Build an AI Agent Team — Step-by-Step Guide & Transcript

[00:00] So the video you're watching right now is sponsored. And honestly, the interesting part here is that I didn't get to sponsor, I didn't research the company, and I didn't even come up with the first draft of the ideas for this video.

[00:12] A team of AI agents did all of that for me automatically. So by the time I sat down to actually look at this deal, there was already a full research report on the company, a safety audit telling you they were fine to work with, and three different video concepts that I could choose from in a small presentation that I could review in under two minutes.

[00:30] And really, all I had to do was show up, pick an idea I liked, and then just start filming. So in this video, I'm going to walk you through exactly how I built this AI agent team, because it's simpler than it probably sounds, and by the end of this video, you're going to be able to do the exact same thing.

[00:45] And real quick, before I get into all of this, all of these agents are running on a platform called HyperAgent. You can see the Mission Control dashboard right here, and they also are conveniently the sponsor of today's video, but I'll tell you a bit more about them in a second.

[00:57] Okay, so to give you a little bit of context here, I get a lot of sponsorship inquiries. They come in through email, they might come through Slack from my team, and every single one of them takes off the same routine. Someone messages me saying, hey, you know, this company wants to work with you, and then I need to ask, are they legit?

[01:13] What would the video even look like? And then I can go and take a look. And what I do is I spend a couple of hours here Googling the company, reading the reviews, checking that they're not sketchy, and mostly trying to figure out what would actually make sense in terms of a video to do.

[01:27] And none of that work is hard, right? But it's just repetitive and it happens over and over every single deal and costs me a lot of time. And that's kind of the perfect setup for what I want to show you here because I don't want a chatbot that I have to sit in front of and prods all day.

[01:41] What I actually want is a team. So a group of teammates that pick up the work on their own, do it properly, and then come back to me when it's finished. So that's exactly what we're going to build and I want to show it to you now. Okay, so I'm inside of HyperAgent here,

[01:54] and you can see my mission control dashboard with all of my different agents. What I want to do is introduce you to the agents I've set up and show you the process and how it works. So I've got all of these connected to Slack. And inside of Slack, I just sent kind of a sample of what I might get

[02:08] in terms of a deal to work with a company. And you see that I sent this, and automatically, my AI agents pick this up. They start going through the pipeline. They start doing an audit risk. And you can see my better agent comes back with a full audit risk

[02:20] telling me what I need to be careful about or if the deal makes sense. And Sam, my presentation agent, wraps this all up into a presentation, I want to show it to you right here, that goes over all of the key things that I need to look at before I decide if I want to proceed with this deal or not.

[02:34] Gives me a few different concepts for the idea and just makes it super easy so that all I have to do is literally open this document and decide do I want to do it or not. Now you can see the exact same process kicked off here for another deal I have

[02:46] related to Trueforge and all of the stuff came back with all of the different concepts and you get the idea. And the way this works is because I set it up inside of HyperAgent. And what I did is I created four different agents here, Ron, Better, Sam, and Marco,

[02:59] that are able to communicate with each other to create this finished result. And from this mission control here, we can view all the agents, see how much they actually cost, and view any improvements or things that they've learned. So let me start by showing you Marco,

[03:11] and then what we'll do eventually is actually build one of these agents, and I'll show you all of the connections. Now Marco is my virtual assistant agent. It handles the sponsorship pipeline. It's kind of the first line of defense, if you want to call it that.

[03:23] And I have it connected up to Slack, Gmail, Google Drive, and Google Docs. Now, what it will do here is it will automatically read the Slack, and any time a deal comes in, it will kick off the pipeline to start vetting the company and moving forward.

[03:35] You'll notice here that we can view all of the activity for it. We can see any instructions that we've provided, and we have the ability to have this delegate to other agents. Next, we have better. Now better is typically the first step in the process here but what it going to do is it going to kick off a full due diligence on whatever company we send So let look at the TrueForge one for example and it will create a kind of outline here

[03:57] We can see it's kind of working computer here on the right-hand side and also its browser, where it tells us what we need to be careful about, if the company makes sense, other deals that it's done, influencers that they've worked with already, so that I have the full kind of what you call an analysis here before I decide if I want to take the deal.

[04:13] Next, we have Ron. Now, Ron is our writer agent, and once we agree that the deal makes sense, it will automatically start coming up with a few different concepts that it can do. If we look at the integrations here, same thing, Slack, Google Drive, and Google Docs,

[04:27] has some instructions, and then if we look at its, I don't know, activity here, you'll see that it gives us a full doc here with a few different ideas that we can use for this video. So I ran 100 agents, whatever, outline, sponsor integration, you guys get the idea.

[04:42] So that's Ron, our writer agent. And then finally, we have Sam. And what Sam does is create that presentation that you saw, so I can review it really quickly. Same thing if we go to the integrations. This time it has access to Google Slides.

[04:54] And if we look at the activity, you'll notice that automatically it can create this decision deck for us. And that's what we're able to view and go through. So that's my team of four AI agents. They work together through this process, again, being triggered through Slack.

