Multi-Agent Building In Claude Code Somehow Got Easier

So, what you're looking at right now is I've got four different Cloud Code agents running for me and I can look at them and manage them from just one view, which is really, really nice. I can choose them with the arrow keys or I could just click on one of the sessions and I can see what's actually going on. And then I can just go back to my main agent view and see what else is happening. And as you start to finish up, it'll tell me the status of them. So, I can just see which ones are done or which ones might be waiting on me for input. Like right here, you can see that this one turned yellow and it says that it needs input. So, I can click into this and it's basically just asking me to approve the plan or give feedback. So, if you guys have ever had like five plus cloud code sessions open where your VS Code looks like this and you've got all these different ones to click through and you don't know which terminal tab is doing what, then this is going to be a game changer for you. So, as quick as I can and as easy I can, I'm just going to break down exactly what you need to know about this and how it works. And right on Q, you can see this is exactly what happens when one of those finishes up. It turns green and it goes down here. Okay, so what is this whole new agent view thing? Well, they just dropped this today. It's called agent view and it lets you basically have one tab where you can not only view all of your sessions but you can move inside and outside of them. So you can basically manage all of your different working agents just in one terminal tab. And I'm actually really really excited about this. So previously when you were running tons of agents in parallel, you probably had multiple terminal tabs open which got very confusing. Even if you had a hook set up to have like notifications when something finished, you would still be clicking around trying to find the right session. Now they did add something really nice that helped with that. They basically added this little recap thing which at the end of every terminal session it shows you what you were working on and that definitely helped. But having this agent view where I can literally just switch back and see what's going on and I could like move into this one if I wanted to and then I could go back out to my agent view and go to a different one. It's super nice. So the one thing here is you do have to use this in the CLI and if you guys are scared of the CLI please don't be. It's a really really nice friend and if you are planning on using cloud code for a while or even like codec or whatever coding tool you want to use the CLI usually just has the most functionality. So if you have been using the extension which looks like this which just maybe looks a little bit more friendly and less scary I would definitely challenge you this week to start using the terminal. Anyways all you have to do now is you press the left arrow while you're in any session and it opens up the agent view. You could also launch a new session with clawed agents but just hitting the left arrow is so much easier. And then when you're looking at all your sessions, you can obviously see what needs input, what's working, what's completed. And then you just do the right arrow to go into the session and then left arrow to go back to your main agent view. Now, one thing that's also pretty cool is if you're working on a session already, you can just do /bg and that will put it into the agent view. And you can launch a new session with claude flag bg and then name the task. But it's so much easier than that because you can actually just straight from the agent view. If I just describe a task right here, that actually just kicks off a new session. So, let's say I wanted to go ahead and do some research on Claude's new agent view and just like tell me about it. As soon as I hit enter, that just starts off a new session. You can see that one was right here. And if I move over, then we can see what's actually going on right here. And that's the exact prompt I just shut off. And once again, I go back out and now I see all of them. So, if you're reading through this doc and you feel like that's like commands that you have to remember, it's really not that bad, so don't worry about it. So, how are developers using this agent view? Well, they listed four main patterns here. So, the first one is obviously scaling the number of concurrent sessions. The second one is managing longunning agents. And they also just dropped a new feature called goal, which if you guys are familiar with the goal feature in Codeex, it's kind of just like that whole Ralph Wiggum loop. So now basically if you do SLG goal, you can set uh an objective for Claude Code to work towards and it basically will experiment and try different things until it hits that exact objective. So if I did / goal and I told it to build me a 3D monster fighting game, it will basically keep going at this. You can see how long the goal has been running. And these can run for hours and hours. they can run overnight and they'll come back once that objective has been achieved. Now, this is not the best way to use goal. Obviously, this was a very subjective prompt. And because you're kind of oneshot prompting, if you use goal, um you know, prompting is super super important and it's usually best if you can give it an objective metric to hit. So, if you think about the way you do like Karpath's auto research, you're usually trying to have it optimized for a metric. So, when you do goal, you're probably going to want to do a similar type of prompting. But anyways, this video is not about goal. The reason why I wanted to bring this up is because let's say you're doing a bunch of different goal sessions. You can just view them all right here because they're probably going to be working for a long time. And if you look on the right hand side here, you can see how long each of these sessions has been running. You can also navigate between separate sessions really easily just using the arrow keys or even just using your mouse. And then you can see what shipped based on the status indicator. And it's really that simple. That is basically the end of this doc. That's all there is to it. You really just have to get in here and play around with it a little bit. Now, the reason I said you don't have to worry about all those commands is basically because the idea is in a fresh terminal you could do claude agents and that would open up that view which we were already just in, right? Or in a fresh session you could do claude-bg and then you could describe a task like um you know build me a 3D monster game. I'm just going to use that same example again and that will basically kick off a cloud agent that gets launched into the agent view. So, if I go back to the agent view, we would see another one which is, you know, build me a game. And for some reason, this got cut off because this is in research preview. You might notice a few bugs. The first time I started launching this up, my computer got really, really slow. So, just keep in mind, research preview, it is going to be maybe a little bit buggy. Yeah, it comes back and says build alone is too ambiguous. And by the way, I was tinkering with it a little bit more and I realized that you have to wrap it in quotes. I tried the uh square brackets, but the quotes is what actually worked. If you see over here now, this is the session that it kicked off and the full prompt got injected. But honestly, I'm not really going to be ever launching sessions using this flag. I'm probably just going to be launching them from here. Or if I launch them from the regular command, you know, like I just come in here and do claude and then I start talking. And if for some reason this doesn't get into the agent view, which it pretty much always does automatically, then that's when you would just come in here and you would use slashbg, which is like slashbackground to put it into the agent view. So, I think the value in those commands is if you're trying to do work in different directories and still manage everything. Because if I'm in my Herk 2 project and I open up a new terminal, this is going to keep me in that project. And if I launch one from here, I'm pretty sure it's just going to create another agent inside of Herk 2. So, if I wanted to CD into a different directory, and then I could do the Claude with a flag like the Claude BG, and then I could say, "Build me a 3D monster fighting game." And then that would kick it off and I'd be able to see it here. And now I'm working inside of different projects. I wish what we could do is we could see right here very easily which directory we're in. But all we'd have to do is we'd have to switch in and we could see this one is in my project up here called a bunch of skills and then this one is in my project right here called pretty sure this one's in the herk too. And so the other way that you could do that is if you're in a different project and I do claude and that just launches your regular claude code. Then let's say I say you know hello but then what I want to do is add this into the agent view. Well, it's still already in there, right? It's still there. I didn't have to do the slashbg. So hopefully that stuff makes sense now. Now, what else is cool is I just showed you guys like entering a different session, right? Now, what if I wanted to delete one, I could just select it and do Ctrl X twice, and that kills it. You can also see on these ones that need input. So, for example, this one says, "Do you want me to save these to project/lin post?" What I could do is just hit space and I could reply right here from this agent view. No, I don't want you to save those. That was just a demo. And I could shoot that off from this view. And now that session has gotten our input. That's basically all there is to it. I'm not going to hammer that home too hard. I think you guys understand it. What I would really recommend now is if you don't really see the value in this, it's probably because you're not running multiple sessions and you're not like doing a lot of work at the same time through cloud code, which means you're not really using it as your operating system. So, what I did for you guys is I built a full course. It's like 2 and a half hours on building your own operating system using cloud code, building your AIOS. So, you're going to want to check out this video right up here. Hopefully, I'll see you guys over there. But, if I don't, I still appreciate you guys watching this video. If you enjoyed or you learned something new, please give a like. 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.