289 | My entire Claude setup for creating AI teams that run my businesses

The inspiration for this episode is the fact that when I woke up this morning while I was drinking my coffee, I already had two podcast guest pictures ready to go. I had six AI news articles summarized in my AI news application. I had six linked in posts, including their suggested images ready for me and Joyce to review. I had a client proposal together with an email to the customer as a draft in my outbox and I had five potential YouTube shorts ready to go for me to review or for my team to review to get posted in addition all based on my previous full podcast. All of that and a few other things before I did a thing on my own. This is the power of having multiple agents teams of agents working for you, not you commanding them, but you setting them up and they're working for you across multiple aspect of the business. And this is the inspiration for this episode. Hello and welcome to the leveraging AI podcast, the podcast that shows practical ethical ways to leverage AI to improve efficiency, grow your business in advance your career. This is a sour mate is your host and I've got a really exciting episode for you today that I wanted to do for a while. If you've been listening to the spot cast in this past quarter since the beginning of this year or the end of last one, you know that I've been diving very deep into developing and deploying and using multi agent orchestration solutions or if you want in simple terms, complete armies of AI employees across Stephen teams across my two and a half different businesses that are basically running more and more things in the businesses. It allows me to do magical things that allows me to grow in a speed I couldn't even imagine possible and that's all while having only myself and my incredible assistance to working in all of these businesses combined not completely true at least on the operational side. It's us on my software company. I actually have a software team and on my half company that I'm sharing. I have a partnership with another company that actually has human employees as well. But the vast majority of the work that happens that I'm involved with doesn't happen by me actually happens through these AI employees teams of AI employees that actually do the day to day work more and more of it. And it's not just about efficiencies and doing stuff that I could have done. It's actually doing a lot of things that either I don't even have a clue how to do or I know how to do it, but not effectively or I know how to do it, but I don't have the time to do it because I'm running to in half different businesses. Either way, it allows me to grow the business go after new initiative, new revenue pipelines in different ways that were not possible to me before I had these teams of AI employees. What I'm going to show you today is how my cloud universe looks like, how I think what are the systems that I'm using, what are the concepts behind the scenes, how does it all connect together and some examples. So it will help you make it more tangible and hopefully to follow that path and do the same thing. So with that in mind, I will share my screen and I will go through some slides and some sharing of actual cloud and other aspects that I'm working on that will help you to understand how I'm thinking and how you can hopefully do the same thing before I get started. I will say one thing this episode is brought to you by the multi agent orchestration course. It's a course and workshops for companies in which I teach everything you're going to see very briefly now, but over eight to 10 hours of going through every single. Tiny component here and delivering it in a way that enables people companies to implement this as well and build teams of AI employees that can help you grow your businesses. We sold the first two cohorts of this in 10 days. We're now in the process of selling cohort number three that third cohort starts on June 22nd and it's selling out fast. So if you want to learn within the next three months, how to do the magic that I'm going to show you today come and sign up for the course now before this cohort sells out. I'm going to keep on selling cohorts, but that means you have to wait basically now an entire quarter unless you join the one that is open right now with that. Let's dive into the actual flow and the concepts that I want to share with you today. As I mentioned in the opening to this episode, when I woke up this morning, I had two podcast guest pitches ready to go. My team of AI news specialist has found and summarized six AI news articles and added them to my application that I use in order to create the news. So every weekend, I had six linked in posts, including images ready to go for this coming week for the team to review. I had a client proposal perfectly written based on a conversation I had with them yesterday, including the email to the customer as a draft in my outbox with the proposal attached to it with the relevant setup in my CRM to know that the proposal is sent and so on. And I had five options for YouTube shorts, the actual videos, not suggestions based on my previous episode that I deployed to YouTube many of the things all of that with before me doing a thing. And this is exactly what I want to show you today is how to build these kind of things. This is the magical aspect of all of this. Now the trick is which most people haven't done yet is to go from a world in which you chat with AI to a world in which AI is your actual workforce. Now it doesn't need to replace your human workforce. It is going to enhance, accelerate and amplify your human workforce. In my case, it is just the two of us, Joyce and myself and literally the rest of the work is done by AI and it's growing every single day because we're building both Joyce and myself. There are more agents and more more teams of agents that are working with you, but you can do the same thing and amplify whatever aspect and whatever team in your company. So the first thing you need to do is a mindset shift. Instead of you telling the AI what to do, chatting with it, copying and pasting things into AI chats and trying to use this in your day to day flow. You need to change your mindset of going from AI is a chat partner to AI is employees, actual teams that can do actual work across almost every aspect. The company and to be fair across every digital aspect of the company. The second thing is