Claude Code: Build Your First AI Agent in 30 Minutes (Zero Code)

Right now, I've got eight unread emails. Watch what happens when I type in slash email agent in my Claude Code session and send it off. This is an AI agent I built in Claude Code without writing a single line of code, and it's exactly what you're going to be able to do in the next 30 minutes. It's reading my memory, voice guide, and references, and it's going to fetch my inbox threads, label every reply, write drafts, and summarize everything it did in a matter of seconds. As you can see, the agent just added labels to all of those threads. If I go back and I refresh, now every thread has a label. It got my inbox completely organized, gave me responses, and it did all of this without my intervention. And if I refresh again, we now have drafts ready for every single email. We can see that it understands my intentions as well. It says, hey, Sam, I'm not doing any podcasts for the rest of the year, so I'll have to pass on this one. I appreciate you thinking of me, though. Good luck with the show. With a smiley face and my sign-off, everything's in my voice. built exactly the way I would respond. In this video, I'm going to walk you through the four levels of building an AI agent so that you can repeat this process every time that you want work handled for you. So we must understand, what is an AI agent? Normally, you ask AI a question, and it gives you an answer, but you still end up doing the work yourself. An agent does the work for you. You give it a job and it uses your apps and your tools to get it done, just like you saw happen with my inbox. And you do not need to be a developer to build one, because you just tell Claude what you want in plain English, and it develops it for you. Every single agent you build will start with a workspace, and that is level zero. Level zero is to set up your agent workspace. When most people try to use AI, everything ends up scattered across a bunch of random chats. So every time they want something done, They start over from scratch and they can't find what they made last time. It's like cooking in someone else's kitchen. You spend half the time opening cabinets and drawers, looking for pans and utensils, but in your own kitchen, you just reach for it because you know where everything is. A workspace is your agent's own kitchen. It's just a folder on your computer that your agent works out of, so everything it needs and everything it makes lives in one place. It's as simple as making a folder on your computer and naming it after the job. If I want my email agent, I will just name this Email agent. There you have it. This is my agent's kitchen, my agent's workspace. It's going to know where everything is and exactly where to put things back. Then what you need to do is open the Claude desktop app. As you can see at the bottom of my screen, I've downloaded Claude to my computer, which makes it 10 times more powerful. After you open the Claude desktop app, open your sidebar in the upper left-hand corner, and then switch to the code tab. So I'm going to select code. And next what you need to do is select choose a folder to work in and select that folder that you just created on your computer. So I'm going to select choose a folder to work in. I'll hit open folder at the bottom of my screen. And now I've found my email agent folder. I'm going to click it and then select the blue open button. Then I will hit trust workspace. This is going to be your agent's home now, and you didn't have to write a single line of code to set it up. So let me show you the benefit of working out of a folder using Claude code. Number one is you can create files with Claude code and it can add it to the folder. And so I could tell it to create a document. I'm going to say, create a document I have a document labeled Drake's Favorite Foods and put... Salmon, ribeye steak, and soba noodles. What you're going to notice is this Claude Code container can create files for us. Just like that, it's created this document. When you're working in your agent's environment, there are a couple of things you must understand. When you are creating things with Claude Code or visualizing things, it can pull up windows like this on the right-hand side. That way, you can actually visualize what it's doing. I want to put save this to My folder. And that will save it to its working directory. So anytime we open up a new chat, it will understand and know all of my favorite foods. Now, this is a silly example. If you have a business, think about your clients. Think about your analytics, all of the information Claude would need in order to become a useful agent. So it saved it to my folder. Now, if I click my folder in the upper right-hand corner, what you're going to see is we have Drake's favorite food right here. And this is just a bunch of code Because it created. A nice fancy HTML file, so whenever I open it, you can see it's this nice styled document. If you want it just text-based, you could also say I want this text-based, switch this file type. to markdown and markdown is a file type that we are going to be using a lot today it is natural language AI agents can read it humans can read it as you can see we have drake's favorite foods salmon ribeye steak and soba noodles now what happens when we click my files in the upper right hand corner is we have a markdown document