How to ACTUALLY Automate Business Processes with Claude Code (Beginner Guide)

In the past couple weeks, you've probably been seeing a thousand different videos on Claude Code, and you might be wondering what actually is it, how do I set it up, and how do I actually use it to automate things. So in this video, I want to run through the most simple way to set it all up, start actually using it, and start actually using it to automate things in your life. So first of all, we're going to run through what actually is Claude Code and how does it work, and then we're going to move on to how to actually set it up and begin to use it, and then we're going to build a real automation with it. So if you stick around till the end of the video, you'll have absolutely everything you need in order to be able to go from complete zero not even knowing what Claude Code is, to be able to actually start using it and automate things. So without further ado, let's just jump straight into it. So first of all, I want to talk about what actually is Claude Code. So I'm assuming majority of people that are watching this video have probably come from my NAM videos or have watched NAM videos because you're already in sort of the AI automation space. So to really simply break down what actually is Claude Code, if we look at NAM first, where in there if you want to automate something, you manually place down all of the nodes that you would use in that workflow. So you kind of plan out the workflow, map down all of the nodes. It might be a Gmail trigger, then a HubSpot step, and then an AI step, etc. And you write all the prompts, you do everything inside of it, and then you run it through, you test it, you adjust things, make sure you remove all the bugs, and then the workflow produces whichever output you're looking to produce. But with Claude Code, essentially what this is is an AI agent inside of an IDE, which stands for an integrated development environment. Essentially what this is It's just all of your development stuff inside of one environment. So don't worry, you don't have to know how to code or really know how any of that stuff kind of works to be able to use Claude Code. But essentially it's just everything you need to do what you would have done in NAN inside of one environment. So Claude Code lives inside of that environment and essentially all you do is you just tell it what you want, so what you want to automate, and it's able to create all of the tools, all of the workflows, and then automatically run the workflows through test them, debug them, and then give you the output that you want. So essentially you can do everything inside of NAN, but all just with natural language. So you can just tell it exactly what you want, you can set everything up, automatically test it, and give you the outputs that you want. So rather than manually building the workflow and choosing each node, with Claude Code, you just describe the workflow and kind of give it guidance on exactly what you want, and it just does everything for you. So now you know what it is, let's jump into how you actually set this up and use it. Okay, so the first thing I want to run through here is the pricing for this. So you can see here, this is just on the Claude website. We have 3 different pricing options. There's Free, Pro, and Max. And the differences between these is just with Free, you don't have access to Claude Code. So if you have a free plan with Claude, you're not going to be able to use Claude Code. You can see it's only available in the Pro and the Max. And then the difference between the Pro and the Max, of course you can see there's quite a big, uh, price difference between these. The Pro, you have access to Claude Code, and this is mainly just going to be for sort of testing stuff out, building individual projects, if you're trying to look to automate like massive workflows in your business or automate things for other businesses, this probably isn't going to be the best option for you just because of the amount of usage that you get with it. But if you just want to follow this video along and just test out some stuff and try building some little things with Claude Code, this Pro plan will absolutely work for you. And then the second one here is the Max plan. This one jumps to $100 a month, and you can see here you get 5 times the amount of usage, and you can also bump that up to 20 times and of course the price just increases with that. So you kind of have more flexibility of how much you want to actually use it. This is going to be for when you start to automate big processes in your own business, or if you're an agency owner like me, you're looking to automate processes in multiple other businesses, uh, the Max plan is going to be better for that. So essentially it's just the usage. With Max you get 5 times more usage than you do with Pro. So going into this, just make sure you already have a Pro or a Max plan and then you're all set there. And now we're going to talk about the actual setup for this. So there's a few different places that you can set Claude Code up in because Claude Code is kind its own thing. It doesn't live in a special environment or anything like that. It's its own thing that you install into your own environment that you choose. So there's a few different ways that we can do this, but in this video we're going to be doing it using something called Visual Studio Code. So Visual Studio Code is just an IDE, which is that thing I was talking about earlier, which is basically just an entire development environment in one app. So everything you need to build apps, automations, workflows, all inside of one environment. And to download this, all you have to do is just go over to their website