you've got to stop thinking about AI as a tool, again, something you use and think about it as workforce that uses tools in order to achieve different goals. And these tools that the AI can use is basically everything you use in the digital world. It is writing documents, it is doing research, it is summarizing, it is doing analysis of financial information and the other numerical information. It is writing proposals, et cetera, et cetera, et cetera, every single thing that you and your team doing the digital space, the AI can do while using similar tools to the tools you are using in order to achieve these goals. Now the real magic happens is instead of building a single agent that does one thing is building multiple agents each and every one of them specializes in something that is very, very specific, which makes it very good in that one thing. And it also makes it very efficient from the memory that it's using because it is focusing on that one thing and then having multiple of these work together to achieve different goals. I'm going to show you several different examples of exactly how this works. But before we dive into from the concepts we talked about so far to the practical actual things, I want to clarify the terminology to make sure we're all in the same page. If you are watching the screen watching this on YouTube or on Spotify, you can look at the screen if not, I will explain everything that we're seeing. So what we're seeing is a little drawing of a company and I'll explain the terminology in the AI agent space and how it connects to what happens in real life. So the first one is an employee and an employee is an agent and agent replaces or represents an employee who can do different tasks in your company. Before when you build agents correctly, they focus on one thing and if you want them to do several different things, build several different agents and then work with them together as we're going to discuss in a minute. The second thing is the SOP, the standard operating procedure is called a skill. A skill is something you teach the AI how to do and as sophisticated as it sounds, it's literally a document. Everything we're going to talk about today, the whole magic of creating this really sophisticated incredibly powerful automation tools and capabilities is all build around what's called markdown files or dot MD as the type of file that you're going to see, but all it is is think about a very simple word document without all the fancy formatting. It's just a word document that is written in a well structured way and a skill is basically the SOP how to do something specific how to do research how to write the document how to design a presentation how to create graphics, how to do financial analysis, how to do that step versus the second step of the financial analysis, how to review an RFP, etc. Each and every one of these things can be a skill and you can have more or less as many of those as you want in order to run your business just like you do with people. Next one is the workstation where work actually happens and when work happens, it's more than just one skill. It is multiple things that needs to happen together, including different tools that you need to use in order for the person at the workstation to be able to do what they need to do and that workstation in the AI space and in the cloud space is called plugins. So plugins are an aggregation of multiple skills and tools and commands and connectivity to other sources and things like that which enable the work to be able to happen as a package and a package that plugin can actually be shared with other people in your organization and if you really want with people outside your organization as well. It's a package of everything that's needed in order to complete a specific kind of work. The next one we have the manager, which is in the AI terms is the org. an orchestrator, as the name suggests, knows how to move data and tasks around across different agents, across different employees in the company, just like a real manager does. So you could build a sequential process where one agent triggers the next. That is helpful in some cases, but it is limiting the flexibility of the process. If you are creating agents that can do different things, and you create an orchestrator that understands the broader goal and what needs to happen, he can be a lot more flexible with which agents to call in which section and scenarios in order to be more effective and more flexible in addressing different situations. So a orchestrator is basically a manager in the company that can hand work and get the output of the work and then continue to the next step of the work with other agents/other employees. The next component is the employee handbook. If you want how the company operates not a single SOP and that is in the cloud universe is called clod.md, but the same exact concept exists in chatGPT and exists in open-cloth and existing any one of the other tools. It is the general definition of how you want to work with that specific environment and that can include things about you, things about how you communicate things to do and not to do guidelines, guard rails, anything you want that will be the way everything needs to work. Not just a specific task, not just a specific employee, but the rules that everybody must obey by needs to be in that environment and that makes it extremely powerful because you can align your entire workforce to whatever structure processes and so on that you want. The next thing is in my image is a badge that allows you entry to additional rooms and that is in the AI world and MCP. An MCP server is basically a way for AI and there's other ways to do this, but it's a way to allow AI access to other universes. It allows it access to your email system. It can give it access to your marketing platform, to your CRM, to your ERP, to third party tools outside your environment and MCP is just a very simple way to do that. You can also connect it to actual APIs. You can also connect it to CLI. There's other ways to do this, but MCP allows it to happen very, very quickly and today with almost zero knowledge you can connect your tools into a huge variety of MCP servers and allow your AI agents, workers, orchestrators and so on access to your universe of tech stack. The last component is a conveyor belt, but actually brings goods in and out and for that I use N8N. N8N is an automation tool that is open source that is