How I Built a Marketing Team of AI Agents Using Claude Code

Key Takeaways

Hello. A lot of you have been asking how I'm using Claude Code as a marketing engine and why. And I'd love to show you how this works and give you a peak under the hood so you can get an understanding of why this is so powerful and why managing context is going to be a better use of your time than managing prompts. Let's get right into it. Going to launch Claude Code by typing claude. I have a folder undercloud which is called agents. And within this folder are several different agents, which are really markdown files that follow a specific format. And each one has the instructions for what this sub agent's job is. So if I go down to the SEO optimization specialist, this block of text is a description of what the agent is. And then if we scroll down, we see instructions. And then there are variable names that get filled out or referenced when this instance is set up for a certain brand. And the other flavor of agents are command line agents. And these are what I use to proactively operate the system. The difference is that I can activate them manually as opposed to sub agents which are activated autonomously based on whatever work the AI is going to do. You still with me? So I have an onboarding agent here. For example, when I type onboard, because it's set up as a slash command, all I'm really doing is activating the onboard agent. And the onboard agent's job is to set up this instance of the marketing engine for a specific brand or client. And the first thing it's going to do is to check if it's already set up or not. And it'll behave differently based on what it finds. Now, we're presented with several options based on the onboarding agents instructions. We could add materials to the intake folder to set up this instance, which is the recommended path. We can do an interactive setup and it'll provide a guided onboarding journey and ask you a bunch of questions or we can do a quick setup if we want to get going right away or you can just interact with it however you want. So if I give it my website, it'll get an idea of what the brand or business is. Now the onboarding agent has activated the website analysis agent and what this means is a separate autonomous session has been spun up behind the scenes that has special instructions about analyzing websites and it has been given my website. It has its own dedicated context to do only its job. Now the website analysis has finished its job and the onboarding agent now has used that to recommend a brand framework that's the best fit for my brand and it's recommending the story brand framework. Other options are the golden circle jobs to be done blue ocean. They're all templates somewhere in the project directory so more can be added and these can be edited or changed or whatever. I'm going to go with his recommendation his and just hit one. My story brand framework architecture is listed here. Here's my messaging framework that's based on the story brand company profile and a foundation for an SEO and content strategy. And again, this is with very few inputs. It works best when there's many inputs and more inputs can be added over time and integrated into the context. And all that really means is there's several documents that are well organized in such a way that anytime there's an operation that's executed, it knows what to pull and where it is. But now that that's all created, I'm going to clear to work with the CMO agent. The CMO agent is your main co-pilot for the marketing engine. And it's activated by typing / CMO. And what that's doing is configuring the chat session with the instructions from the CMO markdown. It's like a system prompt, but the difference is that system prompt has full access to all of the context that we set up. It will know what's in there and what to do with it based on what I ask it to do. And it's set up with core journeys here. And these are predefined journeys that are fleshed out in an orchestration file. So the CMO can read the orchestration file. It says, "Here's the process you follow and here are the sub aents that you need." It's really just a system of organized markdown files that are being activated in a certain order. So let's say I want to do a lead generation campaign. I'm just going to hit the number two. Okay. So we have several options here. seasonal campaign, product launch, brand awareness, lead genen, etc. I'm going to go with lead genen. And now the CMO is assembling the specialist team as I described. And it's giving the details of the team that it's going to assemble as well as an a time estimate on how long it might take to complete this entire operation. Now, the first sub agent that it's using is the conversion flow optimizer, which it highlighted in the plan. Now the content marketing strategist has been activated and it's using the inputs from the conversion flow optimizer as well. And in cloud code if you hit controlt you can get a to-do list of what's happening. And now it's calling in the email marketing specialist. As this is coming together you can see what it's starting to create a lead genen campaign strategy lead magnet content strategy an email automation sequence. And I'll show you what all of this looks like. I just want to note that as I'm filming this, Anthropic released Sonnet 4.5. Excited to test this out. Wow, 21 minutes. We're moving on to the next item, which is ROI and tracking framework. A new sub aent is activated for this. These sub aents can work in parallel sometimes, but in this flow, each sub aent is working off the inputs of the one before, so they're going to run linearly. All right, so it finished. We have a full inventory of all the files that were created and there are several. We have a summary of the outcome achieved and we have our next steps and this system will completely handhold you through the entire development and execution. Anything that you need to do, you're never lost because you can always come back here and use this like any LLM. Let's take a look at the strategy document first. Every campaign comes out with an implementation guide or some master document that gives you the exact steps to follow and lays it all out for you. This guide starts with a conversion objectives and lead goals, an overview of the recommended CPL model, budget optimization strategy, the multi-channel lead capture strategy, etc. And if we look at the other files that came out here, we have an entire email nurture sequence with copy subject lines for each email in the sequence. We also have an email welcome sequence with seven emails all written each with several subject lines that could be split tested. We even have reporting templates for daily monitoring. It even created an HTML dashboard that we can use for monitoring. One thing I didn't mention in the last video is that the marketing engine has a custom API integration with air table meaning that it can take content and push it and store it in an air tableable record. And once that content is in an air table record, it's organized in a calendar view that features an approval workflow. So team members or clients, whomever, can collaborate on content feedback and approvals. Additionally, I can use this to generate images, convert the content into other formats such as blog posts to social media, and then publish directly from Air Table. And that's powered by an innate end workflow that activates one of these automations based on the action that the user chooses in air table. So I hope that this was helpful in understanding the power of marketing context engineering, how I'm using cloud code for marketing and to highlight the endless possibilities that you can get working with AI in creative ways like this. If you like this video, please like and subscribe. I'm going to be putting out a lot more content diving deep into cloud code for marketing. Thank you so much.