here, It's free to download, so don't have to worry about paying anything. And then I'm on Mac, so I'll download it for Mac. Of course, I've already done this, but you literally just hit the button, download it how you would download any other app. And then once you're in there, it will look something like this. And to be honest, when you first load into these IDEs, if you've never used one before, this stuff might look a little bit confusing. But trust me, it's really not too much to worry about. It's just when you first load in, it can be a little bit overwhelming because there's a lot of different stuff and it looks kind of technical. But the first thing we need to do here is actually install Claude Code. And the way that we do that is come over to the sidebar and click this button here, which is Extensions. You can see I already have mine installed, but you would just simply type up Claude Code like this. That says Claude, but there you go, Claude Code. And then it's just this one here. Make sure it's by Anthropic. Click on this. As I said, I've already installed it, but you just hit the Install button, wait for that to install, and there'll be a pop-up that just allows you to sign in with your Anthropic account. Um, and then really, really easy, just click on that, it'll redirect you back to the web, and you can just sign in with your Anthropic account, either on your Pro plan or your Max plan, and then that's it. That's all you have to do to set it up. And then once that's set up, you'll see a little button in the corner here that says Claude. So you can click on that and it'll open up. If we shut this down, this is how our interface looks. So it looks pretty similar to like ChatGPT or any kind of other LLM interface, so it should look fairly familiar. But We're going to get into how to actually use this and all the different features and stuff in just one second. The first thing that we need to do is actually start a project in here, and the way that we do that is we go to this file icon up here, we go to Open Folder, and then this is just going to take you to your admin folder on your laptop or your computer, whatever you're using. And for me, I've just made a folder called Claude Code, and then here I've already got a couple files. For this one, I'm just going to create a new file. So if we go Show in Finder, and I just went ahead and made a new file, just a standard way you make a new file, just right-click, New Folder, and then name it. I've just called YouTube demo, and then we're just going to click on that and click open. And essentially what this does— by the way, when you open this, you'll get a pop-up that says this, and you're just going to say yes, I trust the authors. And then again, we're back in the view that we was before. This time we're in a project, but right now you can see stuff looks a little bit different. We have the welcome page here from Visual Studio Code, and we also have the Visual Studio Code built-in agent, which is kind of like their version of Claude Code, um, and it's not amazing, so there's no point using it. You might as well just use Claude Code. So what we're going to do open up Claude Code here, and then we can just shut down these other windows. And then now this should look fairly familiar to you because like I said, it's kind of just the same as like ChatGPT or any other LLM like Gemini, Claude, whatever it is you use. Um, this is kind of how they look. So just to walk you through what I did with that folder bit, basically that folder is just going to contain all of the different files to create your actual workflow automation or whatever it is you're building inside of Claude Code. So when Claude Code gets running, it's going to create a bunch of different files that all have different functions in there. and it's all going to be stored inside of that one folder. So to give you an example of how that looks, you can see I've got my other Claude code projects here. I have YouTube demo right now, it's empty, but these other ones are full up with files, and these are all files that Claude have created that basically do the different functions and stuff inside of the workflows that we're going to be building. So these essentially are just like projects. So if you're used to using NAN, imagine these as like each individual NAN instance. So if you use like hosting or something like that, Imagine these as just different hosting or accounts or NAN instances. So basically everything from your project, all your API keys, all of the workflows, all of the functions, all of the AI agents, all going to be stored inside of these folders. They're basically just like mini projects. Okay, cool. So now what we're going to do is we're going to do something called adding the Claude MD file. So the way we do that is we just add a new file here and we're just going to call this claude.md. Md. And if you're wondering what md means, that just stands for Markdown, which is basically just a computer-type language. It's just a way to format text. So that's what the md stands for. And what this file is, is essentially our system prompt. So if you're used to already building AI agents and things like that, you should be familiar with system prompts. So this is just going to be essentially a system prompt for our Claude code for what we're trying to build. So I already have one pre-written here because the great thing about this is that you don't really have to rewrite it every time. You can kind of use a standard one. And you'll see how it works. It kind of, um, the variants come from what you do when you're actually chatting with Claude rather than this system prompt or this MD file. So I'm just going to make this a little bit bigger so you can see. This is