complex to use, but I don't care because I don't actually use it. The AI is generating everything I need and creates dozens of these N8N processes, which then knows how to take data from different components in my tech stack and from the things that the AI generates and move them around in the most effective way, connecting to literally anything you can imagine, including servers on Google Cloud or AWS, including databases in different systems, including my YouTube account, including LinkedIn, including anything else you can imagine. So quick summary, we have the employee that is an agent, we have the orchestrator which is the manager, we have the SOP which is a skill, we have the workstation that is a plugin, we have the employee handbook or the company rules, which is in the cloud universe cloud.md, we have MCPs which is the badge that allows us access to other rooms with more information, more capabilities and tools and so on. And we have the conveyor belt that knows how to move goods in and out that is N8N. So this is how my universe and any AI agent universe actually works. Now let's talk about my operating system. What is in my operating system and how does it actually work? So the idea here is that instead of just having these agents, there needs to be structure, there needs to be clear processes, there needs to be an environment for everything to work in the most effective way. And this is the biggest unlock. One thing is to that and start building infrastructure and building the underlying concepts, ideas, processes, protocols which I'm going to share with you right now. This is when I was able to start unlocking significantly higher levels of growth and efficiencies. And these are two very separate things. The growth is doing things that drive revenue to the company. Efficiency are things that are reducing cost to the company. Both are very important. I can tell you from myself and from doing this with multiple businesses that growing the company drives significantly higher ROI than the savings you can get on the savings sides. Doing both obviously is very important because you gain on both sides. I will also tell you that it is very hard to get to the things that are driving revenue before you start working on efficiencies and understand what is possible. So how does that make infrastructure set up? The first thing is what I called the shared work environment. It's the environment in which I work together with the AI. You have to build a working environment in which you and the AI can have access to the same documents to the same information to the same outputs. So you don't have to copy and paste and go back and read and look for everything is structured in the same place. And the way this is happening is actually relatively simple. Everything these models do, as I mentioned before, is written down in Markdown files, these documents. All you have to do is create a folder structure that will work for you that is identical across each and every one of the initiatives that you're running. For me, I have one mega folder that is called Cloud Desktop and underneath that I have different universes for my different companies and underneath that for each and every one of them, I have the specific projects that I'm working on and each project there is the same folder structure until a certain level and then there's a different thing for every project because they require different things. Because of this way, everything I do lives in the same structure everywhere. I know where to go and look for it. I know where to add files when I want to add files and the AI knows the same thing. So when I do something or the AI does something, it lives in a very clear folder structure going back to SOPs and processes and des practices. And this way I can go back and forth between different tools without any problem. The cool thing about this is that in addition to the fact I don't have to explain anything to AI. I don't have to continue from where I left off without any clear passage by remembering myself on retyping the instructions. I can also navigate between CloudCode and CloudCode and I will tell you something more interesting. I can navigate between Cloud and Gemini and ChatGPT and GROC and any platform that I want because that infrastructure that I created is completely blind to which model is running on top of it, which provides a lot of flexibility that we're going to talk about later in this episode. But the key thing here is the ability to create a folder structure that is the same across the board with the same type of files in each one. We're going to talk about what these files are in a minute that everything connects to. So let me show you a simple example and I'll show you exactly how I work literally all the time. So on my main screen I have always these two things side by side. On the right, I've got CloudCode work on the left, I've got CloudCode. In this particular case, CloudCode is running inside of cursor, but you can run it however you want inside your computer terminal on other third party platforms. By the way, you can also run CloudCode inside of the Cloud Desktop app, which we're going to talk about in a minute. But the reason I have it open on both sides is because there are things that I do in CloudCode, which is writing code, creating applications and building more sophisticated stuff. And there is where I spend most of my time, which is CloudCode work, which I have on the right side, but they're both looking at the same folder. So when I tell the left side, when I tell CloudCode to work on a project, I don't need to tell it what to work on because there are set of documents that will tell it what to work on. And when it finishes working, I don't have to tell CloudCode work that we finish that step in the process because it knows because they're looking at the same documents. So this is the magic of creating the right setup and infrastructure of your folder that shared space between you and multiple AI tools. Very quickly on CloudCode work, if you haven't used it, on the top left corner, you have three little buttons. One looks like a chat, which takes you to the regular Cloud chat. The second one looks like these little checkmarks, which is CloudCode work. And next to it, there is like a little code thing and that takes you to CloudCode sessions. Again, I very rarely run CloudCode inside of the Desktop app. I actually run it on other applications, such as cursor or the