# 5,000+ Hours of Using Claude in 20 Minutes (Beginner to Pro)

Channel: AI Edge
Video: https://www.youtube.com/watch?v=lyc7oT62uZI
Duration: 21 min
Language: English
Words: 4664
Transcript page: https://viewrankai.com/tools/youtube-transcript/lyc7oT62uZI

---

[0:00] So, in today's video, I'm going to give you the five levels to use Claude properly, condensing all of the knowledge that I've accrued running my 30-person AI media company and hundreds of agents over the last 18 months, condensed into a single piece of content with the five highest leverage things that you need to know. This video is equally going to be applicable to you whether you're just using Claude to become more productive in your personal life, or whether you are building and scaling a big startup. I personally spend over $10,000 a month in Claude tokens alone. So, as a business, we've been forced to optimize and look at

[0:33] exactly how to use Claude across the business to get the best and most efficient results. So, subscribe to the channel if you want to stay at the cutting edge of AI, and I'm about to make you a Claude expert. All right, so before we get into the more advanced stuff, let me cover the three surfaces

[0:48] of Claude. Firstly, you have the chat. This is what you've probably interfaced with most if you've used Claude, where you can basically ask questions, review documents. There's honestly a lot of functionality now built into the chat, especially with the additional plugins and connectors, which we'll get into later in the video, that Claude has now added to make this even more useful. So, if you just use the chat, you can still extract a lot of value out of Claude, but in my opinion, the way to level up even further, instead of using the chat, is to use co-work. Because Claude co-work is what enables Claude to read

[1:20] local folders and documents on your computer. So, if you're just using the chat, it doesn't have access to any of your context, only the instructions that you feed to a project. Whereas if you use co-work, it can tap into your local memory system and context, and edit local files, which makes it even more powerful. And then the third surface within the Claude desktop application is, of course, Claude code. You can also use this through another harness like Visual Studio Code, or through the terminal, but for simplicity's sake, you can use it on the desktop just fine. And this is the next step up from Claude co-work. So, if you're vibe coding

[1:55] applications, if you're doing builds, it's going to be more efficient and overall a better user experience to use Claude Code. But out of the three, where I recommend people start building their habits on Claude is actually on Cowork because you get a similar user interface to the chat, but you gain the advantage of being able to still code technically

[2:14] and edit local files on your computer. So, I guess which one I use depends on the task. If I'm just on the go and I'm brainstorming content ideas, I'll still use the chat. It's fine for questions and, you know, light document reviews and stuff like that. But if I want Claude to access the memory system that I want it to see, I use Cowork and if I'm doing a bigger build, I use Claude Code. Now, before we get into level number two where I'll actually show you how to configure and manage your own memory system inside Cowork, let me touch on what might be the most

[2:42] underrated feature of Claude and that is the connectors capability in the application. Connectors essentially enable you to level up the chat because it can connect to other data sources and other applications via an MCP, which you can essentially think of as a bridge. So, instead of having to go into Higgs Field, for example, to produce video or photo-based content, you can simply add the Higgs Field connector and stay within Claude, but get access to Higgs Field's models like Seed Dance, like Gemini's Nano Banana. So, you can actually use Claude as a harness through connectors, which basically means the place that aggregates things, the user experience layer. So, you don't need to

[3:21] be hopping from AI tool to AI tool. Most people don't know this, but through connectors, you can actually use GPT within Claude. That is for certain tasks where you think GPT would be the better model. You can also add Google Drive, so it can view and edit things on your drive, which is super handy, especially for our company drives, which we manage on Google Drive. You can add Notion, so it can edit your to-do list and you can use Claude as a personal assistant. And the one I really like is FMP and also Quiver, which enables me to tap into, which I discuss on my finance channel,

