[0:00]You probably suck at using Claude Code, or maybe you're not using it at all, which is even worse. But in this video, I'm going to break down exactly how I use it as an AI founder, I run multiple different AI companies at this point. How I use it myself to automate certain parts of my roles, but that can also be taken to whether you're a student, an employee looking to just automate your whole degree or automating parts of your work so that you do a whole lot less. Or you're a business owner as well wanting to be able to use it for yourself or teach it to your team. This is the exact training that I put all of my staff through so that they can use it and set these up to automate parts of their roles as well. This is probably the best general purpose AI tool for productivity right now, when set up correctly and when used correctly, it can automate up to 90% of the stuff that you do. So I'm going to be breaking down in this video the exact template that I use and give to my staff, them to start and create their own workspace. I'll be breaking down the four biggest mistakes that I see beginners make that stop them from getting the most out of it. And then I'm going to be showing you step by step how you can get the set up on your machine as a beginner, the right way, using all the sort of shortcuts and commands to make this really fast to use. And then show you a real and live example of me using the template to automate certain parts of my own workflows, so you see how it works in action. I haven't seen anyone else on YouTube dropping this kind of source on Claude Code, so let's get into it. So we've got all this to get through, but we're just going to start with the first thing that you have to understand is that an agent, which is what Claude Code is, acts as a digital employee for you. So what does that mean? A digital employee will need to have some kind of tools or access or context or knowledge about what it's doing, right? And the first thing you need to understand with getting the set up and how to get the most out of it is like, Claude Code is going to be your digital employee or your assistant to do whatever you set it up to do. But in order to do that correctly, it needs to have what I call a workspace. So what is a workspace? It is one folder on your computer that contains other folders and context and files and scripts required to do a task or set of tasks that you want to be performed through Claude Code. So in this case, as you can see on screen here, this is what we'll be working with later, and this is uh VS code. Don't worry this is like it's not coding, it's nothing, it's just chatting away to a chatbot in here. And if you can't wrap your head around using one of these IDEs, at this point you're kind of cooked. So this is for those of you who are extra intelligent and uh believe you can wrap your head around a basic tool like this. So this is what a workspace looks like, and this is essentially one folder. You can see it's called Claude workspace here. This is Claude Workspace template that you guys are going to get, but inside this folder, this is just any old folder on your machine. Inside it, there are a bunch of other folders, so we have the commands folder, which we'll go into in a second. We have the skills, we have the context, we have outputs, plans, reference, scripts, and the Claude.md. So when you are thinking about setting up Claude Code to be your personal assistant, to automate parts of your work, you need to create a workspace for it to operate within. And that means, okay, what context does it need to understand who I am, the stuff that I do, the priorities and strategies we're working on right now? Things like these command files, which take a workflow like looking up or researching a podcast guest, and how you break that down into simple texts, which we'll go through in a bit. But basically, a workspace is a folder with a bunch of folders and files in it that you can configure in a certain way, and this is by far the best set up and kind of template that I've found for training my teams on this. Um it's just a great default starting point, it has all the key points that you need to really get the most out of it, and you can use Claude Code using these like create plan thing, which I'll show you in a bit. To very rapidly turn it into something that you can use for yourself and modify it. So for a workspace, I find it's most helpful to split them by a a kind of a key function. At least for me as a founder, I've got multiple different businesses. So for me, I have multiple workspaces, one for each function or area of my my work that I do. So, um I think of it as like a digital teammate or employee for each of these areas, for my personal brand and for the content I do. I have a workspace for that, I have a workspace for my agency, uh helping us with proposals and scoping. I have one for my uh education product, so it has full context of all of the course material that I've created so that I can create new stuff easily. And one for my SaaS as well, which is an engineering kind of prompt engineering workbench, which I'll actually show you guys a little bit later. But in my case, I've split it into different areas. I could try ram it all into one, but I like to keep them a little bit clearer and more separate so that I can really tune the uh, tune the content and context a little bit better. But uh, if you're just an employee or a a student and you're looking to automate parts of your work, you're probably only need one. And it's going to basically take all the job functions that you do. Say you're a marketing person and you need to like do the weekly marketing report, you need to uh come up with a new campaign idea or like do some brainstorming or yeah, analyze the last campaign and figure out what you can do next. All of these are a tasks that you do, that can be turned into a a command or a list of instructions, and put into a command file here, and then you're able to call those very quickly and say like, slash marketing report and it's going to do all of those things. So it's kind of like any other kind of automation, but in this case, you're doing it through just text files. From Claude Code having enough context about you, it can kind of just read these text files, have access to all of these bits of context and documents and folders, and be able to do things for you like maybe pull information in from the web, which we're going to talk about a bit, that falls into the scripts here.