terminal, but it doesn't really matter. You can run it in here as well. On the left side menu, it is very similar to what you know from regular chats. There is the ability to start a new task. There is projects. There's scheduled things so we can schedule sequential things. I will show you examples later on how that happens. As an example, my news summaries happen on a regular schedule. My dashboards get updated on a regular schedule. A lot of things happen on regular schedules. Then you have live artifacts where you can create things that connect to your live data and create dashboards and so on. You have this patch, which allows you to connect to your phone. So a lot of the work that I'm doing actually happens on my phone. The way this works is as long as CloudCode core work is up and running. You can open Cloud app on your phone, whether it's iOS or Android, and click on this patch on the mobile. And then tell it what you want to work on any of your projects. And it will actually trigger the desktop application to start working on things. So if you see down here, I have multiple scheduled things. It's actually not showing all of them. It's showing some of them. And then I have this patch and you can see on this patch, there's like 30 different things just happening in the past three days because every time I have an idea or every time I'm on the run and I need to check up on a process, I can do this from my phone and it will actually happen on my computer as long as it's open and it's running. And then underneath that, there's the actual conversations. In the middle, you have the actual chat and everything that's going on. And on the right, you have on the top right, you have the progress. So the way you work with CloudCode Core Work or with any agentic system is you give it goals and it will define the steps that it needs to do. And in CloudCode, Core Work, it actually shows you what it's doing and it's crossing out everything that it completed, it shows you what folders and context it is connected to and what tools it's using and stuff like that. So that's in general how Cloud Core Work works. Now back to the process, so we talked about having these structured folders, but where are the files that we actually put in these folders and what we're actually doing to create this additional infrastructure? So the first thing that I have that I'm using all the time every single morning, that's my starting point, is my portfolio dashboard. What the hell is my portfolio dashboard is the place that summarizes all the different projects that I'm working on across the board. How does it work? There are several different agents that goes to different aspects of my universe. They go to all my files, so all the MD files, all the different projects because they're structured in the same way. It goes to my clickup, it goes to my email and other places and it looks to what is happening today and it's creating a dashboard for me that allows me to decide what are the highest priorities for me today. Now if I'll show you how this looks like, it looks like. So again, those of you who are not seeing, I will describe what I'm seeing. There's a section at the top that is a very high level summary that shows me I have 31 active projects, 21 that needs my attention, six up and running, live basically 29 ideas in my backlog, five projects on hold and 14 still projects. And I've got a list of the still projects. What are still projects is projects I didn't touch for 14 days or more. Again, this is perfectly fine. There's priorities in a company, but that shows me kind of like things that I may have forgotten that want to go back into. It gives me my top action items that comes from the actual tasks in the actual projects. I'm going to talk about how this works. But again, this pulls it from clickup and from specific folders in my files, which I will show you in a minute. Now I'm going to get the actual portfolio of projects themselves. I have the ones that are in discovery, the one that in architecture, the ones that I'm building and working on right now, the ones that I'm testing and the one that I'm live and the ones that are live. And on the bottom, I have my backlog of ideas and I can see all of them in different categories. And on the very, very bottom, I have the portfolio growth, basically a bar chart that is showing me how many projects existed in every single day. And since the last week and half or two weeks or so, it is also showing me whether they're in discovery or architecture or build or testing or live. And I can see how many projects I'm adding regularly and how much progress I'm making across each and every one of these statuses of the projects. So these projects span across everything that I'm doing across the different companies that I'm running and they work through financial analysis, creating accounting, purchase orders and invoices, working with suppliers, answering emails, working on LinkedIn, doing marketing, infrastructure projects, etc, etc, etc, across basically every aspect of the business. And I'm moving several of them in parallel every single day. So this is one thing that I wanted you to see. This is my dashboard. And how does all of this happen? What's actually happening behind the scenes to enable this incredible knowledge on every single aspect of everything that I'm doing? And that is happening because I have these standard documents that I mentioned before, inside of each and every one of the folders and each and every one of the projects. The first thing that I have is evergreen documents. When you use chat and you use chat all of you regularly, in a regular chat, once you start a new chat, it doesn't know what happened in the chat before. Now all the AIs now have quote unquote long-term memory and they remember things about you, but they don't remember the granular details of everything you talked about in every single conversation that doesn't make any sense. The problem with that is you run out of the context window, if you're going to do everything in a single chat forever, and then the AI goes cuckoo and starts forgetting stuff and start doing weird things and you have no control on what it actually knows and it doesn't know. The solution is what I call evergreen documents. And these are documents that I built for the AI and the AI knows how to use it across the board and it explains what are we working on. What is the project? What are the steps? What did we do