[05:07] There's a ton of settings and configuration we can do, and I want to show that to you now. So now that we've listed those agents, I want to create a new one from scratch and show you how to set it up. So what I'm going to do is create a new thread here, and notice that I can just select Setup Your Agent.

[05:19] When I do that, I can press Start Fresh, and what I'm able to do is just pass a long prompt into this kind of input field here, and it will automatically create a new agent for me. Now, I also can hire a full team of agents.

[05:31] I can have it interview me about the agent, or I can import it directly from something like OpenClub. Now, by default, this model will use Opus 5, but I can change this to use pretty much anything that I want, and that's just going to come from my credits that I have on the platform.

[05:43] Also, we have different tools and a lot of built-in settings. It works incredibly well. So what I'm going to do is just kind of type a long prompt or speak a long prompt on what type of agent I want to create, and then start setting it up. Now, my idea here is that while I have all of this stuff to come up with concepts,

[05:58] it's not really rooted in facts or what's currently working on my channel. So I want to kind of have an analytics agent which can keep track of what video titles and concepts are performing well, so we can use that as input to the script writer to get better concepts.

[06:13] So let's try something like this. I want you to create a new agent called Amy. This agent should be responsible for doing analytics and research related to trending topics to try to figure out what video concepts are working well on Tech with Tim's channel.

[06:27] It should be connected to YouTube, and it should be able to view all of the analytics and stats, And before we have a new video concept created, we should do some research on anything similar on the channel to see what's currently working.

[06:39] We also should research other channels or current trends to see what we can hop on to get the best type of result in terms of title, thumbnail combination, you get the idea. Basically, I wanted to do a quick report that we can pass to other AI agents before we write the script for a video that we have some kind of baseline or evidence to base this off of.

[06:57] It's probably not the best prompt in the world, but you get the idea. And by the way, if you're wondering what I'm using to do this, I'm using a really cool tool here called Whisper Flow. You can see I have 257,000 words. Anyways, it's much faster just to use your voice rather than to type all of this out.

[07:11] And you can see I always give better, more detailed prompts. I'll leave a link to Whisper in the description if you want. But anyways, let's press Enter here, and let's see if I can build this agent for us. Okay, so you can see that it's creating me here. What I can do is view the details on how this agent is set up, and then I can edit it or save it.

[07:27] So for now, let's go to Edit and just make sure everything looks good. but you'll notice that we have kind of profile photo, which we can adjust. We have the name description the trust profile which is pretty important the system prompt on how the agent actually works and then we can choose the model So for now I actually going to go with Fable 5

[07:43] and for the sub-agent model, I'm going to go with Opus 5, so I can adjust these. We'll go extended thinking and keep that in high. Let's go to tools and integrations. We can see it's automatically picked up YouTube here, which is great. Then we can see we have search, okay,

[07:58] browser controls, web pages, and all of these other stuff if we wanted to be able to do that. So it's automatically selected these, which look good to me. For the invocations, we can invoke it through a thread. We can invoke it from other agents, so anyone can run this,

[08:10] or set it up on a schedule or connect it to any type of, what is it, workspace. Then for knowledge, if we want, we can set custom knowledge. We can allow it to automatically learn memories, or we can sync it with something that's already here.

[08:22] So for now, we'll just leave it as is. If there was a skill I wanted to add, I could throw it in here, but currently I think this looks fine. So let's create the agent, and then let's add Amy to our pipeline, and I want to show you how she works in Slack.

[08:34] Okay, so to connect Amy to Slack, I'm just going to go to Integrations here. I'm going to go to Plus. You'll notice there's like a ton of different things that you can connect here. Now I'm going to connect Slack. You can also connect custom MCP servers if you want.

[08:47] For now, I'm going to go with Bot Identity, and then I'm just going to add her to Slack. Now once you've been Slack, what I'll need to do is just add her to my specific channel. So let me show you how it works. But effectively, what I can do is turn on, if we can send direct messages or respond to teammates,

[09:00] we'll leave that as is. and for the channels, we can pick that sponsorship channel we have. So we pick sponsorship demos, we can go read, write, and respond, perfect, and then simply add the channel and finish so that they have the permissions.

[09:14] Cool, so she should be good now in Slack, so if I go here to Slack, you can see Amy was added to the channel. If we said add Amy and do something like, can you look through my most recent videos and tell me which ones have performed the best,

[09:26] it should be able to do this and get back to us directly here. So you can see it actually automatically creates a thread, and I like that they have the kind of tooltips here telling us what they're thinking, what they're doing, and it should go through this now and give us a response.