[7:01]So as a bit of a walk through of uh this, we'll just break it down a little bit further here. We have firstly the commands folder, and you can think of commands as reusable prompts to automate work functions or do things within the workspace. These use natural language to guide Claude Code through a workflow using the content within the workspace. And like I said, these use natural language, just regular plain English or whatever language you speak, to guide Claude Code through a workflow or set of steps, using the content and scripts and tools within the workspace. So these are the main way we're going to be working, as you'll see in a bit, and these are the ways of baking in your workflow into a set of instructions that Claude can do uh on your behalf. Then we have the skills folder here, and so these are kind of like commands, they're kind of merging a little bit at the moment. Uh but skills I like to think of as a as a bit more substantial in a lot of cases, and they're kind of like installable plugins for your workspace, that typically created by others and imported in. And you can go on to like skills marketplaces and try to find ones made by other people. There's a few security risks inherent in that, which we'll touch on a little bit later. But skills are like kind of plugins or packages that you can plug in from other people a lot of the time, and it's going to enhance or give more capabilities to your workspace. They're a little bit more advanced than I would like to go into within this uh, within this tutorial, but I will be covering these two skills here because I think it's good to have it in part of the template. And then we have the context folder, and this is essential context on who you are, what you do, how you're doing it, and why. And Claude Code reads this every session to come up to speed on. So every time we spin up Claude Code, it's going to be reading all of that information, so it fully understands the the big picture. This is something we'll be touching on in a bit, but it's really important that every time you spin up Claude Code, it knows exactly who you are and what you're working on, current state of the strategies, current data that that's come back from different platforms. Context is essential to give it sort of a a base for you to have a intelligent conversation on. And in this template, I've broken it down into the key different context stocks that I've found to be most effective and they all build on top of each other, which we'll go into in a second. Then we have scripts down here, and scripts are like, I know it might sound scary, but they're code snippets that are created by Claude Code. You don't have to write the code yourself. Claude Code can intelligently figure out when it needs to create some kind of code or write a script, and a script is just a a bit of code that can sort of be triggered at a certain point. It might take in some inputs, do a certain couple of things on the internet, uh process some data in a certain way, and then pass it back to Claude Code. So it might be, I write a script to fetch some information from my uh CRM or from some sort of API, like you want to get the real-time weather data. It can pull it from a weather API, so scripts are a handy way, a bit more advanced tool, but something we're going to be jumping into as well. But these are a key thing for pulling external data into your Claude Code instance. And then we have the Claude.md, which is an essential document that describes the entire workspace to Claude Code every session that's spins up. So combined with the context folder, it basically means that every time we start Claude Code, it fully understands the workspace it's operating in, reads all of these different context documents, and knows who you are, what you're doing within the business, what your strategies are right now, any current data. And so it fully understands where it is, what it's doing, and is ready to start taking action like an employee. So just like when a new person comes with the company, where you're hiring someone, you would need to give them all that context. And as you'll see when we're actually working with Claude Code in here, we're starting it up, we're starting a new version of Claude Code all the time, so that we need to give it that onboarding every time it starts up. All right, so that brings us down to the next and probably most important thing you need to know about Claude Code, which is priming and context management. As you may or may not know, the latest Claude models only have 200,000 tokens available for us to play with. So we want to essentially have Claude Code fully contextualized, understand who we are, what we do, what we're, what it's able to do within the workspace, with as minimal tokens used as possible. So as you can see here, this is actually from Claude Code itself. You can run this command slash context and it'll tell you how much context it's got available. How much you've used up already, and basically, we want to be sending and and like interacting with Claude Code with as many of those tokens available. Because there's an effect called context bloat, which means the more tokens are used up out of that 200,000, like the more diluted your instructions are going to be. So you want to have it fully dialed in, fully aware of who you are, what it's working within, and then you're just going to kind of drop your task on top of that. So it's got all this room, it's very clear on what it's working on, and you don't get any of the negative effects of context bloat. So this all ties into something that we're going to be covering here in a second, but first let's go over the the biggest mistakes that people face when it comes to priming and context management and just generally using Claude Code. And the first is using Claude Code like a chatbot. It means you're just sitting there, kind of chatting away like you would with ChatGPT. And this is not really the best way to use it, because you've ended up at these big long conversations, and here you can see, the messages take up a significant amount. In this case, when I took the screenshot, my messages were taking up 26% of the context. So the if you're chatting away and you've filled up that uh that chat history with messages, and like 150,000 tokens are filled up with all of the messages you've had back and forth with Claude Code, then you're going to have this context bloat. Your instructions and messages are going to be so diluted that the chance of it actually getting exactly what you want the first go is going to be massively reduced. So the first mistake is using it