now? What was completed? What is not completed? What building blocks we're using? What third party tools we're connected to for this project? What are limitations that we're aware of? What are things we shouldn't do? All of this is defined in the evergreen document and the document gets updated automatically by the AI or the time. The same document gets updated by Claude Cod because in the general instructions that we talked about in the beginning, the company employee handbook, it tells them to do so. Maybe thing that starts with reading the evergreen document, every step that is made gets updated back into that document and therefore I have consistent memory for everything that I'm doing and I can start a fresh new chat and tell it a project I want to work on and it will be able to tell me exactly what I need to work on. What was completed previously and what are the exact next steps? To make this more practical from a task perspective, I actually broke this down into another set of documents and that's the task registry. So in addition to the evergreen document, there's a task registry and the task registry is as the name suggests, it's just a list of tasks. It says, "What do we need to do? What was completed? What's the status? Who needs to do it?" And that gets updated automatically, dynamically, without me having to do anything and that's why all these tools and it doesn't matter which AI I drop on top of it will know how to continue from any even point I'm in. So these two are two of the biggest tricks and then there is the structure of each and every one of the projects. So in addition to knowing all this information, I want to know how exactly the thing works and the reason I got to this was I started building these automations and they started getting really complex. Those of you who are looking at the screen can see what I'm talking about right now. There are dozens of components in here. Some of them are agents, some of them are N8N processes, some of them are databases. By the way, I haven't built any of them. AI has built all of them and this is what I'm teaching in the course is how to teach AI, how to build all these things so it can build these incredible processes. But I started losing control. I did not understand what connects what and how the whole process works, which makes it harder for me to continue evolving it or to troubleshoot when things break. And so I created another skill that generates incredible flow charts that shows me everything that's happening how the actual process works in each and every one of these projects. Now we'll dive more into this afterwards when I'm going to show you examples of specific projects. General, what's happening now is in my dashboard and I'll go back to it again. So in my dashboard for each and every one of the more developed projects, the one that are mature and are in testing or either live, it has a link to a flow chart. And if I click on the flow chart, it's actually going to show me exactly how the thing works. And if I put the mouse over any of these components, you will tell me what it is and it will show me what are the inputs and outputs of each and every one of these steps. So I can understand exactly how this works. So far we talked about a specific project. As I mentioned, each project has the flow chart, each project has evergreen document and it has a task list. But I also want the system to learn across the board. I want the fact that I learned something in one project. Let's say on how to work with NAD that happens in all my projects. I want the other projects to be aware of that as well. And so there's a higher level document called lessons learned that gets updated from all the agents across everything that they're doing and they're all reading and writing too. So I'm creating an ever growing, improving organization of agents because everything they're learning across the board anywhere gets written to the bigger lessons learned and that drives knowledge across the entire organization, which I wish it was that easy to do with human as well. But it works incredibly well with AI agents. And then as I mentioned, there's the standing instructions, the broader instructions that tell everything how to work. That's that clog MD document that describes how I work with AI. And beneath that, there are multiple levels and layers of additional MD documents, additional markdown files that explain how I communicate. What is my style? Who am I working with? What type of clients do I work with? Each and every one of these things is a standalone document that the AI knows how to call just when it needs it. So it doesn't read all of that because again, you will run out of context in the context window, but it has access to all of these at its fingertips and it can pull information from each and every one of these things as it needs them. So that makes it an incredibly powerful operating system if you want for everything that I'm doing. And it's completely independent from the models on top of it, which is very powerful. We're going to talk about this more in a minute. I want to show you or share with you three of those 31 projects and how they are happening. But before I explain what does the LinkedIn content machine do, let me start with describing the problem. The problem is that in 2025, LinkedIn introduced a new algorithm that they called 360 Brew. What 360 Brew did is it trashed the engagement more or less across the board on LinkedIn. So I went to several different reliable sources, including social insider, hood suite, sprout social, LinkedIn engineering and looked for the information. So here's the findings. Overall, there's been a 50% decline in platform wide post views. There's been a 25% decline in engagement per post. Again, platform wide. There's been a 42% decrease in the rate in which followers get added to people on LinkedIn. And all of that happened because of a change in the algorithm. The LinkedIn themselves did in 2025. I was suffering from the same thing. I was making the same amount of effort, creating the same amount of posts working very, very hard, trying to look for different tips and tricks. And yet I was growing significantly slower and getting significantly lower engagement. And then it hit me that there is a better way to do this. I can do this in a much more scientific way. Look what is working on LinkedIn by analyzing the top performers on LinkedIn, figure out the strategies that they're using