basically how it looks, and like I said, this is like a system prompt. So we're just saying to Claude exactly how we want it to operate. And what I want to do is I want to run you through basically the framework that we're using here. So the framework that we're using is essentially sort of an agent workflow and tool framework. And the way that works is Claude Code in this is going to be acting as our agent that essentially orchestrates everything. And then what it's going to do is it's going to have workflows. So these are going to be files that it'll create with set workflows. And then it's also going to have tool folders. So these are going to be specific tool folders. And what I mean when I say tool folders or tool files, these are going to be Python scripts that actually execute things. So for example, scraping websites, sending emails, anything like that, actual actions. So kind of like the nodes you'd put down in NAN. They're going to be in tool files. And then the workflow files are literally just going to be kind of like this, markdown files of written workflows of how to use those tools. And then Claude is going to act as our agent. So essentially, to give you an example, what would happen if we wanted to look at local businesses in our area and then scrape our websites, we would have workflow files that would show exactly how to scrape the website, uh, what information we're trying to get, all of that kind of stuff. So essentially, what is actually happening in these workflows, just in plain English. And then we would have tool files that would be one to scrape the website, one to use the Google Maps API to look for local businesses, one to generate personalized compliments. Let's say we're doing it for cold outreach, that kind of stuff. And then Claude would be acting as our agent that would essentially orchestrate all of that. So that means when we would say to Claude, find plumbing companies in London, for example, it would look at the workflows folder, see how to exactly do that, and then it would look at the tools folder, and then it would look at all of the tools that it has be able to actually execute that task. So the Claude MD file looks something like this. We've just given it sort of a brief introductory— you are Claude Code acting as the primary agent and orchestrator for this environment. Your role is to execute business workflows and act as an agentive layer using structured workflows— what to do— Python tools— how to do it— and then files as durable output. So basically it's just going to output its stuff into files. So the core operating rules: always follow a workflow, so never improvise large tasks. Select or create a workflow before for acting. So every time it has to actually do a task, it's going to look at all of the workflow files that it has to see if it's able to basically complete that, complete that task using the workflows that it has. If it doesn't, then it's going to plan out a new workflow file, and then it can use that later on if it gets asked to do the same task again. Prefer tools over reasoning. So scraping, parsing, validating, transforming, use Python. Reasoning is for planning, not guessing. So basically, we want to make it a little bit more deterministic and say things like Scraping, parsing, and validating and transforming— use Python tools rather than just using your own brain, because we want Claude to just act as an orchestrator that looks at the right workflows, creates workflows if it needs to, and then also looks at the tools that it has and creates tools if it needs to, rather than kind of doing this stuff by itself. So like scraping and parsing, we don't want Claude to do that itself. We want it to use its workflows and tools to do that. Be deterministic— so the same input should equal the same output. Avoid creative variants unless explicitly asked. If something's missing, create it. So for example, if there's a missing workflow, add one. If there's a missing tool, write it. Missing schema, define it. So the structure— so this is how it's actually going to structure this project. So there should be a workflows folder, a tools folder, an outputs folder, a data folder, and a logs folder. And essentially what all these do— workflows, we've already talked about this, is just going to be markdown files with step-by-step of how to do specific tasks. The tools, these are going to be Python tools that you can actually run to do things like scraping, sending emails, looking things up in HubSpot, that kind of stuff, like the nodes you would use in NAN. And then the outputs, whatever its task is, it would output into a specific folder as well. And then data, so raw or immediate data, basically if it's scraping through large amounts of data and it needs to store that temporarily somewhere, it would go in this data folder. And then logs, so just the tool logs, is optional but preferred, so just what it's doing at each stage it's really, really good for debugging because it itself, like Claude Code, can then look at those logs, um, if a task didn't go too well, and it can kind of see what happened and then fix it. So the first part of this is workflows. All workflows live in this workflows folder and are written in Markdown. For example, audit process, research company, scrape leads, scrape website, generate invoice, analyze documents. So these are all just Markdown files of written processes of how to do each task and what tool to use. If a task does not clearly fit in an existing workflow, create a new one. Keep it concise and reusable. Workflow contract. So the required section. So this is basically how it's actually structuring these workflow files. Every workflow must include a purpose, inputs, tools used, steps, outputs, and