Transcription
Everybody's building AI agents, but most people are throwing agents at random tasks with no structure, no architecture, and no way to make them work together. That's not a workforce. That's chaos.
We built a single module with six AI agents, each with a specific role that replaced 80% of our marketing process. Competitive research, ideation, scoping, scripting, editing, design, all handled by agents that actually collaborate.
In this video, I'm going to break down the three layer architecture that makes it work: modules, agents, and workflows. And then I'll walk you through our content creation module and show you how the system actually runs live. By the end of this, you'll have a framework that you can apply to build your own AI agent workforce for any domain, content, sales, operations, whatever.
My name is Adam. I'm a toprated and expert vetted freelancer on Upwork. I've done over 60 projects there, as well as a further 250 projects through my AI development studio called APG Software. Everything I'm showing you today is a real production system, not a demo. Let's get into it.
So, most people build these agents isolated and not having access to other agents. So, we can look at it a little bit like this. So, let's take the content creation department for example. We might have an agent that does the copywriting, copywriting for LinkedIn, copywriting for video editing, copywriting for whatever the case is, whatever the channel is. Then we might have another agent for ideation. That's more of the the thinking agent. I'm going to show you in a real life example on this in a moment. We might have another one for video editing. We might have another agent for actually scheduling that content via API onto the different channels. So actually scheduling it so you can plan out your week in advance. We've got all these different agents. But typically what happens is when people go to build in Claude code, those agents can't talk to each other. And there's two places to build. Some people are building just in ChatGPT, my GPTs, or they're building in Claude projects or something like that. They've got all these projects which are really helpful, but those projects can't talk to each other. So that's one use case. Or people are actually jumping into a CLI and using Claude code and building folders here. And in this case, they can talk to each other, but people aren't organizing them correctly.
Now you see this CCS folder. That stands for Content Creation Suite. And under here, you'll see six agents. You'll see the content strategist, the copywriter, the creative director, the editor, the publisher, and the video editor. That replicates a, you know, I call it enterprise grade. I just mean large company, but what a typical like production grade marketing or content creation suite would look like or team has these actions. And then within there, we those have access to commands. So if we take a look at the video editor for example, we can see that this has access to the following commands. It can, uh, it can ingest videos. It can perform audio analysis. Can transcribe videos. It can perform visual analysis. So tell you what's happening on the screen of a video. It can clip and automatically edit videos. So clipping out dead noise or that kind of stuff. Uh, it can edit long form edit. So it can motion graphics, B-roll, that kind of stuff. Can edit short form videos. It can extract clips from other videos to use as B-roll for short form content. So, we can see this is a list of commands that it has. And you'll see that it actually points to a workflow. So, in our workflows folder for the video editor, we can see it has these actual tools available to it: audio analysis, long form edit, short form edit, transcription, video clipping, etc.
But when it's editing videos, it can pull information from the other agents, and so can the copywriter and other things. So, for example, when we look to, uh, to plan B-roll extraction. So, let's say we have a long form YouTube video and we want to pull B-roll in from, let's say we're talking about the latest in the AI news and we want to pull videos and extracts of articles and stuff like that to refer to. We can go back to and refer to the copywriter agent which built the script and we can say, "Hey, look, what are we, what are we actually talking about in this video? What do I need to extract?" And these agents can actually talk to each other. So that's the key thing. That's one of the key things that's getting solved when you actually build a proper suite and a framework to use Claude code within is that persistent memory and being able to actually centralize information and use it across different agents. And that's what creates an actual workforce. And that's where we can actually build an entire content creation suite, not just individual agents that do specific little things.
So, in our case, we're using this content creation suite and we started building this last Monday. So, in 10 days, we've built an entire prototype and that edited the last three videos and posted content all across my LinkedIn. Last week, we did 50,000 in impressions over LinkedIn. Last week, I think we had over a thousand comments across a number of different posts. Some of them were lead magnets, some of them were more personal style posts. Uh, and on YouTube, I had one of the best weeks, if not the best week on YouTube that I've ever had. Uh, I think almost 10,000 views, which is a fair bit for my channel at least. Uh, and a number of subscribers. So, and what that, what that content creation suite did was it allowed me to ideate ideas, do competitor research, uh, clip up the the videos. So I might have some of those recordings were like an hour, two hours long, would clip up all the the dead space, the noise, etc. Uh, clip up the videos, then go ahead and create blog content, create email campaign content, create LinkedIn posts, both video, carousel, image, and raw text. Uh, and then schedule it across the different channels to post whenever I wanted to. All within one module that operates inside of Claude Code in an IDE using this organized structure and there's persistent memory across each of these.
So if we take a look in the configuration of these agents, sorry, the memory folder, we can take a look. We can see the different, so we call it the sidecar. But what we can see here, for example, let's take the ideation agent, so the content strategist. We can see that it's got persistent memory about our brand guidelines, our content history, so what we've posted previously, our ICP profile, and instructions on how to run. So this is persistent memory that can be accessed whenever that, or whenever a content strategist is enacted. So this is like the next layer that dramatically improves the speed of how you can generate. Because with tools like this, we can't, if we wanted to, you know, update the history of what we've posted, if we had a specific my custom GPT or tool to post, we have to continually update the system message and the prompt. It doesn't have that clear persistent memory anywhere near as organized as if we were to actually set up a system in Claude code to to build this. And so we end up getting, before we have like data silos and things not talking to each other, and afterwards, we end up getting all of these separate agents that can work together to create content and push content into a separate section.
So, the hierarchy for this is we call it a module. So a module, you can think of as as a business process and a business case. The content creation suite is a module. Within the module, we have six agents, which replicates a development team. But let's say we took the delivery process for example. We might have in that module, we might have a project management agent. We might have, let's take the delivery process in a real estate company, and I'm not a real estate agent, but I'm just going to guess what their processes look like. Uh, you might have someone that drafts proposals. You might have someone that builds ad copy, um, and, uh, web pages for particular properties, etc., etc. So you have a number of agents which typically replicate like human job roles within a module. And then within an agent, they, they can access a number of workflows. And actually multiple agents can access the same workflow in some cases. So an example for a workflow could be, uh, let's say for example, generating images using Gemini. Multiple agents might generate, generate images. For example, maybe the marketing agent would generate, generate images, but then maybe you also have a customer service agent that might want to generate diagrams to assist in the explanation of certain topics to customers. So that would also use the same workflow.
And so that leads me on to the three-layer framework, which is what I just described. Um, but again, we have the three-layer framework, which is modules, agents, and workflows. And that's the real key tip that's going to level you up from just what I would call, you know, entry level. So I think entry level is just using ChatGPT. Level two is being able to configure projects and custom GPTs, giving them system messages, uploading files for them to access. And then level three is what is this framework here, which is being able to actually build entire modules and content departments to, to be able to, you know, even further increase the efficiency of what these AI agents can do and the context that they can access.
So in our content creation module or suite module, which is the specific module that I'll be demonstrating here, we have the content team. Now we have the content strategist. So this strategist has the brand, the ICP, and the research within it. Then we have the copywriter. So this is writing all of the copy. Then we have the gate or the editor. Now what the editor is doing is it's reviewing all of the content to make sure that it's targeted towards our ICP or ideal customer profile. So every piece of content that we create, we can get the editor to scan it and make sure that we're not going off topic or it's not rubbish content and it's all relevant and it's adding value to our target audience. Then we have the creative director, thumbnails, B-roll, styling, etc. We have the video editor. So, this one's specifically working on short form and long form video editing, clipping analysis, uh, that full process, which is pretty incredible. Uh, and then the publisher, which is working on actually scheduling that via API to the different tools and managing the content calendar and the calendar history, uh, and future.
And you'll see that that is represented in these agents. Uh, and when I show you through the actual data, so this exact video was actually generated using this, uh, this framework. And so if we come down to the projects, you can see here that building an AI agent workforce is just what I've called this. And if we go down to the strategist agent, you can see that first I did competitive research across the AI niche in the AI space. I built this document that scanned a lot of the top creators. It's come down. So, if I open what this looks like, some of the top people in my niche and the kind of content they're pushing, it's, it's figured out outlier videos. So, outlier videos are things that people are really interested in watching right now. Lot of stuff about AI agents, all that kind of stuff. So, there's research, keywords, all of that kind of stuff to check what's trending in a specific niche. We used the competitive research agent or workflow, sorry, to do that.
Then, we ideated a concept again. So, this was, this was done by the content strategist. It built me an entire concept for what to, what video to create. In here, it's talking about how it, it's measuring how it targets my ICP. Uh, and it's building me all of this strategy on the concept to build using those two documents. I then passed them off to my copywriter agent, which has then gone ahead and developed the script. So, this script actually corresponds, I don't read directly off a script, but it does build the, it builds the intro and an idea. So, basically just dot points, cuz that's the way I like to film things, is build dot points on an on a general structure, and it's built an intro and everything. And then what it's actually done is it's gone ahead and built the diagram that you're looking at right now using Excalidraw. So, it's called the, uh, it's basically called a workflow that has gone ahead and developed this entire workflow, which was generated in one prompt with AI. So, that's what that's done. And so, that's a bit of a live demo on how this works.
But the next things I'm going to do is, as you saw before, I've got the video editor agent. I've got a number of other agents. So, as I'm doing this, I'm going through and I'm refining. So, I actually had all of these. If you go and look at my video editor, video a couple of days ago, I've had all this built, but it wasn't as organized and as clean as this new structure. So, what I've done is I've gone and rebuilt it using the BMAD framework, which is just really good for organizing your data really, really easily. Because I also find sometimes like when you build something the first time, you notice all these areas where it falls short. And so what I've done is I've come back and I've gone and built it from scratch again to try and make it a lot cleaner in the second build. And the stats are pretty obvious. If you look at my LinkedIn, which is Adam Gooda, you can check my, you can check my progress over the last seven days and some of the posts that I've made, but I've pushed some really good numbers there. You can check YouTube. For me, I'm trying to offer as much value as possible. Uh, but I've been consistent in posting there and those, the speed in editing and the speed in being able to ideate, do research, and build content ideas to push out has meant that I can push more quality content out, create more videos. Um, and those numbers have gone up as well.
And so the idea starts with something like content creation. And for me, the, the idea is to automate a lot of the, the stuff that, uh, is non-human. So, ideating, competitive research, stuff that you don't really want to do manually because I'm running an agency at the exact same time. And I don't want to, I don't have full-time availability to be a content creator, but I love making content still. And if I can, you know, speed up some of the, the processes involved in doing that, it, it frees me up to push out more content because I can get more done in a smaller amount of time. So, the first thing to do is build this. But then, I've also been using this in my agency to build systems where I can manage clients. Uh, I can generate proposals. Everything else becomes really, really easy. But for a lot of people, that step in itself is too big of a leap and it's a barrier to adopting something like this. So if you want to get on this stuff early, that's the main thing to get your head around is how to actually use Cursor, uh, Claude coding Cursor. And I'm going to be releasing a video on that very, very soon on how to actually get that set up. Once you have that set up, you can follow the steps in some of my other videos to go ahead and start building your own. But install it, give it a go, use the B&B custom agents creation, which is in the method, to go ahead and create your own content suites. You can literally run them by running create module. That'll create a module. Create agent will create an agent, and create workflow will take you through the steps to do exactly that. I show that in my one of my most recent videos. Go through those steps and create your own workforce to automate a lot of the things that are taking up your time that don't need to be.
I hope you found this video interesting. If you did, consider clicking subscribe because I'm going to be pushing a lot more content on Claude Code, BMAD, and how to build AI agents that are actually helpful. Thanks.