[3:49] all sorts of data around stocks, crypto, my portfolio, so I can get live data based on stock earnings calls and market movements. The issue is if you just use the Claude chat and you don't use any connectors, you are doing yourself a disservice because you're just going off what Claude knows. Whereas, there are data companies like FMP that are literally just built around niche financial data. So, if you use the connector, every time you chat to Claude, you're going to get much better responses. For example, I can use Claude via FMP right now to analyze Nvidia's earnings call. This will tell me everything I need to know about the last

[4:24] earnings report for one of my investments. And if you want to know how to add a connector, just go into Claude settings, scroll down, click on connectors, and then you can search from a pre-made library of connectors. You've got Gmail, Slack, Google Calendar, pretty much anything that you can do on the internet can now be done by Claude. Honestly, one of the most underrated tips because everyone's talking about Claude code or the Claude chat and not addressing the fact that Claude actually needs access to information. This is the best way to get the most accurate information out of Claude and it's now inbuilt into the app. All right, now you

[4:54] know how to get more power out of Claude, but unless you give it the right context about yourself and about your business, then you could have all power in the world, but it'll just be useless. I think we've completely moved away from a prompt engineering environment. Remember when you had to like sit down and write these massive prompts and focus on giving it context and all this information because that's what you previously had to do to write a good prompt. Well, that environment is completely shifted now. Well, now I blurt out a lot of my prompts. I either do it through voice, through WhisperFlow, or I just write badly

[5:25] put-together prompts because it doesn't matter anymore. Because as long as Claude has the context about my business, it'll be able to apply what I'm trying to say to my real context. So, instead of putting time into learning how to prompt, put time into building a memory system that you truly own, which more importantly can be then ported to any LLM. If you want to use GPT, if you want to use Gemini in the future, whatever you want to use, you will own your own memory and Claude won't own your memory. Because unless you own your memory, Claude is a black box. Obviously, if you've ever spoken to

[5:53] the chat, it might remember things about you, but you don't have any sovereignty over what it's remembering or why it's remembering a certain thing. You might notice sometimes Claude, you know, is really weird, it'll bring up like a random fact. And then when you're talking about buying a new car, it'll bring up, "Oh, but you know, remember the new camera you were supposed to buy?" It'll just like inject all of this random stuff. So, to get around that and to have full sovereignty and ownership over your own memory, what you want to do is create a memory system. So, there

[6:18] are two ways to set up a memory system. The first way is to set it up locally, which I'll show you, and the second way is to set it up on the cloud, which I'll also run you through. Now, both have pros and cons. Locally is easier to manage and is probably what I would recommend for building your first context and memory system. But then, if you're going to create applications in the future, and if you're really hopping, you know, if you're more advanced between multiple LLMs, you may want to consider putting it on the cloud. As long as you have good protection, like two-factor authenti-

[6:45] cation. The other benefit is, if I don't know, I spill water over my laptop and my laptop dies, I have a backup of all my memory and I don't lose it. Maybe if it's like a basic context profile on yourself, it doesn't matter so much. But if you were to build out a full memory system for your business and something happened to your computer and the file got corrupted, you'd be distraught. So, that is where maybe considering a cloud option makes sense. Or of course, a lot of people that just have their memory systems locally and then every now and then they'll do a backup of their

[7:10] computer and they'll upload it to a drive or something, you know, so they have a cloud backup. Firstly, going to create a local memory system, this is what I'd recommend. Just create a folder on your computer. This is my AI brain. It essentially has context, so this is a file with all the context about myself, my business, what I do. To create this for the first time, all you need to do is do a voice prompt insertion with Claude. Ask it to ask you questions about yourself and about your business and it will write an entire MD, which is a markdown file. Don't do PDFs because

[7:38] Claude finds it harder to read. And also, make sure when you create an MD file that you scrutinize and you go line by line through each line and you put a bit of manual work in up front to make sure that what it's saying is correct because what's going to happen in the future is every business decision, every prompt is going to be based on those facts. So, you know, if you say that you have a $100,000 net worth and it accidentally transcribes, you know, a million-dollar net worth, then all future spending decisions are going to be based on, you know, 10x your actual