validation. So quite simple. So basically just mini system prompts inside of files on how to do specific tasks. And then the rules for tools. So how the tools are working. So how the work is actually done. One responsibility per tool, callable from CLI or importable, structured inputs and outputs, no hard-coded secrets. So we'll get onto the hard-coded secrets bit in a minute. So examples of how the tools would look would be Tool Scrape Web, Tool Scrape Leads, Parse PDF, Extract Fields, Generate Invoice, Validate Data. And these are all Python scripts, which means they're reusable. And essentially with NAN as well, with like the different nodes you'd use, on the back end, they're all just Python tools. So essentially, you can think of it as it creating like its own NaN nodes to execute tasks. If a workflow requires manual steps, that's a signal. Write a tool. Standard behavior loop for every task: identify the correct workflow— so when it gets an input to do something, identify the correct workflow to do that— confirm required inputs exist from the user, i.e., me in this case, use the tools to execute each step, write outputs to the outputs folder, and then validate the results, and then stop. And then kind of like the third layer of this, now we have the workflows, the tools, and we have an orchestration layer, which is just your job. So intelligent routing, read the directives, call execution tools in the right order, handle errors, ask clarifying questions if needed, update files with what you learn. You're acting as the glue between the user's intent and execution. So after everything I spoke about, this should now be pretty self-explanatory of what Claude Code is actually doing. It's using all those workflow, uh, markdown files that it has, and then it's using all of the Python tools to actually execute the user's request. So essentially, it's just acting as the glue in between the user's initial intent, so what they're trying to do, and just the execution of that actually getting done. And then we have a few extra things down here. I like to kind of, um, curate mine to be a bit more business process focused, so optimized for things like audits, research, data extraction, lead gen, document processing, ops automation, repeatability. So maybe slightly more boring stuff, but this is stuff that you would actually use in a business, um, and things that would be automated. So that's pretty much it for ClaudeMD. And like I said, this is just acting as its system prompt. So now we get into actually the Claude Code interface here, and you can see the first thing, there's 3 different modes down here. So right now it's ask before edits. This just means before it tries to edit anything, it has to ask me and I can just confirm or deny. And then we have an edit automatically where it just basically doesn't— it can just edit automatically. And then we have a plan mode as well. So always the first thing we do is we're going to say to it, basically, look at the Claude MD file and set up the environment. So it has to just set up at the beginning just blank folders for like the workflows, the tools, and that entire structure that I just showed you. So we're just going to say to it, look at the Claude MD file, set up the environment first, and then we'll have a look at what it's done. Look at the Claude MD file and set up the environment. So this is going to be set to edit automatically because this should be pretty simple for it. It doesn't really need to ask us for permissions or anything. So we'll send that off, look at the Claude.md file, and set up the environment. So what's pretty cool here is you can see it actually like thinking through everything, and it basically just tells you what it's doing each step of the way. So let me start by reading the Claude.md file to understand the project setup instructions, and then the actual action it took. So it read it. Now let me check what already exists, and then it checked in there, and you can see not too much. And then now it's kind of doing a checklist here of what it's going to do. So create the directory structure with the workflows, tools. You can see here these bash commands as well. You can just allow this because it doesn't edit absolutely everything automatically. The bash commands are something you usually have to, um, actually give it permission to do. But for all of these, you can just hit yes, and it kind of knows what it's doing here. So like I said, yeah, it's given us a checklist here. So of the create directory structure, the workflows, the tools, the outputs, the data, the logs, it's going to initialize a Git repository and it's going to set up a Python environment and.env file, then create the initial workflow templates. And you can see as it's working through, it's updating the to-dos. So it's done that first one, now it's doing the second one, and then done the second one, now it's doing the third one, and then done that one, now it's doing this one, and then done that one, now it's doing the last one. And then it should, yeah, just kind of like verify everything, make sure it actually did its job, and then yeah, it'll give us an output that looks like this. So environment is set up, Here is what I've done. So the directories are workflows, tools, outputs, data, and logs. And you can see them all up here. So data, logs, outputs, tools, and then workflows. And right now, um, kind of just empty apart from like the workflows where it's got a template in there, as you can see. So it's used the template that we gave it of like how to structure everything and just given itself a template. So when it's creating these new workflows, it can actually look inside there to create these files. And then it created a new Git