like a chatbot, we are not going to be doing that as you'll see in a bit, and the result is that you're going to get inconsistent and low quality outputs due to lack of structured prompt processes. So we want to be baking our workflows or the tasks that you do into reusable prompts. You don't want to be like, uh hey, I want to do this today, can you do this for me? Because one, you're wasting a whole lot of time writing that out, and two, it's going to be different each time, right? You want to make standardized uh command files, which I'll show you in a second, it's just a simple markdown file, walking through the steps that you want Claude Code to take. So that you have a structured and reusable prompt, it's going to save you tons of times and ensure that it's consistent. And so yeah, the solution here is to create reusable commands. Mistake number two is not contextualizing Claude Code at the start of every new session. Like I was talking about before, if you have an employee, you need to give it context on who you are, what it's trying to do, what business environment it's operating in, what the goals are, what the values are, the business. All of this context that you have to give to someone in order for them to actually function. You don't give a new employee that context, how could you expect them to do well? It's the exact same thing with using an agent like this. So the result of not giving context is going to be poor quality output. You know when you like talking with Claude or ChatGPT and it's just like not quite got your tone of voice right, or it's just missing a bit of the picture. It's like, no, well, I'm not doing that right now with the strategies, because we've had change, or like this is a new quarter, and you always have to go like, what's called in the loop, and you're chatting back and forth, just giving it nudges to get the final outcome you want. When you fully contextualized it, and you've used one of these reusable commands, it can fucking nail it, like get it right on a button the first time. And that's when you think about how much time you're wasting in chatting back and forth, the 70% completed one might take you half an hour to finish, whereas the 95% one might take you two minutes. So if you're going to do this stuff, you need to do it right, because there's a massive difference in the amount of time spent. And so the solution to mistake two is to always be running a slash Prime command, which we're going to go into in a bit. But basically, whenever you start a new Claude Code session, we're going to run slash Prime. It's going to look at the context and the template, like we have up here. It's going to basically read the Claude.md, it's going to read the context here. And then basically, we have an intelligent base to have a conversation on, which we'll be getting into in a second. Mistake number three that I see noobs making is not utilizing planning and implementing loops. This means using a Claude command to say, hey, I want to do this within the workspace. Can you write a plan for how we could create this? Like, hey, if I have a a new function, say I'm a marketing person and I'm using this, and I've want to add a new command or a new task that I want Claude to be able to do for me. Some people will try to do this all manually or just chat in the in the chat with Claude Code to try to get it to do it. But that often means that it's not going to fully consider everything in the workspace, and when it sets it up, it's not going to be fully aligned with everything else. So you start to get this janky workflow. When you have it set up the way that I'm going to show you in a second, it can fully analyze how everything else is set up, and when you say, hey, I want to add this in, it will go, hmm, okay, well, I could go search the web and figure out what different tools or what bits of code I could write, and then it can actually implement and create those scripts for you without you having to do it. So the planning is really important so that you can kick over to Claude Code, the brain work of having to figure out, hey, what scripts should I plug into this particular workflow. All right, so those are the biggest mistakes, and like I said, all of these are kind of already factored into the template. So we'll see those in action in a bit. But I want to give you this diagram here, which I think is really important, I have to teach those all the fucking guides to my team. So this is the most important diagram, not only Claude Code, but basically any kind of AI usage these days. And it's all around context stacking and how you can get things brought up to speed so that it can have an intelligent conversation. That's what these guys are doing up there. They're having a an intelligent conversation, but they're doing that standing on top of all of this context. So I think of it in layers, like we need to start with the Claude.md file, which is just a text file, and it contains the purpose of this workspace and kind of gives it an orientation of like who it is, who you are, what the hell is going on within this workspace. Then on top of that, I layer information about the business, whether it's your own business or whether you're working for one, um information about you, how you relate to the business, and where you fit in, what your roles and responsibilities are, so the what, the who, and the why. Then we have strategy, so what is your latest like this quarter strategy? Without knowing that, it's going to not have the full context, or maybe be misaligned with the objectives that you have right now. And then I like to add on top of that, when possible, the current data that relates to that strategy, whether it's your website views or YouTube data, or any kind of data that relates to what you're trying to do within that workspace. And the reason we have this slash Prime command, which is going to be run every time we start Claude Code up, is so that it understands everything here, and we can start to have an intelligent conversation on top. It's like, I'm going to bring you up to speed, here's all the documents, but it's important that we have these set up correctly, which is what we're going to be doing in a second. And so this is incredibly important to understand, because this is the key to getting the best results by having it fully contextualized about you and your business and the stuff that you do. Now, the thing is, we don't just have a conversation on top of this, we actually start to build workflows on top of it. So that is where we get to what I call the workflow layer. Maybe let's bring one of these guys down here.
[21:40]So what we have here is our Claude Code here, standing on top of all of that information, right? Every time we start a new session, we're going to be bringing Claude Code up to speed, and then we're giving our buddy, or I mean, I'm not going to assume the brother's gender. But we're going to be giving him, or her, or it some tools to use, and these are in the form of markdown files, which is just text files. Example, we're going to be doing one in a second called analyzing competitors. So if I want to like for my own podcast, sometimes I want to pull a competitor's podcast, and I want to dig into it, figure out, hey, what are they doing, who have they recently uh, who have they recently interviewed, and be able to figure out what sort of their positioning is, what are their links are, so that I can do a bit further research on my own. All we need to do to create these reusable tools is to create this markdown file, this text file, give it a sequence of steps, like create a competitor analysis report on ex-person, start by researching why, or using why. Because Claude Code has research built in, use the Z script, so going back to, I think I have it up here, we have scripts as well. So we can reference these scripts that we have in here, so hey, I want you to pull in some external data with this script, and then use the skill, which I'll show you in a bit as well, how we can pull in other people's skills that allow us to add things into these workflows, so like using the PowerPoint skill to be able to create a PowerPoint out of this data. So you can see that automating your work is literally as simple as a sequence of uh instructions written in a natural language, when your Claude Code agent has access to the right information and the right tools. We don't just give one, we can stack a whole bunch depending on any function that you have in your job or role or business. We can stack them, so I'm other another one for me might be a market scan. Can you please scan my competitors, I can make a list of my competitors, like 20 people. It's going to read that competitor's file, it's going to go out and scan all of these, and then collate kind of a whole market scan, say, hey, for this on the Sunday, I want to see uh where the state of the market is, what sort of trends are popping off, do a scan of the market, come back and give me a presentation on it. And from here, the possibilities really are endless. You can get it to do basically anything you want, but that is the model, all right? So this is what we're working towards, a priming command that primes them on the context of who we are and what we do, and then we build tasks and workflows on top of that, that come in the form of reusable prompts. And so without further ado, let's jump into this Claude Workspace template that you guys will be able to get for free in the first link in the description. Uh you'll be able to download that and then follow along or just watch this, and then go have a play around with it after, that's probably a better way of doing it. And so in order to open this up and start using it for Claude Code, you're going to want to go to your favorite browser and search up VS code. You can come here, and then come up here and click download, whatever machine you're on, Mac or Windows, and follow the set up instructions that come with it. I'm sure you guys can figure that out. And then the next thing you want to do is go to Claude Code, click on Claude Code here. And then once you have VS Code open, you're going to want to go to file and open folder, and then you're going to want to click on the Claude Workspace template folder that you just downloaded. So it'll probably come in a zip. You want to unzip the zip, and then it'll should just be a folder, and then you'll go file, open folder, and then you'll see something that looks just like this. So we've open the folder, the root folder of our workspace, and in it there are other folders. The next thing we need to do is get Claude Code installed. And what we can do is go to the terminal tab at the top here, click new terminal. And then we get this thing down the bottom here, and this is important that you get very familiar with what's on screen here. We have the side panel, which has all of our files and folders, and then up here is our editor, so whenever I open up a file, it comes here. I can kind of like command W, uh to close the tab, or I can click the X. If I've changed it, I can command S to save it, and then close it. I like to have my screen split 50/50 here, so I've got this up the top, and I've got these down the bottom. This is your computer's terminal, and we can run commands here to get it to do things for us. I won't ever complicate it more than that, but what we want to do is head over to Claude Code, so just Google Claude Code. It'll come up with this website, we want to copy this, head back to VS Code, and then you can just paste it in here and press enter. Now it's going to take you through a set up wizard and ask you to do a few things. Pretty damn straightforward to do. All you're going to need is either a Claude Pro or Mac account, so whether you have a a paid subscription for Claude already. Uh if you have that, then you can just choose to use the Claude uh account that you already have. If you don't have one and you don't want to get a Claude account, you can go to console.anthropi.com. You can sign up for a developer or a business account here. And once you've created your account, you can go to settings, set up your billing. You'll want to add a card on file here, and then add five bucks into your balance, uh just so that it's showed that you've credited it and your card is good. And then you'll be able to go to API keys down here, create an API key, copy that, then go back to VS Code and select the API key option and paste it in. Once it's all set up, then you should have something like this, and you'll be able to type in Claude. And actually wait, I will not jump ahead there. If you want to clear this, so say I've written up, I've said a whole bunch of rubbish here, if I go command K, then it's going to clear it, that's a nice one to know, just command K. Um you can right click and go kill terminal if you want to start fresh and just start a new terminal here. But what you want to do now is type in Claude, which is the command to start Claude Code, and it's going to get it running here. So here we have it, I know it might not look like much, but this is your gateway to insane levels of productivity. The first thing you need to set up with Claude Code is something that I found to save so much time, because as I've said, we always want to be spinning up a new version of Claude Code so that we're starting with the minimum amount of context bloat possible. As you can see here, I've just asked Claude Code to uh look up these commands that it already has set up. And it's going to write a