[8:07] net worth. That's obviously a big issue. So, you want to make sure that each line is actually, you know, true to you. The next thing you want to do is create an instructions.md file. This file tells Claude how to behave. So, you can put in preferences like, you know, always make your responses super analytical, abbreviated, don't ramble, don't confirm my bias, think strategically, and very importantly, in the instructions file, you want to put a line saying, "Notify me every single time you make a change to my memory.md file and make sure that every substantial new piece of information is used to update materially the memory.md file." So, what is the

[8:47] memory.md file? This is essentially Claude's brain. So, it's going to have all the context on you and what you tell it. So, every single time you give it new information like, "Okay, I just landed a new big client for my business." or, "You know, I'm pivoting more into SaaS products and I'm going to

[9:00] have SaaS attached to the info product." You would tell that to Claude and if you have the instructions set up correctly, your memory.md file gets automatically updated. And every now and then, what I recommend is you manually go through the memory.md file and you call it down just to make sure that it's manageable cuz the problem with Claude or any LLM for this matter is if the memory context is too bloated, sometimes Claude actually gets confused and the response time gets slower and it burns more tokens because it's reading through this five-page memory.md file. So, what you want it to have is just the core memory. Then, what

[9:31] you can do is in the instructions.md file, you can have a note that there is another folder attached called, for example, AI Edge YouTube, which is one of my business, or whatever your business is called, that has additional business context, only read that folder when I'm asking a question applicable to the information in that folder. Because within that folder, that's where you can have all of your business context and data. So, this is quite a light example for this video, but you could literally have an MD file or HTML mind map of your

[9:59] entire business, your funnel strategy. Like, this is where you can just dump and save all of the more complex information about yourself and your personal life, or [snorts] you know, your business if you're running a business. And if the instructions are set up correctly, then it won't be reading through that entire folder every single time that you need a response to a question, or once again, it'll just bloat the system. So, essentially, creating a memory system comes down to context, instructions, and memory. If you do that, and you plug it into Cohere, by selecting the folder that you store all of this information, every single time you do a prompt on

[10:31] Claude, it can edit that folder, and it can review that folder so your responses are tailored to you. This is the single biggest improvement that you can make to using any AI because it actually has the context on you. Now, the second option, as I said, is the cloud option. I won't go into too much depth, but you can build your own custom system on Superbase using Postgres. The way to do this is basically just ask Claude to build it for you, it'll be able to build it. Or secondly, you could use a purpose-built tool like Supermemory. Or if you're using an application to vibe code like Base 44, they also have their

[11:01] own proprietary memory systems. I think Supermemory is great for people that want a unified experience across all of their LLMs and apps. So, it's worth considering. Superbase, you have some memory sovereignty. If you ask Claude, once again, what is best for your needs, depending on the complexity of your business, it'll be able to tell you. But, I would recommend, if you have not built a memory system, to start with a local folder so you understand it, and then upgrade to cloud later. All right, now it's time to move on to level three, which is the first level where Claude is literally going to be doing the work for

[11:30] you. And I just want to remind you, because now things are getting more advanced, that the entire cheat sheet from today's video will be available for free down below in my free school. This is a brand new community that I just launched to help you level up with AI. It has all of my free resources, and I think we're building an absolutely amazing community there. So, if you want the cheat sheet for today's video, which includes the setup guide for you to set up skills, set up a memory system, and everything I've talked about that you can actually just put into your own Claude to make the information from this

[11:57] video applicable, then check out the free school down below and come be a part of the AI Edge community. All right, so level three is extremely important. As I said, this is how you genuinely start automating your life and business, and that is Claude skills. Because a memory system means Claude knows you, but skills learn how you actually complete specific tasks. I believe this is the highest leverage feature in the product alongside connectors, and almost nobody uses them properly. So, right now I'm going to quickly show you how to set up a Claude