repository, and then it did a.gitignore, and then excludes this stuff here, as you can see. And then did a.env file. And what this.env file is, by the way, is because we don't want all of our sensitive data like API keys and permissions injected into all of this, basically we just have a separate file which is just going to store our API key. So when we actually ask this to build workflows, let's say it needs a, um, you know, like a Google places API key. It would put it in the.env, so it would just put a space in there for us to paste in the API key, and then it's kind of separate to everything else. So that way we can easily manage it, store it, and it's not getting exposed in any kind of way when we take this public. And then it created the workflow template, and then it also set up the Python environment. So basically just the environment we're in right now can actually use Python. Cool. So now that's what we do. The entire environment is all set up, everything's ready to go. And now we're going to ask it to actually create a workflow. So we're going to set this to plan mode, and I would always recommend doing this, set it to plan mode before you actually get it to do anything. Um, this way it kind of just plans everything out, sees what it can do, what it can't do, and then gives you some options as well, uh, with like the different ways that you might want to do it and make sure everything's all clarified, you're all on the same page, and then you can like tell it to actually just go do that. So this is what we're actually going to say to it. I want to build a system that can scrape businesses in my local area find their emails and phone numbers, and craft a personalized outreach message based on my offer. For reference, I run an AI agency and I offer custom automations to automate primarily admin tasks to save time and money. I want the results to be outputted in a Google Sheet that I can easily import into my cold outreach software. Plan this system and let me know any API keys or permissions that you need. Perfect. And then we're just going to send that off. It's going to run through here, uh, basically do some planning, and it'll probably come back to us with some options and things that it needs clarifying. And then once we just give the go-ahead on that, then it will be good to kind of just leave it, let it do its thing. It will be able to create all these workflows, test them all out, test out all the Python tools, and then just basically give us a finished output. So we've got a pop-up here saying allow searching for this query: Google Places API pricing capabilities 2026. Yeah, so it's just looking to search up some stuff here. So, uh, the Google Places API pricing, the Hunter.io pricing, and then the best local businesses directory. Cool. We can say it can search all of this stuff, uh, so we're just going to hit yes for all of these. Perfect. Uh, so obviously just had to do a little bit of research there on pricing and the best methods to do all of this kind of stuff. So pretty much with this, if you don't really understand like the terminal commands it's doing, all these bash commands and everything like that, you can pretty much just say yes to them. Don't feel like you need to be an expert and know exactly what it's talking about. Obviously that's why we use Claude Code. Okay, here you can see it's asking if I have API keys for Google Places, Hunter.io, and Anthropic API keys. So, um, yeah, so we can use all of them, that's no worries. And it's asking which cold outreach tool, uh, do I want. I'm just going to say Instantly because that's what I usually use. And then what location and business type do you want to target first? And we're going to say I'll provide later. So we want to be able to actually say to Claude Code, like, uh, search for X business in X area and then it will just be able to do that for us. So it's a little bit more dynamic rather than kind of like hard coding in the exact location and the exact industry that we want to search for. We just want to be able to say to Claude Code, uh, find these businesses in this area. Okay, cool. And then it's just given us this full plan here. So you can see it thought for about, uh, what was planning for maybe 5 to 10 minutes, something like that. So quite an in-depth plan. Um, and this is what it said to us. So you run an AI agency offering custom automations to eliminate admin tasks for local business You need a repeatable pipeline that discovers businesses in a target area, finds their contact details, and generates personalized cold outreach and exports everything as a CSV for direct import into Instantly. So it's giving me here, this is going to be the architecture of it. Very cool. And then the files to create, it's going to create the workflows, uh, Scrape Leads, the tools, Tool Discover Businesses, Tool Enrich Contacts, Tool Generate Outreach, and then Tool Export Leads. So you can see here, this is pretty much how it works with the tools and the workflows. This is going to be the workflow and use one workflow file because it's pretty much only doing one workflow, and it's going to be the master workflow defining the full pipeline. And then these are the tools that we would need to actually deliver on this workflow here. So the Google Places API business discovery, email phone enrichment, Claude API personalized message generation, and then the JSON to CSV export, and then the.env, which is going to, uh, have all of the API keys that we need in there. And then here it gives us all of the tool details. So I'm not going to bore you and read through all of this, but it gives us all of the tool details and what they're all going to do. Uh, and then this is going to be the CSV output format. Cool. And