What Is an AI Agent? Build Your First One Today (No Code)

Why drink out of a normal cup when you could have this strawberry cup? Hi. So have you been hearing about AI agents and you're just kind of like, how do I build one? I don't have one. I'm behind. How does everyone else have one? Is everyone running their business with AI agents but me? Well, don't worry, friend. I'm going to help you today. My name is Robyn Anderson. My friends call me Jam. I'm the CEO of Jam AI. I run a really cool community called Club Jam where we have small business owners building stuff, creating agents and growing together every single week. So let me get into what an AI agent is and what it's not. I'm going to share my screen. Okay. So an AI agent is really trying to be the same thing as an employee, right? And let's talk about how you would get an employee and how you would create an AI agent, because it's actually really similar. So if you were going to hire someone, you would have to write the job description, right? You'd have to say what the job is and how they would do it. Then you'd have to interview someone and you would hire, and then you would do onboarding, right? Because let's say you hired Sarah from down the street and she's an intern. She doesn't know much about your business, right? She doesn't know who your customers are, how much your products are. So you have to do onboarding. You have to train your employee on your business and their role. You also have to give that employee tools to do their job. So maybe Sarah needs a laptop, and with that they're gonna need a Gmail address, and with that they're going to need access to your HubSpot account, right? So you have to give them all the tools, and then you are not around all the time, right? So you need to give Sarah access to content. And context. So think of SOPs, right? So standard operating procedures for different parts of the role. So if you're not there and a customer comes in and has a return, how would Sarah know what to do if you weren't there? Well, you'd have some sort of SOP for that situation. The funny part about an AI agent is that an AI agent needs exactly all of these things too. And that's how you would build it. My favorite thing to do is start with thinking of that dream employee. Who would it be? And what would they do? So write the job description before you start building the agent. What is that job that you need done? So let's say for context today, we need an agent that does digital marketing, so we could think of you know, blogging, SEO, and your Meta ad management. So you would write a whole job description about that. And then you don't have to interview it, right? You can skip that part, but you do have to do onboarding. You would have to give this agent access. So if it's doing marketing, it has to have access to your Meta ads. It has to have access to your social media profiles. It has to have access to your website or your Substack, wherever you put your blog, right? And so in order to do its job, it has to have access to these tools. And then it also has to have SOPs and instructions on how to do each one of those things. So how do you have it create meta ads? Well, do you already do meta ads? Because you could kind of write down your process, but I actually like to do it a little bit backwards. And in fact, I like to have AI interview me about how I'm doing the job, and then I will ask it to improve my process and write the SOP. So this is step 1 of building an AI agent. It's a lot like hiring an employee. So now you have the the job description, you have the tools, where do you build it? So you can use any sort of large language model or harness that you are familiar with and that you like. So the most popular ones right now for non-technical people are Codex, ChatGPT, ChatGPT Work, so that it has access to your computer. Or Claude in Claude Code. Side note, I'm recently just doing a couple consulting jobs with Copilot, and those come with AI agents as well. So there are some AI agents that you're seeing showing up in these large language models that then you can customize. An example of one was a daily planner. So it's kind of like an executive assistant, and it goes into your calendar and your email and creates kind of a plan for you. So once you figure out your large language model that you're going to use for today, we're going to use Claude as the example. You would have to create the agent in Claude. And for that, I'm actually gonna share my screen and show you how I do that. So here I am in Claude right now. I run the desk— the desktop version, not the web version. And I work mostly in Claude Code. However, I do no coding at all. And how I get started is I actually upload the notes notes that I just wrote about my employee, or I actually will have a conversation and say, hey, this is the type of employee I want to make. Will you help me do it? And it goes back and forth and it actually creates the overview of the agent for me and with me. And it also helps me think through what sort of tools and connections it might need. And here's the part where people miss. Okay. And this is the most important part. The agent can live in Claude, right? It can know that it has to do these things, right? So it's AI with a job. But I like to have a file system on my computer where these agents actually live and store information. So when you work inside Claude, it has a memory about you and what you do. And if you want to learn more about how to set up the memory and learn more about the context and how that works. I have a training called Zero to Claude in Club Jam, so check that out. But it has a memory, but the memory has a limit, right? There's only so much it has room for. But if you think about building agents for every department of your company, that would fill it up pretty quick. So what I do is I actually use a tool called Obsidian, which I will show you. And so Obsidian is nothing more than a file structure on my computer. Obsidian is a free app. It's been around a long time and it's just a trusty file system. The reason I like it is because I am such a visual learner and Obsidian puts everything right in front of me and I could add images, videos, links, not just files and, and things like that on here. Because like, if I wanted to save an entire page of links that I want my marketing agent to go to, I would have to create