[12:24] skill and how to use a Claude skill. What you need to know is that a skill is essentially a folder with instructions in it. So, whenever you complete a task, Claude knows exactly the steps and the preferences that you have in order to complete that task in the most efficient manner. So, you know, if I want to script a video like this, every single time I work on a video, I use a Claude skill, which learns what I like because I keep giving Claude feedback, and it updates the skill every single time. So, I can have an entire script spoken like me, formatted like I would, and

[12:54] presented in the way that I prefer. And every single time I do a script, I give it feedback on how it did, so it gets better and better and better. Or you can have a copywriting if you're trying to get sales leads or if you're trying to write posts for X, whatever you're trying to do, where you can train it to speak like you. Obviously, it's easier for me cuz I'm creating content, so we can just take the transcripts, and every time I film a video like this, we can feed it to Claude so it learns my real

[13:14] intonation, my real vocabulary choices. So, copywriting for a content creator is easier, but you can mimic this just by asking Claude to ask you some questions, and you can just speak like you normally would, and it can learn how you speak. And you can improve it over time. And the possibilities go way bigger than that. You could build a Claude skill to scan your competitors. You can build an ideation skill. You can build a skill which hunts properties for you, finds opportunities in the e-commerce drop shipping space. Anything you were doing, you can automate with a skill. And now I'm going to show you how. So, what you

[13:42] want to do is you want to go {forward slash} skill creator. This is basically going to set up the creator that's going to ask you questions in order to train a skill. This is best used when you can either verbally through voice transcription explain how you like to do something. Or if you want to feed it the data which it'll ask you for, like documents, PDFs, files, MD, whatever, to actually train the skill. But something that is very cool that Claude recently launched which made skill creation even easier is this feature right here. If you click on the plus button next to the chat, you can actually now record your

[14:14] screen. So, it'll look at what you're doing in real time. And if you transcribe what you're doing, it'll also be listening to your microphone. And you can run it through an entire process. So, what I did is I ran it through my entire competitor scanning process where I sit down, I look at competitor channels, I look at what performs well, and I train the skill to automatically look at what makes it bad video a good video. I talked it through the entire process. And now, whenever I want to scan YouTube, it searches the search term that I'm interested in, for example, Claude. It looks at all my

[14:41] competitors. It analyzes the data on a reoccurring basis. And then it comes back to me with proven vetted content ideas. And I trained it just through this screen recording. Or for example, if you're in a job where you have to do data entry, you can do the data entry like let's say you work at a finance firm or an accounting firm. You can do the data entry and you can explain why you're doing what you're doing. Because maybe Claude isn't the best if you just give it a spreadsheet because it might not fit the criteria that your company is asking for or it might not fit the

[15:08] exact work flow you're trying to achieve. But if you've walked it through and you've spoken while showing it what you're doing, it'll pick up on those little nuances. So, this is kind of crazy, but something I would recommend that you actually do is whenever you're working on your computer, try and record as much as possible. Like every time you're doing a task, try and record it

[15:25] and explain to Claude what you're doing. Because now it doesn't only just hear, it also sees. And this is a massive opportunity for you to be able to literally automate your work. And as your work evolves, you can keep doing more recordings to update the skill over time. Now, anytime I want to do certain things like analyze competitors, like writing, like video scripting, now I can literally just use my skills and I can

[15:44] have it done exactly how I want it done. This saves so much time. Yes, there's a bit of upfront investment, but it's an absolute game-changer. If you want to get the most out of Claude, instead of what most people are doing, which is just opening new chats with no context. This point in the video is going to get even better with level four and level five, by the way. You've just stacked a memory system on top of skills to automate work, but now let's get into the really fun stuff. All right, so everything so far still required a bit of manual work. Even skills, technically