little file for me here, because as you see here, it's asking me like, hey, do I have permission to do all of this stuff? Naturally, Claude Code is quite hesitant, and it's always asking permission to do certain things. Now you can say yes, allow all edits during this session, but it gets a bit repetitive, and honestly waste a ton of time. So what these two commands here are basically ways of starting Claude in what's called dangerously skip permissions mode, or YOLO mode, or whatever you want to call it. Basically, this means that Claude is not going to ask permission to do things. It may sound scary, but if you have it within a a walled context like this, it is completely fine to use. I mean, I'm not taking any liability for that, but I've never had an issue, none of my team members have had an issue. Just don't put stupid stuff in here, don't hook it up to a live production database, which you won't be doing anyway. So you are good to do this, but what this gives us is allows us to skip over all of these, hey, yes, no, yes, no, uh, like yes, you can do that. But because we're going to be spinning up Claude Code so often, you want to be able to start it in that YOLO mode every time, and we also want to automatically run our slash Prime command. So as I explained before, our Prime command is really important because we're going to be using this at the start of every Claude session. So this is one of our commands over here, just a bunch of text, and Claude is going to read this and do what it says. So it's going to basically look at all of the files in this project, get an idea of all the different files and where they're located, it's going to read our Claude.md, which is our base level of context. It's going to read everything in the context folder here. So dot slash means in the root directory, so dot slash context means in the root, we're going to go slash, and then go into this context folder here. So if you close them all down, we've got slash content, we've got slash outputs, and so on. So this is just saying go into context and read everything that's inside it. And then after reading, come back to us, giving a brief summary of who I am, what this workspace is for, and so on. And because the Prime command is what we want to have run every time, we want a way of easily starting Claude and YOLO mode, and also ensuring that it always runs the Prime command. So if you see here, this will be included in the project file for you, so you guys can just copy this, and paste it into Claude Code when you're just getting things set up. So I've already got it set up, but this sets up two aliases, CS and CR. CS is for Claude safety, which just runs the Prime command, so this is if you want to run it in that permissions mode where it's always asking for yes or no. I basically never run it in CS, but it is there if you want it. But that's going to Prime it, and you're ready to go in safe mode. And then I have CR, which is Claude risky, and that's running Claude, uh skip permissions, or YOLO mode, and then gets it to automatically run the Prime command. So what you guys want to do is to copy this, and head to your power Claude Code down here, and say, hey, can you set up this please? Paste it in, it may ask for permission a couple of times, but trust me, you can just say yes. And then you'll have the ability to do this. So if I kill this terminal, open up a new one, and I go CR, hell, I might even do two side by side. So you can see, I can go CR, and CR, and spin up two of them side by side. You can see bypass permissions is on, that means YOLO mode is on, and it's automatically running a slash Prime. So you can see, it's going to read the workspace context. It's going through this prompt here. You can see that it's reading all of those context files. And then we get a summary back. So it's really important to get Claude to send that information back to us, and give us that summary. It's like, hey, this is a fresh Claude workspace, it has all of the placeholders, my role is to serve you as an agent assistant, here's the workspace structure, here's what is yet to be defined, and I am ready to assist, and so on. So that is just it repeating back to you, confirming that it understands the environment it's working in. That's so important to make sure that we're standing on that on that base that allows us to have an intelligent conversation. And so here we have two instances of Claude Code side by side, prime and ready to go. And if you want to be able to create more, like when I'm doing development, I'll have even more of these up. This is just me pressing uh command and then the button below delete, command backslash, and that allows you to very quickly open up a new terminal. So command backslash, and go CR. So that's the workflow you're going to be using a lot, because sometimes you want to kick off a create plan. Sometimes you're going to want to kick off an implement, but that is the basics of how to use this workspace. Now let me walk you through a little bit more on each of these documents and why they've been set up in this template this way. So we'll start with the context, building from the base up, we have the Claude.md file here. This is a template that's going to be changed around by you as you go. So when you are making changes, using the create plan command, it is instructed in here very clearly that uh the Claude.md should always be kept up to date with what is in the in the project. So that is already baked in, you don't need to worry about that. But this gives Claude, every time it spins up, Claude is going to read this and also the context file. This is basically the overview of the whole workspace and what it's supposed to do. Then we have the scripts, which you're aware of. Reference is a folder that you can kind of figure out what you want to put in it. It might be a an outreach template that I give it from the emails, then it could look at a list of your competitors. But reference is just for any docs that don't fit into the other ones. Plans are for using the plan command, it will automatically put the completed plan into the plan folder, which you'll see in a second. We have the outputs, so in this case, you're going to see me creating like competitor analysis reports, and they're going to be put into the outputs there. Then we have the context, and each of these four documents have a purpose, so the business context is kind of layer one. Uh I've got a bit of a, kind of a primer in here for you guys to fill out, because you need to get this information really dialed in, so you can fill these templates out yourself, or with Claude's help. We've got the personal info, which is who you are, and how you relate to the business. You've got the strategy document on like what sort of quarterly or yearly or project-based strategies are in play right now. And then you've got the current data. And for current data, you can either copy information off of your analytics or whatever it is, or you can connect it up using a script. Do it in a more automated way, pull your latest data and update this document. But I'm not going to touch on that within this tutorial. The last things we need to look at are these two commands here, create plan, which I'm about to use in a second. This is your kind of workhorse uh command or prompt, that allows you to do stuff within this workspace in a very, very efficient way. It basically says create a detailed implementation plan for changes to this workspace. And then when you run this command, as we will in a second, it takes the input that you put in, and goes, okay, now I'm going to flesh this out into a big plan. It's going to read the whole workspace, understand what files are where, it's going to again refresh itself on the workspace. It's going to write a full plan, breaking down why it matters, the specific files that need to be changed. So you can really comprehensively change the workspace, without you having to chat back and forth and do all of this messing around. And once it's created that plan, we then pass it into this implement command, where it's going to take that plan file, and then ask Claude Code to fully implement it and give us an update at the end of it when it's done. So that might not seem super clear, but let's just jump into having a play around with it. And I think things will start to make a lot more sense. So here we have Claude Code, and we're ready just to start giving it some inputs. So to use any of our commands, you just go slash, and then type in the command, so create plan. I'm going to press tab to fill that out, and then I can type into it what I want it to create a plan for. So in my case, I'm going to talk into it. I highly recommend if you guys aren't already using something like Whisper Flow here, uh Whisper Flow is something that allows you to very quickly uh talk into your computer, it's going to transcribe your voice, and immediately dump stuff in. And when you're working in this kind of AI workflow, workspace, uh environment, being able to talk into it and dump a lot of information onto Claude Code, um or any of your AI tools, and let it kind of figure it out from there, is a very, it's not super precise workflow, but it's definitely a lot faster than typing. So here we go. So in this case, for me, I want to turn this workspace into one that I can use to manage my podcast. I have a podcast where we have guests on, talking about their successful AI business, what's working, what's not, new tools that came out, and so on. And part of running that podcast is doing competitive analysis. So I'm actually going to swap out a few of these files here, and just delete these, because uh you need to go and change those and update them to be uh related to you. In this case, I can pretend like I've already done it, because I have. And I'm going to put these in as context. Then actually, because I've added that context in, things have changed around, I'm going to kill these terminals, because they've been primed with the wrong information. I'm going to start up a new terminal, I'm going to go CR. And so while that's priming, I have two documents in here. One is the about Liam and about me and my businesses, and then I have our podcast strategy and a bit of context on that. So what I actually want to do as the first step to set this up is to write a plan that's going to analyze the two documents, and then update the Claude.md, and anything else that it thinks is relevant to be aligned with this purpose of the workspace. So this is the first step that you guys are going to have to do. So I can go slash create plan. Okay, I want to set up this workspace as a podcasting workspace to help do things for my podcast. I have provided context files about me and about my podcast, and what I want you to do is put together a plan to uh update the Claude.md and any other parts of this workspace, to properly reflect the purpose of this as being a hub, where Claude is going to assist me in doing all the key functions of running my podcast, things like competitive analysis, and market scans. All of these are things I'm going to build in future, um so probably don't include those just yet, because I'll get you to plan those later. But can you please make a plan to figure out how how we can update this workspace to be fully aligned around this podcasting purpose?
[43:40]So basically, I'm just saying this needs to be fully aligned around this new function, and we're going to see what it says. And if you want to get the cool icons like I have here on my folders, you can come in and find different themes. I think, uh material theme is what I have, or icons. I don't know what it's VS Code icons or material theme icons, but it makes them look a little bit cooler, and it's kind of nice. Okay, so here you can see, it has created the plan for us. If you press command and click, it's going to open it up. And here it has written a plan. This was definitely overkill for just changing the MD file here, if I'm honest, but it shows you the workflow that we're going to go through for these changes. So as you can see here, the planning is extremely thorough, and it's absolutely overkill for what we're trying to do here. But it shows you just how detailed it is. It's about these are the new files to create, these are the files to modify, the the key decisions made, gives all the context needed. And then what we're able to do is, we're able to then use the slash implement command, and then actually implement this plan. So I copy this, paste it in here, and then if you run this, it's going to make all of those changes for you all in one go. It's going to make a big list for Claude to do, and it's going to tick, tick, tick, tick, tick, until it's all fully reshaped. But in this case, I'm just going to say that was overkill, can you please just update the Claude.md file? But you get the idea. If you're making major changes, you're always going to want to use a plan step, take that plan, pass it into the implement command, then it's going to fully implement it throughout the workspace, making sure everything is fully aligned and changed at the same time. Now, if we go to the Claude.md file here, we can see that it's changed. This is the command center for running AI for the rest of us, Liam's interview podcast focused on AI from a bootstrap business perspective. And it's given context around what this workspace is, and the commands that are in it as it currently stands. All right, so now that our Claude.md and everything about this workspace is fully aligned around what we're trying to do here, I'm going to command S to save that. And you can see the plan has ended up in the plans folder here, if you ever need to go back to them. I'm going to save that too. And now we get into the fun bit of actually taking a workflow and sort of putting the cherry on top out of all of the stuff that we've done. We fully contextualize the workspace, we've added in our context documents. Now we get to go into the fun part of actually creating our custom commands to do things for us. The example I'm going to use here is an analyze competitor command, which is going to take in a podcast name, or any other information about a podcast, might be a person's name. And then we're going to get Claude Code to research the person, using an external API like Apify to pull more information in about real-time uh data on the channel's videos. And then use a PowerPoint skill to be able to turn that into a PowerPoint presentation for me. So we're going to do a lot in this, all in one go. You could, and probably should, as a beginner, chop this into baby steps and say, hey, I just want to start off with just searching the web. And then I'm going to add in the external data, and then I'm going to add in the PowerPoint, but we're going to just YOLO it and try to get all four done at once. So what I'm going to do is spin up a new Claude Code command backslash there, CR, get a fresh one going. Then I'm going to start writing this up actually, so create plan, I want to, all right, so here's the plan. I've already done a bit of thinking on this, but if you're a bit of more of a beginner, and you don't really know that these things exist, as I'm going to go through and and point some of them out, you can kind of just say, hey, I want to do this, can you figure out how, like, in it say, search the web to figure out the best way to do XYZ, or I want to connect to this, and pull the data in, can you figure out the best ways or the best options? In this case, I'm saying, I want to create a new command called slash analyze competitor, which will do the following. It'll take in the name of a person or a podcast, it'll use one Claude Code deep research agent, which is a thing where it can prompt that to to say go into background research mode. That's much more powerful way of searching, it's going to like deep research for Claude Code. Um so I want to trigger one of these background research agents, which will go on the web to find all the links of the podcast, background information, positioning of the plot, information on the host, etc. Then I want to use the MCP integration skill to look up an Apify YouTube scraper, which can scrape their channel to get the 20 most recent videos and get real-time data for each. Then I want it to write me a research report in MD to read that provides a full competitive analysis breakdown. And then finally, I want to install the PowerPoint or PPTX Claude skill, and use that as the final step to turn that research report into a presentation for me to skim visually. So doing tons and tons of work here, that's the point of creating a plan, is that we can get Claude Code to go and do all this thinking, and come back to us with the plan on how to implement it. So as you can see, Claude Code is going away and doing all the research for me on how to create this, or how to automate this part of my workflow. Now, here it is downloading the Claude PowerPoint skill, and also the MCP required for the YouTube scraping. And here we go, it is uh created our plan here. So we can skim through it. It wants to do this, it's got a good summary. But it's made all these key decisions about the background agent, the MCP integration, Claude skills for PPTX. And then it's got step by step tasks for it to take action, to actually do this. So we are going to take this now and go, there we go, if you just press tab, it already had it it filled in for us, and we can send that off to do the work for us.
[51:03]So you just kick back and we wait.
[51:11]So here it's asking us a few questions. For the Apify API token, I'm going to head to Apify and grab an API key for this. So Apify is a great one for you guys to have a play around with. Very easy to sign up, very low cost as well, but it gives you access to tons and tons of scraping abilities. Very easy to integrate through MCP, which is what we're using for this. So create an account on Apify if you want to. While we're waiting for that, I'm going to go to the Claude Skills marketplace. This one here, skillsmp.com. So this is like a marketplace for Claude skills, if you want to pull some functionality in. In this case, I want to find a PowerPoint one. It doesn't look like Claude Code is able to find the one that I like to use. So we can come on to this marketplace here. Great thing for you guys to do is just come and have a a look around on this. I'm going to go back to home. Then I'm going to go PPTX, and here, this is the one from Anthrop, I'm going to download this as a zip. While that's doing its thing, we can come back to Apify here. You'll need to set up your billing, my case, I'm just going to go to settings, going to go to API and integrations. And our skill has successfully downloaded. I'm going to create a new one, say an expiration date so you guys don't steal it. And then I'm going to copy this, head back over here and say, here is my API key for Apify. And then I'm going to unzip this agent skill, and I'm going to move it into my skills folder. Then I say, also, I just added the correct skill to the skill folder. So read that, understand it, and get it integrated into the build.
[53:39]To pause it, as you saw me pause it there, it's just pressing escape, and it will stop Claude Code. You can give us some more information, then it will continue on. So it's going to continue with this implement command here, just now to the bit more information that I've given it. All right, so here we can see it is going through and doing all the steps as laid out in the plan. So it is created the MCP configuration to connect us into Apify. It has created the output directory. So if we go into outputs here, we have competitive analysis. It's made a new folder inside of our outputs, which has made sense because that's what we want them to put. And now up here, we should see any second the the command that it's writing is going to pop up here. We'll be able to skim over that before we actually put it into into action. So there we go, analyze competitor, it's written the whole command for us to use. Form a deep competitive analysis, it's going to take in some information that we provide it about who we want it to research. Prerequisites around setting up the Apify correctly, then it breaks down the workflow. First step is deep research, second is Apify scrape, then report generation, and PowerPoint presentation. Then it walks you through exactly what it should be doing at each of these steps, which is great. This is a really comprehensive prompt, breaking down everything it needs to do. And as you can see, it's also updating our Claude.md. And here, I'm a bit confused as to why it's made a script. So we can say, why did you make a script? Can you just do it for me?
[55:49]So you need to get used to just chucking everything back at Claude Code, and it will basically, because it can run these commands for you, it doesn't need me to go and do it myself. Boom, there you go. And so we have new one Prime and ready to go here. It should have full understanding of I'm Prime and ready to go, let's go slash analyze competitor, Lenny's podcast. So says our command here is all ready to use. I'm going to right click and close out of that, and now we just sit here and wait as it does all of our work for us. So we've just automated, well, we'll see how it goes, but I'm pretty confident. I will, it's going to go and follow all of the steps in this analyze competitor file, and we should get back in this outputs folder, in competitor analysis, probably a folder for Lenny's podcast. It's going to do a research. Oh, there you can see, it started a background agent, so I can go control O to see that. So the deep research agent is cooking away. All right, people, I had internet issues finishing that off last night, but we are back. And I'm here to show you the output that we got from this. So we ran the Prime, it's started the uh deep research agent, it searched in the web for any information about uh Lenny and his podcast. Uh deep research agent came back. Then it's going to prepare the MCP integration, uh making sure everything's all good, the MCP and Apify are ready to go. I did have to play around with that Apify API key, so I highly recommend, I'll leave it in the template for you guys, but here in a .env file is where you can put your tokens. And this is going to be a sort of persistent place for for you to store all your API keys so that your Claude Code agent can grab the information from there when it needs it. So once I've got that working properly, then it uses the Apify uh MCP to pull a whole bunch of information about the channel, uh using the scraper. We have this YouTube scraper that we are using that uh Claude Code figured out that this is the best one for us. It's able to do a whole bunch of things, but for us, it is taken the YouTube URL of Lenny's podcast, and pulling back all the information on that's latest 20 videos. Then it pulled a whole bunch of information back, got the 20 videos, let me examine the data structure, and then put together the competitor analysis markdown report, which we can look at here. Remember, click uh holding command and then clicking is what opens those up, and we have our report here. What you can do with the markdown file is right click up here, and go open preview, and then you get it nice and pretty and in proper markdown formatting. So massive reach and product manager space, Silicon Valley bubble is a real gap, uh premium monetization, he's got a newsletter, his name, his background, his handles, his positioning, the go-to resource for product managers, etcetera. All of this is great, bringing me up to speed on what this guy's podcast is about and how often he posts, episode links, one to two hours, frequency, one to two episodes per week, high polished, well edited, and so on. So this is great for me to just come up to speed very quickly, and then we have the final outcome, which is what we're really looking for here, which is it then taking that markdown report that we have here, and all the context that has, and turning it into a presentation. So that instead of me having to read that big document, I can just pull up the presentation, and we have that. Uh here it's created a couple of folders on the left hand side here, in our outputs folder. I'm just going to command click on this, I'm going to drop this down, and here we go. Overview, great, average views, posting frequency, engagement rate, top performing videos. So all of this is real data that's taken from YouTube at the time of running this. Content strategy, competitive analysis, doing a swap analysis of them, which is great. Strategic recommendations, positioning moves, and so on. So this is a full breakdown of a competitor analysis in a spreadsheet that's been able to be done through this analyze competitor command. And this is really just the start of what I'd do with this workspace, if I was turning it into a podcast hub for me. I can do market scans, I can do guest research, hey, I want to research this guest. Apify even has scrapers on here. If I try to find them, YouTube channel extractors. So I could integrate this Apify actor through MCP, and that's going to then be able to take a guest. Say I'd scan Lenny's podcast, figure out what guests were popping off, or maybe I'd do a market scan and figure out that, hey, these certain guests keep getting high views. Then I could say, okay, research this guest, and it will kick off a similar research process to what we just did here. And then I can say, great, can you go like slash reach out to, then it could look at an outreach template that I give it from the emails. Then it can use the YouTube channel email scraper here, pull the email from their channel, and then send them an email, or at least draft it for me before I'm able to tell it to send it. So I hope you guys can see just how crazy this can get, how much you can build into this. But these workflows are really the key thing that I wanted to get baked into your brain here. The importance of the priming, right? Remember, we need to be priming our session every single time when we spin it up. We want to be using new sessions of Claude Code as often as possible, rather than chatting away non-stop. We want to be able to say, slash, do this. It's as fresh as possible, minimal context. Remember, we've only got 200,000 to work with. And then we have the tools on top of it, and we're going to be using those to fire off the actions. And this is our task layer. So I hope that's been eye-opening for you guys on how I'm using Claude Code, how I'm teaching my teams to use it. This set up has completely transformed productivity within our organization. So I highly recommend you guys give it a go. The first link in the description down there will be a link to get this template, and just follow the steps that I've showed you in this video on how to set it up. Remember that, we have the shells alias here, which you need to copy and paste to Claude Code when you're just getting things set up. So then you'll have quick access to the sort of YOLO mode and priming setup here, so you can spin this Claude Code up as quickly as possible. So yeah guys, that's about all for this video. I think I'm going to leave it there. There's a ton more I could show you, but this will give you the sort of starting skills to and and the starting template to have a crack at this yourself. And yeah, like I said, I highly recommend you actually give it a go, because this is one of, this is what everyone who's been super productive is using right now. And if you can't figure it out now, you'll probably going to get left behind a little bit this year. So guys, thank you so much for watching. Leave a like, subscribe if you enjoyed it, you want to see more of this stuff, and check out up here if you want to watch another video from me, that's just as good, if not better than this one.