[09:38] Perfect, and you can see we got the metrics here, at least the top performing videos from the last 30 uploads, including all of the shorts that I have on YouTube. So this is working, but now what I want to do is I want to start chaining these together

[09:50] so that these different subagents can work in parallel. So basically what I can do now is save Amy, and I want to go over to just create a new thread, and I want to instruct my agents, I can pick any of them or just talk to a general agent,

[10:02] to kind of adjust the flow now to include Amy in it, so that that's going to be another step where it looks at the analytics before it starts drafting the ideas. So let's do something like this. I just created a new agent called Amy. I want you to adjust all of my other agents so that they will

[10:16] use her as well, and this will kind of be an intermediate step, so before we start coming up with the concepts and the scripts, we look at videos that are already working and use that as evidence before proceeding. Put that in the plan. This is something that Marco probably will need

[10:29] to handle for us. Okay, and then they should be able to actually adjust all of the agents for us automatically to throw Amy into that flow, and then I can show you the whole thing running in parallel, which is super cool inside of Slack. Perfect, so you can see it actually created a

[10:41] plan for us here. I'm just going to approve this to proceed. It's going to ask, should Amy's evidence brief be a hard gate on wrong or advisory? Let's do advisory only so it doesn't block. Okay, perfect, so it looks like we're good to go here. What we can do now is just update the

[10:54] because it's made all of these changes, and with that now it should kick off the correct process. Okay, so now that everything's updated, I'm just going to trigger this flow to go again by just doing kind of a sample pitch here in Slack. So let's just

[11:07] type Marco on this, and this is an example of like, okay, here's a sponsorship post that's coming in, here's a budget, whatever, this is just completely made up. But let's press enter, and let's see the pipeline kind of start triggering now. So

[11:19] you can see Marco is going to start working, and what I really like is that so you can view the full process of everything that happening and all of the delegation that Marco is about to do So you can see it listing the spotable agents executing the integrations right Searching through the integrations whatever you get the idea

[11:35] And it should actually even update this sponsorship pipeline view that we have the keys track of where all of the deals are in the different stages. So let's kind of watch Slack here and see what happens. Perfect. So all this is going now, it's going to take a second

[11:47] because I have to delegate to a bunch of agents. But if we go to the command center, we should be able to see actually all of the processes running here. So you can see Marco's going, Vedder is going, Amy's waiting on something, Ron is here. And you can see all of the costs, which I really like.

[12:00] Also, all of the improvements of these agents. And it's going to take a few minutes to complete. So while it does that, I want to show you the self-learning capability that's inside here, which is super interesting. So if you scroll down to memories or to learning here, especially learning is what I want to go to, and you go to improvements.

[12:16] So what you can see here is that this platform will automatically learn new skills. You can see there's a few ones here, as well as memories, as well as prompts, and even creating different agents for you while you use it.

[12:28] So if I click into this one right here, there's a rubric that we can use to automatically assess and make sure the agents are performing properly. Same thing with all of these different memories, which I can just press save on. So if we go back to learning, I can go into the rubric, I can see any of them that already

[12:42] exist with different criteria that can be applied. So it's super cool and it's automatically learning and improving for you. And then if you want to see the memories of the agents, you can click into it, see the personal memories and any individual memories that the agents have learned.

[12:56] And then, of course, if you want, you can add skills, you can make custom ones, you can build in, what is it, pre-built ones from different providers. And then you can go to teams here and you can actually set up teams, add other members so they can use your agents.

[13:09] In my case, mine is in Slack, so I can configure the agents so other people can directly chat with them if I want to do that. or I can add them right here to HyperAgent so they have access directly inside of this platform. Cool. So we're going to wait for this to finish in Slack.

[13:21] We go back to the command center. We can see kind of the status of the agents. We can see Ron is active right now. So hopefully in a few minutes we'll get the result. I'll show it to you, and then we'll continue. All right, so the pipeline is actually finished here. You can see that we have the Neon deck, as I asked for.

[13:36] It gives me caution, tells me what I need to look through, gives me a few different concepts, and you get to the point. And then it gives me a suggestion on which one I should pick here. if I should go ahead and proceed with the video. So the point is, pipeline's working.

[13:48] We've added in Amy now as the other agent. Of course, it does take a long time to run because we're going through a ton of different ones, but we can optimize the pipeline, change it as we want. And if we go to the sponsorship pipeline, now you can see that we have the Neon deal added here,

[14:00] whereas before, we only had two. So like I mentioned at the beginning, everything you saw in this video was built on HyperAgent. And the way it works is that each agent runs in the cloud on its own login. And you can connect it to the tools you already use,

[14:13] like you saw here, Slack, Gmail, Google Docs, Airtable, whatever, there's a few hundred different tools you can connect. Now, you can message the agent just like a teammate, you can put it on a schedule, and you can let it launch different triggers, like your inbox, for example,

[14:26] and come back with actually finished work. It can do pages, decks, documents, briefings, whatever you want, and you don't have to directly chat with it to do that. It can handle it automatically. Now, if you want to build your own team, you can sign up for a paid plan using my link in the description,

[14:40] and you get $100 in bonus credits to get started. And honestly, my advice is just to start with one simple agent to automate an annoying job for you, and then you can start building out your team and build out these different roles.

[14:52] So that's my team. You saw exactly how it worked. It's taking a bunch of time for this video, and I'm sure for many in the future. This is a really cool platform. I like the way that it's set up, especially the mission control and orchestration. If you want to check it out,

[15:04] you can do that from the link in the description. Anyways, I'll see you guys in another video. .