and apply it to my content. And this is exactly what I built with the AI content machine. And the results are these. Since the launch of the AI content machine, which again, uses my content, it doesn't generate stuff on its own. And it applies best practices that it is testing and verifying on weekly basis on LinkedIn. I've seen 147% growth in daily impression. That's almost 1/2 x of the amount of impressions I saw before. I see a lot of people. a 37% increase on engagements per day, and my followers are increasing a 21% faster per day than it did before I launched this system. So while the overall average on LinkedIn is declining dramatically, I'm seeing a very nice growth. And again, I was seeing that decline before I deployed this. So how does this thing work? So with this, the easiest way to understand how the system works is actually to look at this really complex and scary flow chart that again, the AI generates explaining how it's doing and what it's doing. So the first thing it's doing is it has what I call the creator universe manager. It's actually continuously looking for the top AI creators on LinkedIn. So these are the people who are in my niche, in my industry, my competitors, and it's building a database of them and it's tracking what they're doing every single day. What are they posting on? How long are they post? Are they using video, image, or text? Are they writing on news or on tools? Or what is the mix of the things they're sharing? How long are each lines in each and every one of the posts, et cetera, et cetera, et cetera, each and every one of the things that he does. And then there's a whole intelligence and data layer that uses this information to define what is going to be my strategy. What are the things that are working right now and how to analyze my content based on these new ideas? It saves that information and it creates a strategy. That then is being used in creating the plan for the following week. So you can see here that I have the weekly content orchestrator that works everything, more or less connected to everything around it. And then we have the content strategy generator that updates the strategy every time the universe of creators that it's tracking is doing something different. And then you can see that there's a strategy optimizer that keeps updating that all the time. And that generates a content calendar generator that generates the content calendar for the next week based on the new things it learned from looking at all these people, which means it has a scientific approach and an on-demand currently relevant approach to how to post on LinkedIn. Now those of you who are thinking, oh my God, I can do this for YouTube and I can do this for Facebook and I can do this for Instagram. The answer is absolutely yes. The same exact process. All I have to do is replace the type of people it's looking at and the platform and it will do the same exact thing across the board. But then it goes and actually generates the content. How does it generate content? As I mentioned, it's not making stuff up. It is actually using my voice, my content. And that's obviously the magic behind all of this. How does it know my voice and my content? I have a podcast that has two episodes every single week it can pull from with my voice and my content. I have the Friday AI Hangouts, which is my community of people who are just like you who are in the business world that wants to know how to use AI effectively. We meet every Friday at 1 p.m. By the way, if you're interested, come and join us. It's an incredible, incredible group of people that is constantly sharing ideas and practical deployments of AI and learning from one another. Every Friday 1 p.m. Eastern time, but that's another piece of content that is available for you to look at. And every time I have an idea of something I want to talk about, something that frustrates me a cool thing that I found. I just open my phone and I click on the button and I record a voice message that goes into this process and gets processed and becomes a post as well. So all these things, it's taking my actual voice, my experience, my background, all the knowledge that it has in the MD files and it creates content that is me, but based on the best practices of the best performing people on LinkedIn or like I said on any platform. And then the final step, it actually delivers it so it deploys it. I'll show you exactly how. There's a human in the loop step, which we're gonna talk about in a minute and it also learns on its own. So it's looking at what is working for me that may not be exactly the same thing as for these top creators and it is continuously updating the strategy that is happening in the very first step based on the results that we are achieving, we meaning my assistant, myself and the AI in our posting on LinkedIn. And I showed you before the results that it's generating. So this is project number one that I wanted to show you. Project number two is the proposal pipeline. I get approached several times a week by companies who want me to do workshops for them or people who want my consulting capabilities and so on and so forth. There is a Zoom call that's happening to meet with these people and learn what they need after they fill up an initial form. There might be some email exchange back and forth between me and these prospects or existing clients who are looking for more stuff. And all of that triggers the generation of proposals again without me having to do anything other than verifying it in the end. So let's see how this looks like. So this is a much shorter pipeline but it is still extremely effective. So let's see what's happening here. First of all, it is scanning my Google calendar. How does it have access to my Google calendar? Through the MCP we talked about before. So I have an MCP connector to my Google calendar and it sees when these type of meetings happen. So either specific types of client meetings or specific kinds of meetings with prospects. Then it goes through N810 and pools the actual summary of the meeting from Fathom. Fathom is an AI tool. There's many other tools like this who transcribes and summarizes everything that I do. It then evaluates whether based on the summary the client or the prospect actually asks for a proposal or I suggested a quote. It doesn't have to be a verbatim. Please send me a proposal. It understands from the context. And if it is, it actually pulls the entire transcript. And