that we can actually read in here so that's very nice so although this is kind of a silly example i want you to think about what we just did and how this agent is going to begin to work. Because now, whenever we open up Claude Code, we can put it in this folder. It can create files, it can read files, but a folder can only hold what's on your computer. And your email, or your calendar, or the tools you use, those aren't files or on your computer necessarily. They're a tool in the cloud. But Claude can't see any of that right now. And if it can't reach it or it can't see it, then it definitely can't automate it. Level one is to connect your tools. Let's get into level one, which is connecting the tools that you need Claude to read from and write to. If we wanted drafting emails for us for this email agent example, what I'm going to need to do is connect my Gmail. So in order to do that, you need to open up your sidebar and then you need to go to the customize tab. From here, you can select connectors, and connectors are what give Claude extended capability to make it actually useful. And as you can see, if you look in the upper left-hand corner, there are over 9,600 connectors, different tools that you can connect to Claude. As you can see, the entire Microsoft ecosystem. You have Stripe, Shopify, HubSpot, Salesforce, Monday.com. You have literally everything known to man. that you can connect your Claude account to so it can begin taking actions for you. And so whenever you want to create an agent, you need to connect tools in order to make it more powerful. And so I'm going to connect my Gmail. I'm going to select the Gmail. As you can see, it's already connected, but I will disconnect just to show you how this process works. Whenever you want to connect to a tool, all you have to do is hit Connect to Claude. It's the exact same with every single tool. You're going to hit Continue Connecting. And then it's just an authentication page. You sign in like you're signing into the actual thing that you want to connect to. And for me, I want to handle in my support inbox and I want it being a support agent, thing that can give refunds, a thing that can give discounts if it actually meets criteria worthy. And so I'm going to give it my support inbox. Then I'm going to hit continue. I'm now connected. And so when I go back to Claude, we can see that I'm connected. The even cooler thing about this is you can decide what Claude can touch. If you're nervous about giving AI your inbox, for instance, that's fair. So start by only connecting. What this agent needs. The agent we're about to build will also ask me before it sends anything. So I'm going to be always the one hitting send. The only thing it's going to do for me is draft my emails so I can review them. And I have all of my drafted emails sitting there so I can just approve rather than write. So if I click this button right here that says your tool permissions, I can go to this page, which allows me to change what certain tools have access to Claude. And tools are things like this. Do I want Claude to be able to get draft emails from my account? Do I want it to be able to list draft emails? Do I want it to be able to list labels? There are two types of tools. We have read tools and we have write tools. Read tools are where Claude can only read that information and bring it back to the chat. Write tools and delete tools are actions Claude can take to your account. So can it create emails? Can it send emails? Can it apply labels? You get to fine-tune. granularly go through and always allow have it needs approval or block the tool entirely for each tool that you want for me forwarding emails i turned off i don't want it forwarding emails to anybody replying right away and sending emails i have turned off completely so my agent doesn't send anything without my approval first now the unique thing about connectors is once you connect it to your Claude account it's connected to every folder that you create and so If I wanted to, I could go create a different agent that uses the same email in a different working directory or workspace for the agent. And my account's already connected, so I don't have to worry about connecting it again. Now, let me just say this to Claude. Check my email and give me my five recent emails while blocking out any sensitive information. I want to show this on video and I can send it off. And what I like doing when I connect my tools is first testing them out. So right now I'm in my email agent. I'm just testing this out. The agent is not an agent right now. It's just a working directory where we will build the agent later. Right now, it's fetching my five most recent emails, redacting sensitive details, and then presenting me a clean summary here in the chat. And it pulled in all of my recent emails, redacted all of that information. And so I want you to think about this. It can pull in my emails and I can feed Claude my reasoning. I can give it things like my favorite foods, right? Or the way I like to talk or how I respond. And then it can draft responses for me based on the emails that are coming in. But seeing your emails is definitely not the same thing as handling them. Right now, Claude can read my inbox, but it really has no idea how I would answer any of it. And that's what the agent is for, and that is level two. Level two is to build your AI agent. This is the part most people think takes and requires code. When they hear build an agent, they picture a developer I'm writing a program, but an agent in Claude Code is really just three different files sitting in a folder written in plain English, and you don't even have to write them yourself. I wrote a prompt called the Agent Builder. That does it for you, and it's free in the description. What I'm going to do is open up my sidebar and hit New Session. Keep in mind, I'm still using my email agent folder right here. I'm going to paste in my Agent Builder prompt. Again, I left this down in the description. I'm just going to quickly go over this and everything that it creates for you and everything that makes up an AI agent. So first off, we have this prompt right here. I want to build an AI agent in this folder. The job I want to hand off is, and then this is where you need to describe the job that you want to hand off in a sentence or two. For me, it's, I want you to label my emails and write replies in my voice. So this is my job description that I want to hand off. I want to create an email agent. Yours might be different. Frank, yours might be. A sales agent, yours might be a personal assistant, virtual assistant, whatever you want. I then say, I'm not technical, so explain anything technical in plain words the first time it comes up and do all the setup for me. There are five steps in this agent builder prompt that I want to explain and I want you to understand. Step number one is it's going to interview you about this job. It's going to ask you one question at a time, eight to 12 questions, in order to gather enough information to handle a job like you would. Step number two is it gathers examples from your tools. I tell it to look at all the tools I've connected. If any of them hold past examples of this job done well, tell me what you found and ask before you save anything. So it's going to use the tools that we connect in order to gain information and knowledge about the way we work. Number three, ask me for anything else you need. Tell me which files would help you do this job. And so since I'm doing an email agent, that would be things like policies, FAQs, templates, prices, my course transcript files, so it knows what to tell people who are asking questions about that, anything in between. And then this is the part that is super important that you must understand. I then say, build the agent as three files, create these all inside this folder. And it knows what I mean by this folder because we set that folder as its working directory. We have file number one, claude.md. This is the first thing that Claude will read anytime you work. A claude.md file is an instruction file. Anytime Claude enters a folder and sees this claude.md, it's baked into its backend instructions to look at that and follow whatever. The claude.md says, this is the agent's personality. This is the agent's rules. This is where everything is learned about us for the way that we want this agent to work. Claude.md. In this process, after it interviews you, after it gathers examples and asks you for anything else, it's going to create this instruction file based upon that interview. Claude.md is super important to have in your folder. When we create this agent and run you through this entire process, I'll show you all of these files as well. The next thing I want it to create is a skill for the job. A skill file, by the way, these are all like things that I didn't make up. These are all. Anthropic created. Claude understands what each one of these naming conventions is. Claude.md, skill.md, and the third one's memory.md. It understands how to work with these. This is not something I made up. A skill is an automation folder. And so it's going to save the skill in this folder so it understands the step-by-step way for doing the job every single time. What to read first, the steps in order, what a good result looks like, and how to use my examples. For example... If I'm having it do my emails, I want it to write like me, so it needs to store the past examples somewhere that it can go and reference those using the skill. In a skill file, all it is is literally a step-by-step workflow for the way that you do things. So like step one, open the email. Step two, find any unread emails. Step three, open the unread emails. Step four, draft a response in Drake's voice using all the files you have access to. That's what a skill is. It is an automation document. And the deeper you learn about skills, the easier your life is going to be. I promise you. I then say, give it a short name. I can type with a slash. So in the Claude chat, if you come down and you just type in slash. What you're going to have pop up is a bunch of things you can run and you can create skills that run for you. I've got my daily brief skill. I've got my Don Draper skill that writes high converting sales copy. I've got my skill creator skill, which was built by Anthropic, YouTube deep dive, podcast summarizer, presentation creator. I've got so many different slash commands that all I have to do is type in the slash command, hit enter, and it runs the job for us. So in this case, It would go and scrape my email and respond and create drafts for me. Then the third thing that this prompt is going to create is a memory.md in the main folder. And this is where Claude is going to write down