the API— and then the API keys required, uh, the Google Places API, Hunter API, and then Anthropic API. Tells us how exactly to get it, and then also the pricing as well for this. So it was cool as well with the Hunter, it's told us that you get 25 3 searches a month, but scraping fallback if exhausted. So basically, if we don't want to spend any money and we just want to use those 25 searches, but it goes over 25, it's actually going to build a fallback to just manually scrape it and get their phone number and email. So very, very cool. And the estimated cost per run, so for 20 leads, it's going to cost us less than a dollar, and it does the breakdown here. Very, very cool. Then the build order and verification, so all looks good to me. We're just going to say yes and auto accept, which which just means it's going to be able to actually basically implement this plan and it's just going to auto accept everything. So right now it's just given us a to-do list with everything it's going to do. It's going to update the.env with the API key placeholders, and then in there we'll put in our API keys. And then it's going to create the Python tools, as you can see, and then it's going to write the workflow in the markdown file. So it's already done the.env file. So actually now if we look in.env, you can see, yeah, in here is where it's going to put— is where we're going to put our API keys. So it's just got placeholders in, uh, Google Places API key, we put that in there. Hunter API key, put that in there. Anthropic API key, put that in there. Awesome. So what we're going to do, we're going to leave this, let this run, and we'll come back when it's done. Okay, cool. So it's just finished all of that, and it's basically told us what it's done and then our action items to actually get this working. So all 6 files are built. Here's a summary of what already. So the discovery tool, the enrichment tool, the outreach tool, then the export tool, and then the workflow here. And then our next steps are just add our 3 API keys to the.env file, and then we are literally good to go. So if we quickly look what it's done, if we open the workflows folders here, we can see, uh, workflow scrapeleads.md. If we open this up, I'm not going to bore you and read through this entire thing, but this is just the exact workflow on how to do what we're asking to do, which is scrape local businesses, enrich them, and then write a personalized outreach message. So it tells Claude all of the inputs that we need here, um, all of the tools that we want to use, uh, the steps— so a step-by-step of what to do, uh, and then how to output, and then how to validate as well. And then if we shut this down, go to tools here, you can see these are all the different tools. So discovery businesses, enrich contacts, export leads, and then generate outreach. So in here are just Python tools to do everything it needs for this workflow. So if we just open this.env file, now we need to add in our API key. So I'm going to go ahead, grab all of those and put them in. By the way, if you ever don't know how to get the API keys, it'll give you a brief, uh, thing here on how to get it. But you can always literally just say to Claude, how do I get my Hunter API key or Google Places API key, and it will just tell you. Okay, so I just went ahead and grabbed all of these API keys. And by the way, at the time of recording, all of these will be deleted, so don't worry about that. Um, but now they're all in there, so we pretty much have everything that we need, um, in order to actually get this going. So if we make sure to save this by hitting Command+S and then close this down, now what we're going to do is we're going to clear this chat. So if we just go to forward slash clear— by the way, when you hit forward slash in here are all of the different functions that you can do. So these like additional functions and quick commands, as you can see. Uh, we're not going to get too much into that in this video because you don't necessarily need it. But don't worry, later videos will come out where we get more in depth into this kind of stuff. So we're just gonna do clear and then clear conversation. It just starts a new conversation. If you wanna look at the old one, you can just come up here and look at it. So now we're gonna actually try to run this through. So we're gonna say, I want to find 20 marketing agencies in London. That's literally it. We're just gonna say, I wanna find 20 marketing agencies in London. We're gonna change this to edit automatically and we're gonna send this off. So usually for this, it might be worth, um, just for your first run through doing it on the planning option so that way it plans everything out. But just for this video, because it's pretty simple, I'm just going to do it so it just runs it straight away. So first of all, it looks at the workflows.md. So basically looks at all the workflows it has. It's going to look at this one, the Scrape Leads. As you can see, it's just read Scrape Leads, and then it's going to read all of the Python tools that it has and basically get familiar with them, and then it's going to actually execute this. This for us. So right now it's executing all of this and it's just going to get us to allow some bash commands. So the to-dos here: discover 20 marketing agencies, enrich them, generate outreach messages, and then export to Instantly AI with a CSV, and then a report summary. Cool. So we're just going to let this run through. If anything pops up, obviously I'll tell you, but if not, then we're just going to let this run through and we'll see the results. Okay, so one of the cool things it's done is it's now found that— so it's found the 20 businesses, it's enriched 20 businesses. Now