a document and then put it on my desktop and save it. And it just isn't as easy to access. Also, everything it creates is an MD file, which is how large language models can read and learn really quickly without burning tokens. And so this is my Obsidian file, and Claude helped me create my vault, which is called Sutton. And inside my vault, I have all these different agents. So these are the agents that run my company. I have 2 humans at Jam AI, and then we have these other AI agents, and we literally run as if we are a team of 20. So this is a beautiful document. I did not write it. Claude wrote it with me, and we built this entire structure and folder system together. How I did that was I actually went over to Claude, and all you have to do is click on— show you this. All you have to do is click on this little folder icon, and you could give Claude access to your desktop. And so whatever you click in here, it will have access to that entire desktop. So right now I just clicked Obsidian and inside it has access to all of the different vaults I have. So vaults in Obsidian are basically just folders. And so Sutton is the one for Jam AI. So now when I have this connected, it has all of my different agents and everything that it needs to have so much memory on my business. And not only that, it can see everything that those agents have made. So I could simply say, find the presentation that I made on AI 101, and it'll look through my file and folder structure in Obsidian, and it'll bring it to me, tell me when I made it and where I could find it. So it says, here's the source in my vault. So it's in Stasi, and I actually like to build webpages for all of my presentations. So there it is. The other really cool thing I can do is I could have my agents work together. So I have my Jam AI Obsidian vault, and I could say, have a meeting between Bozma and Carl and make sure that we're on track for our Black Friday push that will start in late September. So it says, okay, I'm digging into your Obsidian vault. So it has like all the stuff that you make, cuz I know you guys, we're all creating so much content in these AI tools and most of it goes unused. So I love that when I create content, it knows to push it to that agent's file and it's always updating too. Like for instance, I moved platforms a while back from email marketing. And so it went and it updated all the agents on how we work now. It updates the SOPs for me and it updates the agent's memory files for me. So right here it says, good, got the context. Now running the meeting, briefing both chiefs in parallel. Running agent. While that's running, I wanna show you a sneak peek of what it's actually looking at from a context perspective. So Bozoma is my marketing person. Carl is this dev. So you can see in here when I open the Bozoma file, there's so much stuff. So much stuff. I have images, I have blog topics, I have everything that an entire marketing team, not just one person, would have to know and understand. So it has images, it has one-sheeter ideas. It has my, my blog entries. It has a whole thing of my website migration that I did. And you can see over here, it's not super organized. And I think it's funny because at first it drives me crazy, but then I realized this really isn't for me. This is for AI. So if it makes sense to AI, then I don't need to worry about how it looks and how it's labeled. And then if I go over to Carl, same thing. Carl has his own ND file. He has a brief that, um, Claude made for him. And then we have different offers and launches, and it has all of these things that it has to know about. The other cool thing is that I have a whole outreach program that Carl runs. So he does cold ideal client profile prospects. So I have a AI tool that continues to feed this list of prospects, and Carl will write them cold emails every day. But the really cool part is he actually researches who they are and decides which offer they should see. So if I have a bunch of real estate people, of course he's gonna show them our real estate offer. If they're small business people, he's gonna show them our Club Jam. And if it's a big team, he'll show them our Claude Sprint for Teams. So it's pretty cool how they know how to do that. Back over here, it looks like Bozma and Carl are still in their meeting. So while they do that, I'm going to refill my strawberry drink. It actually looks like they're having a disagreement, so they're working it out. I hope that I can figure out what they're, what they're having conflicts about. I hope I don't have to fire anyone. Okay, so they're not 100% done, but this is why you need AI agents and why you have to give them so much context and so much information and update it all the time. So one of the things that they're blocked on or that they're arguing about is that number one, this, uh, new link that I'm running for my Black Friday, it's not on the sitemap at all. So I was over in Claude Code developing stuff, making stuff, going hardcore on my, uh, marketing, right? But I didn't follow my own SOPs because I'm not a robot. And I didn't go back and say, oh, now that I created this page, I should add it to my sitemap, right? I didn't think like that. I thought, oh, this is a really cool offer. I'm so excited that I'm so prepared and getting this done so early. And then they also found out that my landing page has no meta pixel on it. So I wouldn't be able to track any of the ads that I run. And then there's also zero email capture fields on the product page. So that goes against some of the rules that we have for our SOP in marketing, right? So that's just so cool. They have all the information as if I'm working with a CMO. Or a business development director. They know all the rules. And yes, I should know the rules. It's my business, but I also have so much going on. That's why people have employees, right? So my employees right now are saving my butt and helping me recreate my Black Friday offer. All right. So let's go back to what we've covered so far. So what we covered so far is we have agents. And then if you want to build one, you need a job description and tools, right? What are they going to do to— what are they going to do when they have the job? How will they do the job? And then you have to go and create the context MD file or the brief that you saw, which is the instructions for the agent, but then also SOPs for the jobs it has to do. And that's something that your AI will build. And then you have to have a place where it lives. So the place could be somewhere on your computer, just a file structure. It could be Notion, it could be Google Drive, or like I'm using Obsidian. It can also be like GitHub as well, if that's something that you wanna use. And the next thing I wanna talk about are how once you have that agent, it can create routines and workflows. And then I'll finally end with how this whole system helps you be platform agnostic. So let me show you just the routines and work. So we're back in Claude. These guys are still working, but I'm gonna show you a workflow. So in Claude, you can go to routines, and in here I have a couple that I really like. So I'm gonna show you a really simple one. Every morning at 7:00 AM, I have a routine where basically Claude is going into Bozoma's folder. Right, my marketing director's folder. And she's supposed to write a blog based on this SOP criteria, right? So these are, these are the rules. I want it to follow all of these rules for AEO and GEO and SEO, right? I want to be found in Google searches. I want to be found in large language model searches. And so this was what I told it it had to do. In order to write a blog post. And then it's put publishing and deploying on its own. So it also has the tools, right? It has access to my connector. So it goes to GitHub and then it goes to Vercel for production, and then it goes live. So it's actually finding my blog topics inside Bozoma's folder. It's picking the topic that I've already approved. It's writing the blog in my style according to my rules, and it's publishing it every single day. And then it'll notify me and tell me that it's live. So you could see today it ran at 7:07, and here's the blog that it launched. I'll show this to you. So here's my blog and it's literally posting every day. Based on things that I've trained on. So because Bozoma and my other Obsidian files, um, and agents have access to all of my training content that I've done since 2023, it's taking my content and rewriting it every day. So I never miss a day of blogging, and I am expanding my digital footprint every day by doing this. And the larger your footprint, the more you'll be seen. Another workflow that is super important to me is I have an AI agent named Vicki, and she's pretty much in charge of all my other agents. And her job is at the end of every day, 8:37 PM, I go to bed early. She runs through all the chats that happened in Claude Code and makes sure that every agent has updated its briefs, its files, and created new content from the conversation. In my Obsidian folders. The last thing I want is to work so hard and then have it all in a complete system. And that kind of leads me back to my next point of being platform agnostic. So when you have a tool like Obsidian or Notion and you're constantly backing up your content from your large language model, it allows you to be free from any specific platform. So that's where I was going with this, and let me just kind of explain what that might look like in the future. So everyone's really into Claude right now. ChatGPT is having a moment again. They're kind of back. Codex is wowing us. And I think 6 months from now, we're going to have one place where we can get the full picture of all of these models without having— because we kind of have that now, but it's— we don't get the full picture, right? We can't do everything that we can use Claude desktop version for if we're using its API. So in the future, I think we will have the ability to do that. So I do think it's really important to get to know all these tools, understand what you like about them, how you use them. There's so many. We train on Manus, NotebookLM. Guys, Copilot had to come back too. It's not as bad as it used to be. Um, but we would use whatever harness this is and we can bring in API keys from Higgsfield, the photo generator. We could bring in API keys from HeyGen. We could bring in all these other tools and it'll all just happen in this one spot. And the coolest part is no matter what we're putting in here on the back end of this, Don't you like my drawings? I'm like so artistic. On the back end of your computer, here's your computer on some sticks. You have your Obsidian, or you have your Notion, whatever you're using. And so whatever we're pointing at it that day, from a large language model perspective, you don't have to retrain all of these. You don't have to tell it who it is. Or who you are. It already knows because it's reading this big file structure, which is your AI agents. And so at the end of the day, if you're breaking down what an AI agent is— I type too much. An AI agent is a file. It has content in it, like past examples. If you're starting out, past examples. Of what good is. It has tools, connectors. It has an SOP, has context for its job, and it has some sort of LLM on top of it that is making it alive. And that's it. That's what an AI agent is. And once you have it set up, you can create one for Uh, marketing. Here's kind of the ones I see people do: marketing, sales, um, executive assistant. We have operations, we have accounting, we have HR. I have a community one, like whatever you need in your business. But it all starts with— and this is what I train people on— it all starts with going to a coffee shop, get away from a computer, and actually write down what roles need to be filled. The last thing I'll leave you with is stop asking what the AI can do. Stop asking that. Stop asking what Claude can do in ChatGPT. Start worrying about what systems you need. Start worrying about what tools you have and why. A lot of times when I do this exercise with my community, people will realize that like They're paying for stuff and signed up for stuff they don't even use. Somebody had a Mailchimp account, they've never logged in. So make sure you have your systems in place. Figure out what you need to make that system work better, and then put the large language model in front of a complex file structure. And there you have it. You have an AI agent. I hope you found this helpful. I'm sorry my handwriting is so bad. I probably should have been a doctor. If you wanna learn more and learn in an amazing community of other small business owners, check out Club Jam, and I can't wait to see you inside.