then it has a whole process with several different agents. Again, you can see 60 different agents here that actually will research about the client, will learn exactly what their needs will figure out, how that aligns with my offering. And then you will write and draft a proposal. And then it does three additional things. It will create a draft email in my Gmail account and will attach the proposal. It will upload it to the relevant place in my Google Drive. And it will update my CRM with that, all without me having to do anything. So when I wake up in the morning, in my morning brief, I have a message saying, hey, there's a new proposal ready for you to review. And here's the link to the proposal. Here's the link to the CRM. And here's a link to the email, the draft email that's already in your outbox ready for you to review. And if you'd like it, just hit send and you're done. All of this happens basically 30 minutes after the meeting is over. And the reason for 30 minutes is if you can see in the trigger, it gets triggered by a Google Calendar invite. And we don't always actually finish on time. So it actually asks it to wait 30 minutes after the meeting is over to go and run this process just to be sure that the meeting actually ended. And it has access to the transcript. So this is another project. We're going to look at a third project, which is my AI weekly news. If you've been listening to the spot guess for a while, you know there's two episodes a week. One like this, where we go into practical things and how to do things in AI, usually with really incredible guests that share how they do things in AI. And the other every Saturday that shares AI news. How does that happen? I need to find news. I need to cure it. I need to filter news. I need to sort news. I need to summarize news. I need to do all these things. And there's an entire universe of agents and other processes that make that possible. Looking at the flow chart of that, you can see again, multiple steps. There's a step that starts with an input that starts, if you're by me, dropping in articles into a Slack channel or the things that it is researching on a schedule on its own. There's a very, very long list of n8n processes. You can see there's probably 15 of them here that do multiple things as far as grabbing the article, analyzing the article, figuring out whether it's relevant or not relevant to view my audience. And then based on the decision, sending it through a specific pipeline to get analyzed. And then eventually it gets deployed to an application, which is a web application that Cloud Code created based on my requirements that I keep on updating every single week that has all these articles where I can highlight and shift stuff for our own and decide on the sorting order, which it's recommends on its own. And then actually produce the podcast by allowing me to review this in the right order with the right highlights, with the right focus. And it also produces the newsletter that tides to it that I know a lot of you read as well. So this pipeline does a lot of that work again while I sleep. And I just have to verify, see the things, and obviously employ my judgment to what I think would be more interesting and less interesting. So liking the other places, you'll see a lot of collaboration between myself and the AI in that universe. Now, and this is actually a good segue to the really important fact that despite the fact there are dozens or maybe hundreds of agents working in orchestration doing all these things, I'm still in control, the human. You, the people who work for you with you are still in control. And the way I do this is I have human checkpoints for basically everything that I'm doing. As I mentioned before, in the proposal, the proposal doesn't actually get sent. It gets placed in an email in my outbox as a draft for me to review. But there are much more granular things that I'm doing. And the way I'm doing it, I'm actually doing it inside of ClickUp. So obviously, if you don't know, ClickUp is similar to Notion or Asana or Jira or Monday.com or Microsoft Planner, it is a board or multiple boards and multiple panels in which I can define different tasks and define who's in charge of them and view the statuses of all of them and communicate about them and so on. So I work with all these AI agents in the same exact way I work with human teams. I have a board for each type of process. And in that board, there are steps that are human for review and there are steps of actual work that the AI agents do. So in this particular case, we're looking at the a at the LinkedIn content production process that we discussed before. And you can see that there's planned posts that the AI created. You can see prompt reviews that I asked the AI to create a new prompt and it created a new prompt for it. What do I mean by a new prompt? If we go into any of these posts, you will see that it has multiple components. It tells me which pillar of content in belongs to. It gives me a lot of information about it. It has the actual suggested post that again is based on my information. And you can see down here on the bottom, you can see where it comes from. So you can see it says which pillar of content. This is AI tools. You can see the overlap score. The overlap score basically tells it. How aligned is the content that it found from my content? How aligned is it with the best practices and the things it found with the best performers on LinkedIn right now? But what you also will see, this one is actually not a good example, so I will look for another one, that is actually already produced. What you will also see is that on the bottom right there is a prompt in the chat suggesting what should be the image for that particular post that again, the AI decides based on its best practices and the strategies that they develop, should it be a video which it can create on its own based on my original videos or should it be a image or just text and so on. So it suggests that and if I approve it and all I have to do by approving it is dragging it to ready for graphics and then it goes and generates the graphics. And that might be a simple flow chart, this would be a photorealistic image, it can be whatever it thinks is the right thing. So I will actually give my comments, how do I give my comments, I actually write in the chat of the task just like I would do with a human and say, oh, I don't like this, I like that or I will just rewrite the prompt myself or edit the prompt and send it back and it will generate a new kind of graphics. So I work in this board as if I had a team of an entire marketing team working on this including designers, including researchers, including content writers, including stylists, including all of that. So I take it in the back end to my brand guidelines to how I want to communicate to all the files that we talked to before and then he knows how to do it down here as an example, you can see a voice note. This is one of those where I just had an idea and I clicked a button and I recorded and it does the same thing. It goes and creates a post based on me just speaking away and you can see the transcript of what I was speaking and you can see that it actually wrote an actual post based on that that can go and be posted on LinkedIn. So I just did it myself and you can see here that there were several different variations of the actual graphics because I sent it back and I didn't like it and I wrote comments and then it went and fixed the graphics until I was happy with it. This healing in the loop aspect of it is a critical critical aspect of actually staying in control of what it is that you are trying to do. And so while the AI does the work, you still have full visibility to everything that it is doing. And now to the critical aspect that I talked about before, you do not want to be locked into either Claude or Chachypeti or Groc or whatever the next thing that's going to show up or Gemini because of two reasons, one redundancy. If tomorrow something happens to Chachypeti and it's down for three hours and your entire operation depends on it running, you are screwed because for three hours your entire company doesn't work or critical aspects of it don't work. And three hours is not five minutes where you can recover from that's a very long time to be down. So by having the ability to switch to a different model and everything still keeps on running gives you that powerful redundancy. More or less every single process that I built is built with redundancy on top of it so it can switch to a different model and still keep on running. And the way it works, if the process gets stuck and the API doesn't respond, it tries again. If it tries the third time, if it doesn't work, it goes to its secondary model and keeps on running from there as if nothing happened and then an hour later we'll check again if the original process that is its first choice runs and we'll check again another hour until it's back up and then we'll switch back to the original model. So redundancy is one. The other one is improvements. These models keep coming up with new models every single time. The new models get better and in many cases cheaper as well so you can get better results for less money. So why not do that if you build everything locked into one environment, you're stuck with that. And so that's another reason to build it the way I'm describing it here because the actual model doesn't matter. I can switch it with very, very little effort and still have everything running because everything is in that file system and everything is in those processes that I created and the infrastructure I created. The bottom line is the mode isn't the model. People think, oh my god, you're doing so great because you're using Claude. No, I'm not doing so great because I'm using Claude. I'm doing so great because I built this entire system that I'm sharing with you right now that currently runs mostly Claude by way not only. But images as an example that you see in this presentation that Claude created the presentation, the images are created by Nana Banana and image 2.0 by OpenAI. It generates images with both of them based on what you think will fit the slide. It looks so both of them. It suggests one. I get to pick the one I like. And that's how the process works. So even in this presentation that I'm using here, there are several different models in the background that are running to do different things while allowing me to make the final choice. I wanted to share with you today quick recap and a quick reminder. If you want to learn really how to do this, not so far as a 40 minute overview, but rather how to do every single step and to get the actual prompts and to get the actual links to things and to experiment and to learn how to build these kind of systems. Come and join the multi agent orchestration course as I mentioned the next session starts on June 22nd and it's selling fast. So if you want to join that, you need to act today. By the way, if you sign up between now in the end of April, which is just a few more days when this episode goes out, you get 200 dollars off. And the link to how to sign up is in the show notes and the code for the 200 artist count is on that page that is in the link in the show notes. So if you're interested, come and sign up today. The first two cohorts sold up in 10 days and this one is already selling pretty well. So if you want to join this would be a good time to join. I can tell you if this didn't tell you enough what you just saw in the past 40 minutes, it changes everything you know about how businesses run, about what you can do about the level of efficiencies, about how much growth, about what kind of teams you can have and so on, limitless. That's all I can say. So if you want to learn this, come and join us and that's it for today. I hope this was eye opening and educating at the same time. I was trying to reveal as much as I can in the time that I had. So if you don't want to join the course, at least it gives you ideas on where to go and what to do next on what kind of infrastructure you need. If you like this episode or the spot cas in general, I would really appreciate if you do two things for me, one go into your favorite platform, whether it's Apple podcast or Spotify and give us a review and the second one, which is a lot more important is share this with other people who will benefit from this on your phone or wherever it is using this as a share button. If you want to click on that button, choose your favorite tools of communication, whether it's email or Slack or WhatsApp or Telegram or just regular SMS messages and send this podcast to people that you know can benefit from this so they can learn as well. They would appreciate it. I will appreciate it and you will feel good about doing the right thing. That's it for today. Have an awesome rest of your day and have great.