[16:09] you need to sit down and prompt a skill. What I'm about to show you next is the best way to use Claude in 2026, and that is loops and scheduling. I'll focus on loops first. A loop is essentially Claude doing a task on repeat without you triggering it based on a preset of variables and instructions. There are two commands that I love using in Claude. The first one is forward slash goal. This tells Claude exactly what you are trying to achieve. For example, I want to build a personal habit tracking dashboard with the following constraints. And then I'm going to use forward slash loop. Based on my entries

[16:45] every day and my behavior, automatically adjust the dashboard to better optimize my productivity. That's one example of a loop. An even better example is for content. The example I showed you before, you can actually use a Claude skill with a goal and a loop. Let me show you. Goal, come up with the top five video ideas based on what is

[17:02] currently performing well on YouTube. Loop, every 24 hours, you need to specify the duration, look through the top ranked videos with the following search terms, and update the list on a spreadsheet ordered by your highest conviction ideas. Use my video research skill to follow the exact same mental framework that I trained you on. So now it's using a skill which is trained on a loop, so telling Claude how often, what variable, what interval it needs to repeat that task, and with the goal in

[17:30] mind, so it knows what the end goal is. The way the new Mythos class models work, and GPT also works with this as well, is they are so smart now that they are able to self-correct their own work. So even if you're vibe coding an application, running loops to do audits is a massive game-changer because it can actually vibe code for you on a loop automatically. And if you want to schedule a task, you can go into the scheduling function of Claude, and you know, have a set time when a task is scheduled. For example, every single morning I have a research report which is sent to my inbox based on what

[18:03] happened in the market overnight, connected to the FMP connector from before, that actually has my context on my own portfolio. So every morning I see exactly what I need to know about my own portfolio. That is done on a scheduled cadence. You don't need to run a loop alongside a schedule. So you could just schedule a task on its own, but for tasks with shorter intervals, I prefer to run loops. And for tasks that are more like daily or weekly, I prefer to run schedules. But forward slash goal, I use nonetheless to actually tell Claude what I am trying to achieve. And now let's move on to the final level, the

[18:34] master level, and something that honestly in my business took a long time to crack, but I feel like we're finally getting there and it's so exciting. This is the creation of agentic workflows. Now technically in level four, a loop is [music] an agentic workflow because it is going to spawn agents on your behalf. But to level up, you can create a system which orchestrates, especially for complex tasks, automatic behavior and automatic model selection. For example, you can have Claude Fable, which is like still the smartest model on Claude, but it could change. They're always updating new models. But right now I use Claude Fable as the orchestrator, and then I

[19:07] have Opus 5, and then I have Claude Haiku and Sonnet based on how complex the subtask is to complete work so I can actually reduce token cost. So essentially have different models doing different jobs. And what you want to tell Claude Fable is to spawn sub-agents to complete a certain task. You combine that with the scheduling feature that I talked about before or the looping feature and you can actually have tasks running automatically. I did a video on this recently but something that's really been helping me in my work is graph engineering. So basically breaking down a process, for example, a video production, you know, all the way from

[19:39] ideation to publishing the video, I will write down each step of the process and I will visualize in a mind map graph who and what model needs to be involved at every step of the process. I'll also put in where a human needs to be involved. Like I'm the bottleneck a couple times in the process but I can visualize which agents are used at different points in the process. And if you create this for every single task and create a skill for it, Claude will be able to orchestrate

[20:03] your agentic workflow on your behalf. This is the ultimate way to do things and I've done a full video on this. So if you want to check out my graph engineering video, I'll link it below. And if you want the full setup guide from today's video, it's available down below for free. Come and be a part of the community and you'll get access to it. You can put it straight into Claude and you will automatically be able to apply all of the knowledge I gave you today to your real workflow and set up

[20:24] your own memory system from scratch. Hopefully today I've helped you level up your knowledge of Claude. These are the things that have really moved the needle for me but of course these models are only getting better and better so make sure you subscribe to the channel for more content like this and when I learn things I'll share it with you to make sure we're at the cutting edge. See you

[20:40] in the next video. Peace out.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt