters that could spawn. It'll go into some requirements and it will go into some things about how to actually use it, but if you want to test it out, then hop in here. And here is where unfortunately it says that it's right now Windows-only testing. And on the roadmap, you can see that there's lots of things they want to add, like having Cloud Code Agent teams, because right now it's just like parallel agents. So anyways, once you have installed that, what you're going to do is open up a folder. So let's just say you come to the Explorer and you open up a project that you want to be working in. So here I just opened up my Herk 2, which is basically like my second brain executive assistant type of deal. It's really weird, but if you want to get this project to work, your folder cannot have a space or a period in it. As you can see down here, this used to be called Herk Space 2.0, and then it was Herk 2.0, and now it's Herk-2. So make sure you don't have those things in there if you want the Pixel Agents to actually pick up the activity log and actually reflect what you're doing. You're welcome. I had to learn that the hard way. So once you're in here, you're going to open up the terminal. So if I'm on Windows, I'm going to do Ctrl+~ and this opens up the terminal. And then here is where you can see I have a Pixel Agents tab, which opens up the little workspace. Now what I want to do is I want to see half and half. So I basically grab this and drag it to the sidebar, which lets me look at it over here. And then all I have to do is click on +Agent. and that basically opens up a Claude Code terminal on the right-hand side, and it spawns in an agent right here, as you can see. Now, right now this agent is idle because it's not working on anything. So if I go ahead and say, please spin up a research agent and just find out what's going on with small businesses and AI adoption, and I go ahead and shoot that off, then this agent should pick up that task and it should also spin up a sub-agent. I'm gonna add another one that opens up another terminal, and I can add more and more agents so we can see that we could watch all 5 of these different terminals run at the same time. And then what you can do in here is you can go to your layout and you can change things up. So I could move around this plant, for example. I can move that right here. I could go to furniture and I have storage, I have tech, I have decor, I have all this kind of stuff. I could literally just put a monitor in the middle of the kitchen floor if I wanted to, and I could keep adding more and more. So it's kind of funny. You can fully customize the way that this looks and the colors of like the floors and the walls and things like that. And the other setting that you can really play around with would be opening the sessions folder to see what's going on, which opens up like a folder like this to see activity log and stuff. You can also go to the debug view to see what the agents are actually working on, and you can turn on or off the sound notifications, which is pretty cool. You could obviously use something like hooks to do that either way, but it's nice that that is now sort of native with this extension. So I'm not gonna spend too much time here really. I think you guys all understand exactly what's going on. You can see that this agent Just finished up. Great timing. I thought I should wrap up this video by actually just talking about like, what is the point of this? Is this just like useless? So my first point is about entertainment. I think that there is value to having this as just like kind of something that's fun. And the reason I say that is kind of twofold. The first one is I, a lot of times, like I said earlier, have two monitors up and I'll have Cloud Code going over here and I'll be working on other things here. And it is kind of nice to be able to look over and just see like how many agents are up or, you know, you know, do I need to load them up more tasks? But the other piece of it is I think that it's visual, and that might lure some people in who aren't super technical or who are intimidated by looking at a terminal. I think it's very similar to how NN started to blow up, is because they visually had some good elements that people felt like they could actually understand just by looking at. And that kind of leads into my second point, which is about that is exactly what AI coding is lacking, is visual elements. Because kind of like I said earlier, what we're looking at is a terminal or, you know, basically a a chat window when in N8N or something like that, you're seeing visual workflows, you're seeing the process of the data and you're seeing things be built or, you know, changed. Now obviously Pixel Agents isn't that, it's not a representation of what's they're actually doing. It's just like showing you how many are working. But I definitely think it's a step in the right direction. 'Cause I think ultimately where I'd love to go is to a place where Claude Code builds me, you know, workflows and automations and it's showing me it building them. In an N8N-style interface where I can see the flow. And then we can talk about multi-agent monitoring. There's a lot that can go on with different agents, whether you run them in parallel in different terminals or you've got subagents or you've got agent teams. And it was kind of cool in here that, you know, each of these people are their own terminal. And so if I see 4 of this girl, I know that that is 1 main and then 4 subagents, whereas all of these are just different agents running in terminals because they are different things. Parallel agents, you know, 4 different terminal agents. Subagents are a parent delegating tasks to different subagents. Those subagents can use different models and they have, you know, their own independent context. And then you have agent teams, which Pixel Agent doesn't yet actually show, but agent teams work together a little bit better. They don't share context necessarily, but they share a task list. They can talk to each other and they understand when the other person is done or the other agent is done. And even Boris, the genius who is kind of like behind Claude Code, said that he's using, you know, 5 agents in his terminal at all times and then maybe another 10 on the web or something like that. So he's always working with tons of different agents. So I am really interested to see what Cloud Code does around sort of like more native visualization. So real quick, let's hit on the security of this plugin because you should really only be installing extensions that are verified or that you know are safe. So before I installed it, I just looked into it a little bit. The publisher is named Pablo De Luca. He's a verified publisher on the VS Code Marketplace. I looked into his background and he is a real person. He is a co-founder of a company and he has a GitHub repo with 1,300 stars and over 100 forks. So I think that that means that he's got a little bit of credibility to him. He didn't make some sort of fake profile to put something up there to steal your data. So we looked at some key things like outbound network calls, data exfiltration, shell command injection, suspicious dependencies, credentials, secret handling, file system access, and remote script loading. And we didn't see anything. So it doesn't look like it sends data anywhere. It doesn't collect anything. It doesn't run any commands and everything is staying on your machine. So that is why I felt confident to actually install this and test it out. So I just wanted to share my experience from a security perspective. So overall, the way I feel about this is that Pixel Agent just tells me that an agent or agents are working. What I actually want to know is what they're building, what decisions they're making, and whether it's about to do something that I'd agree with or disagree with. That's the real product that I think we need when it comes to visualization, because us as humans, we're moving into this world where we really just have to be really good managers and keep our agents and our subagents on the right path. And the best way to do that is be able to just see all the time what's going on, stop things before they go wrong, and proactively keep giving them more work so that they're never just sitting there idle. But anyways, that is gonna do it for today. If you guys wanna dive deeper into all this kind of stuff and nerd out, then definitely check out my paid group, AI Automation Society Plus. We've got over 3,000 members who are building with AI every day and building businesses with AI. So it's a great place to be if you have similar goals. But if you guys enjoyed the video or you learned something new, please give it a like. It definitely helps me out a ton. And as always, I appreciate you guys making it to the end of the video. I'll see you on the next one. Thanks, everyone.