it's going to write our outreach messages, and it's just got a little pop-up here saying, what offer/service should the outreach message pitch these marketing agencies? We're just going to say AI automation services. The cool thing about this is, is once we say that once, it's now going to be able to update the workflow file, and, um, in there, that now it's never going to basically ask that question again because we've said it once. It's going to update the workflow file to basically put that in there. It knows when it does it again, it will have to, you know, offer AI automation services rather than the other options that it gave us. Okay, cool. So as you can see, it is now completely done. So it gave us an output of what it did. 20 agencies discovered. It found 20 out of 20 emails. The email source, 19 came from Hunter.io, and then 1 came from the website scraping which is really, really cool because obviously Hunter couldn't find it, one of these cases. So Claude actually just went ahead, used the tool to just scrape their website and then find the email. So outreach message generated, and then it also found 19 out of the 20 phone numbers. Really, really cool. One thing I also want to point out that I thought was really awesome that it did— um, mistake on my half, I created a new Anthropic account for this video and they give you a certain amount free API usage because I hadn't connected my card to that account yet. And the API key that I uploaded, basically that Anthropic account had run out of credit. So one of the cool things it did in here is you can see that basically it was erroring, it didn't really know why, and it found out that, um, my API key had no credits remaining. So it actually just generated the messages itself. So it read the tool on how to generate the messages, and then Claude Code itself actually just went ahead generated those messages for me, and then it just carried on as normal, which is really cool. So it gave us the output here saying your Anthropic API key has no credits, so outreach messages were generated using personalized templates instead of the Claude API. So absolutely amazing. And then now to actually look at the outputs, what we do is we go to this outputs folder here. You can see we have a leads CSV. If we open this up, in here is all of the leads, but obviously right now this isn't, um, how we want it. You know, it's not— it's like in a CSV file technically, but right now it doesn't really look very formatted and we can't actually do anything with this. We can't upload to Instantly. So really, really simple on how to actually get this file. So I'm going to upload this into Google Sheets so you can see a little better, but the way you actually get this file is if we go to import something in here and then we just browse our files. When we go to Users, Admin, and then for me it's Claude Code and then in here all of our projects. This one was called YouTube Demo. All we do is we go to YouTube Demo and then we just go to Outputs, and then the CSV file is right there. So it's literally already on our computer in that file, which is really, really cool. That's why having these projects in these files is such a cool way of doing it. So we can literally just upload that file, import the data here, and then we can see what it gave us. So this is the email, this is the person's first name, the last name. You can see it's actually finding like um, not just like office@ or anything like that. It's finding like the personal emails of these decision makers, which is really cool. Then we have the company name, their phone number, their website, their location, rating, review count, business type, uh, subject. So it's going to be the subject of the email body and the actual outreach message. Hi Sheena, came across The Good Marketer digital marketing agency London researching marketing agencies in London. With 138 reviews at 4.8 out of 5, it's clear your team delivers strong results. One pattern we see with agencies at your stage: the manual work behind client onboarding, reporting, and lead qualification starts to bottleneck growth. We build AI automation specifically for marketing agencies— things like intelligent chatbots that handle client onboarding, automated lead scoring, and content generation pipelines. The goal is to free your team from repetitive ops so they can focus on strategy. Awesome. And then it gave us the email confidence score as well. So basically just how confident it is that that email address is the actual email address of that person. You can see this one's 50 here. So this is likely the one where it relied on the web scraping and it just found their contact email, which obviously isn't the best, but Hunter couldn't find anything at all. So as a backup, it just used web scraping here and it still found us an email. So it's still better than nothing. So also want to show you ways to host this and have it running so you don't have to tell Claude every time. But of course with this one, um, there's a little bit of nuance because you want to tell it, uh, you know, different requirements like the amount, the location, the industry, everything like that. So we're going to save it for another video on how you can actually host this on its own server and have it running sort of automatically. But I hope you guys enjoyed this video, and I hope this kind of— if you was a little bit hesitant or don't really know how to get started of Claude Code, as I was myself when I first started using it and first sort of heard everyone talking about it. I hope now after watching this video you are able to actually get set up with it on your own and start to automate some things. That's it for this video, guys. I'll see you in the next one.