lessons from my feedback. So when I'm telling Claude things that it's getting wrong Or if I notice it's doing emails in a certain way that I don't like, that it's getting wrong. I'm going to tell it that, and it's going to store it in the memory.md, and this is Claude's brain, so it's going to read it every time it tries to do the job for us. So anytime the email agent will be invoked, it's going to look at the memory.md, take in all of our notes so that it doesn't make the same mistakes twice. So those are the three files that make up an AI agent, Claude.md, skill.md, and memory.md. And finally, step five, I say, explain what you made in plain words, just so we can understand what's going on, especially if you're non-technical, you need to understand this stuff. When you're ready, you can send it off and it's going to start doing the interview. And after the interview, it will create. All of the files for you. And so I'm going to hit enter and send it off. And so look at this. It says, which email account should the agent work in? And roughly what lands in that inbox? It's question one of 10. It's going to interview me until it understands exactly how I do the job. And any extra information you want to give, do not be shy. Just give it the information until it has everything it needs to do the job like a true employee would, or like you would do it. All right, I've ran through the entire interview. I've given Claude all the files that it's asked for in order. To work like me out of my inbox and this is what the prompt ends up giving you. These three files that I just described. Claude.md, Skill.md, and Memory.md. And so these are what this prompt produces and it works very good. Let me take a look at my email agent folder to show you what has been created. I added a lot of files in here that it asked me for. I gave it a lot of information and it created files for me. Number one, I gave it all of my module transcripts. So if anybody has a question about where something is located in a course that has paid me, then what I can do is have my agent go in and find the exact spot a lot faster than I could. So this is going to help the customer experience a lot more. And it's just good when your AI agent, in my case, it's good that it knows my entire product and everything I teach. So I've given it all my module transcripts that people get access to when they purchase my course. I gave it reference files, like my voice guide. Product facts and course outline. These are all just Markdown files. Markdown's super easy for agents to read, easy for humans to read. And these were created from the interview with Claude that it had with me using the Agent Builder prompt. I even have examples of me going in and helping people with certain problems from my reel. Inbox so it knows how I talk and it knows how I handle things. You might be thinking. Okay, we've got the Claude MD We've got the memory MD. Where is the skill MD located? Well, that is in a secret folder on Mac in order to find your secret folders. You want to hit command shift period And once you do that, a hidden .claude folder will come up. This is the way that Claude likes working with its skills. It understands because it has this baked into its backend instructions. So that's what this agent created. Whenever it creates a skill, it will add that .claude folder. I'm going to click into there and then I'm going to click into skills. And then as you can see, it will organize your skills in this way. So I have my email agent. If I wanted to create other skills or other routines in this folder, Then it would name it differently. So this is my email agent skill. If I click into here, we now have the skill MD. And so we have all three of our things that we need. A Claude MD, memory MD, and a skill MD. If I click into the skill MD, I will show you what this thing looks like live. Gives it a name and a description. It says email agent. Labels Drake's unread emails and writes draft replies in his voice for him to review. Use when Drake types slash email agent or says do my email. Sort my inbox, label my emails, draft my replies, inbox runner, anything about triaging or replying to his inbox. So the agent builder built this for us. And then all a skill is, is just step-by-step instructions for the email agent to go through every single time. So step one, read first every time in this order. And then it tells it to read the ClaudeMD. It tells it to read memory, reference voice guide, product facts, course outline. Basically, hey, get up to date on Drake and who he is and what he sells. Step number two, every single time I run this, by the way, it will go through these steps. This is what a skill is. Get the emails. Shows how to get the emails in four steps. These are its instructions. It's basically like a huge template prompt that the agent understands how to use. Now your agent is built, but you haven't seen it do the job yet. And you shouldn't trust it with things like your inbox, your calendar, your CRM, your reports. You shouldn't trust it with these things until you've seen it work through a couple of times and you've run it manually and worked with it in order to get the results that you want, the exact results that you want. And that's level three. Level three is to run and test your agent. The biggest mistake people make is putting a brand new agent on autopilot on day one, and then they find out a week later it's been doing the wrong work and the wrong job the entire time, or it's doing the job wrong the entire time. Think about teaching someone how to drive. You don't just hand a new driver the keys and leave. You would sit in the passenger seat with them for a while. You watch, you tell them what to fix, and once they stop needing you to say anything, that's when you might want to let them drive alone. That's what running your agent locally is. Locally just means it runs right here on your computer, only when you ask it to, so you can watch it every time it does a job. To run it, you type slash and then the name of the skill that you just created. So for me, that would be slash email agent, and then it will go through its process, find my emails and draft replies for me. I'm in a new session. And what I can do is I can make sure that I'm in my email agent folder so it has all of the data. And then what I can do is type in slash email dash agent. And the agent builder prompt created a slash command for us. This is usually what happens when you build a skill. If it didn't, in the exact same chat thread that you created your agent in, just tell it to make it so you can use a slash command in the prompt bar. As you can see, if I want an agent to go in my inbox, Draft replies to all of my. unread emails in my voice. All I have to do now is type in slash email agent. And hit send. The goal is to just use this manually for a while in step three, even a week or two. I mean, this is very minimal work. Before you start trying to run this thing 24/7 in the background without your interjection, I recommend just doing it the way I'm doing it right now, where you come in here, you type in the slash command whenever you want to go do the work normally, and then it goes and it does the exact same thing every time. You adjust the way that it works. You still have a hand in it. And you're developing a system that's going to save you more time in the future. So always keep that in mind when you're running this thing and correcting it. All right. And right now it's adding labels to the threads. So if I refresh this screen, you can see my agent just made a bunch of drafts for all these and also labeled them. So we have draft ready. We have customer support, noise slash sponsor pitch. We have a win here, hot leads, refund request. It did all of that in a matter of about 60 seconds. And so we have eight threads processed, seven drafts written, two of them need me, one labeled only. My agent just did all of that work for me. Now, all I have to do is go in here, and this is what I recommend you do for every workflow, even if it's simple. Just go in there and check how the drafts sound. If I click this podcast invite, it says, hey, Sam, I'd be down for this. How I run AI Foundations on Cloud is one of my favorite things to talk about. Then it gives me a little confirmation here. Drake, confirm if you want to do it, then add a few weekday times that work. I didn't offer any dates. What kind of solo businesses make up most of your audience? That's going to follow up the question. And then responding. Now, maybe I come in here and I'm like, you know what? I don't really feel like doing any podcasts for the rest of 2026. Maybe I could come in here and I could say, I don't want to do any podcasts for the rest of 2026. So change that draft and add that to my memory so you remember. All right. And this is the type of thing and the type of way that you want to go about communicating with Claude as you're going through and changing things and just making the entire setup your own. Making it useful requires this type of interaction with the model. This is why we have all three things in place, like our memory file, our Claude file, and our skill, because it reads them all before it works. So now it changed Sam's draft to a decline. So if I go back to my email... So now this person asking me for the email to do the podcast is now saying, hey, Sam, I'm going to pass on this one. I'm not doing any podcasts for the rest of the year so I can stay focused on filming and building. Appreciate you thinking of me, though, and good luck with the show. A little smiley face and then Drake at the end. I personally love this. There isn't anything else I would have changed about it. And I like how I added the smiley face because that's exactly how I talk in my emails. So it's natural. It doesn't look like AI. Now, if you catch something bigger, like a step it should always take every single time it's running the process that you don't find it taking, tell it to update your skill file as well so the step-by-step gets better along with it. Now, once you begin to trust it and you go in there and you run it manually every single time and it's giving you amazing results, then there's only one problem left. The problem is it only works when you're at your computer and when you ask it to work. And so that's level four. Level four is to put your AI agent in the cloud. If you have to remember to run your agent, it's one more thing to do on your to-do list. And the whole point of this was to take things off of it. On top of that, a local agent can't do anything while your laptop is closed. Think about a coffee maker. You can walk over, press the button every single morning, or you can set the timer the night before and wake up to coffee that's already brewed, sitting there waiting for you. Now think about the coffee maker change. You just stop being the one who has to press the button. And to do that with an agent. It needs to live somewhere that is always on, and that is on the cloud. Now, the first step in getting your AI agent on the cloud is setting up a GitHub account. A GitHub is basically a cloud repository of your local folder. So that agent that you just created locally in a folder, it's basically just a place where you can store your local folders and files. Now, engineers use this for code bases so that they can share code across developers, But we can use this as agent developers just so we have. A backup repository is what they call it of our agent and instance of it in the cloud. And once it's there, we can also run it while our computer's off. What you can do is you can be in your email agent and you can just ask it in Claude code. This is the beautiful thing about Claude code is you can install things on your computer through the terminal because Claude code has access to your terminal. And this thing is intimidating, but it's super powerful. So you don't even have to look at it. Claude can just use this in the background. And you can ask it to install GitHub for you once you are signed in. So I'm going to ask, is GitHub installed? If not, could you install that for me? And I already have GitHub installed, but this is a question that you would want to ask. GH, the GitHub CLI, is installed, and I am already signed in. It has permission to work with my repos. Repo is like a repository, which is a folder, and workflows. It says this email agent folder isn't a Git repository yet. If you want to put it under version control or push it to GitHub, I can set that up. I want to say, yes, put this in a private GitHub repo, and then I will send it off. And what that's going to do is take all of our local folder elements, all of this thing that we just created, and it's going to store it in the cloud. And then the cloud can interact locally as well so that they stay in sync. So anytime you want to make a change locally and you do make a change, you can say, Okay, now push that to GitHub. As you can see, it just created files for us, .git and .gitignore. It says the email agent folder is now... is now in a private GitHub repo right here. So I can go here and I can check it out. As you can see, we have email agent and it's our exact same folder structure just in a GitHub account in the cloud. Now, once you have all of this set up, your next step is to create a scheduled routine. And a scheduled routine can be created by just typing in a simple prompt, Claude to create one for you. So I can say, Hey there, Claude. What I want you to do is I want you to create a scheduled routine for me that runs every single hour for my email agent skill. That way it's running whether my computer is on or not. I want you to build this routine for me, set up the schedule, and make sure that it runs out of this repository with everything handled. That way, whether my computer's asleep, whether I'm traveling, if it's shut, my emails are still getting the drafts sent and handled. All right. So very simple prompt. I'm just telling it, Hey, the keyword is scheduled routine. And then you can tell it when you want it running. It runs every hour, every weekday, every other week, every month. You can tell it when to schedule a task for, and it will do it. In this case, I want it on every single hour. I'm going to send that off. All right. And it's set up. So it will now run in GitHub. That was how we needed to set that up because GitHub is a cloud environment. And our local system is tied to a computer or tied to some sort of device. And so what I can do is I can go, I can open up my sidebar. I can hit the schedule button and we now have email agent running hourly. You are Drake Sriracha's email agent running as an unattended hourly cloud routine. Nobody is watching this session. So never stop to ask a question. And then it gives all of the instructions. It tells it to run our skill. You can change the schedule if you want. You can add things to your cloud environment. You can, if you have the Cloud app on your phone, set up push notifications. So every time it runs, it gives you a notification to your phone. So you can see if you have any new drafts that you need to check and things like that. But this is the final level of making this thing truly agentic is putting it on a schedule. Letting it run, and hey, eventually, maybe it's not doing drafts, but maybe you get so confident that it auto-sends. I don't know when I'll get there or if I'll ever get there, but it is possible. All right, and with that, you've built your first AI agent from start to finish. So to recap, set up your workspace, connect your tools, build your AI agent using those three files I gave you, test it until you completely trust it, and finally, then put Put it on a schedule. And with all of those steps combined, you've built a real AI agent without writing a single line of code. But once your first agent is running, you'll notice something pretty quickly. You're still doing a lot of the small jobs yourself, and you're not sure which one to hand over next. I made a whole video on finding jobs that are quietly eating your week. I'll leave that video right here, and I show you how I hand them the Claude one at a time. Watch that one next.