TLDR
Claude can be used far beyond simple Q&A by leveraging its seven levels: choosing the right mode (chat, co-work, code), crafting effective first prompts, connecting external tools (Gmail, Firecrawl), spinning up sub-agents for parallel tasks, using skills (predefined recipes), managing safety and permissions, and running a Claude Code operating system for autonomous dreaming and optimization. Most users only scratch the surface, but mastering these levels can dramatically increase productivity and output.
Key points
- Claude has three usage modes: chat (quick answers, no file access), co-work (file access but fenced), and code (full access, best for building).
- The first prompt is critical; instead of a direct question, ask Claude to interview you to clarify goals and specifics.
- Connecting Claude to external services like Gmail, calendar, and Firecrawl gives it 'hands' to act on your behalf.
- Spinning up sub-agents allows Claude to perform multiple research or debate tasks in parallel, increasing output 5-10x.
- Using skills (predefined recipes from GitHub or custom ones) lets Claude build websites, apps, and systems with one command.
- Safety and compliance are managed by setting granular permissions per connector, following the principle of least access.
- The Claude Code operating system tracks spending, recommends skills, and autonomously 'dreams' improvements based on your history.
Tools mentioned
- Claude Code
- Firecrawl
- GitHub
- Graphify
- Gmail connector
- Google Drive connector
- Data Girl
- Glider
Techniques
- First-prompt interviewing
- Sub-agent parallel processing
- Adversarial debate with multiple agents
- Skill creation and reuse
- Least-access permission management
- Autonomous dreaming via operating system
Takeaways
- Use Claude Code for full access and building; don't be intimidated by the word 'code'.
- Always start a conversation with a prompt that asks Claude to interview you for clarity.
- Connect external tools and use sub-agents to massively boost efficiency and output.
- Set up the Claude Code operating system to automate recommendations and save money.
Transcript (captions)
Most people use Claude like a Google search, which is only 5% of what it can actually do. I have spent over 1,000 hours in Claude building, experimenting, and I have seen every master class. And in this video, I'll show you the seven levels of Claude that you need to know step-by-step and the mistakes that you have to avoid so you can save more time, grow your business, and get light-years ahead of everybody else. And if you're new, I'm Jack. I built a similar stack startup with a gazillion customers. Now, I'm building my own AI companies, and I share here the things that work. [music] So, if you haven't already, grab that beautiful coffee and let's dive straight in. So, these are the seven levels of Claude, and they get increasingly more important. Starting off with level one, which is where we use Claude. So, there are three ways we can use Claude. One is chat, co-work, and code. Chat is for quick answers, but doesn't touch your files. Co-work touches your files, but you're fenced in. And full access you get with Claude code. Don't be intimidated by the word code. It is the best builder. You talk, it builds. And so, inside the app here, you can see we have co-work in the middle, chat on the left-hand side, and then we actually have code here. So, download the app down below. Now, code is what we're going to be doing this entire video because it is the full unleash version. Don't be put off by the word code. And so, in the bottom right, we can pick the model within Claude. Opus 4.8 is the Albert Einstein of the model family, okay? Use this for the most capable intelligent work. Sonnet 4.6 is your day-to-day workhorse, and 4.5 is for fast, very easy work that's going to be very, very, very token light for you. And the next to this, we have this slider that ranges from low effort all the way up to max. For most things, I use either high or max. And then beyond that, we have one step, which is ultra code, which I'll explain later. Level two is talking to Claude properly. Now, this is so important to understand, but 95% of people don't even understand this. If you ask Claude, for example, there's a car wash 50 m away. And this was an example from Andrej Karpathy, one of the preeminent experts in AI. It will tell you to walk. I even proved this myself. I said, "Hey, there's a car wash 50 m away. How should I get there, drive or walk?" And it says I should walk to the car wash. It's 50 m. That's like 30 seconds. And the point here is that Claude basically executes well on your question, but it isn't necessarily thinking around corners. And but we can change this by the way that we communicate with Claude, which is why the most important prompt of any conversation is the first one. And you should give it a prompt like this. Hey there, the thing I am trying to achieve is my stated intention is building a strategy for my YouTube channel. I want you to interview me and ask me questions so you and I can be specific and clear on what goal we are trying to achieve. Challenge me and ask as many questions as you need to. And as you can see, the kind of questions it asked, what's the channel about? What's the current state? What does success look like? These are levels of detail that you don't get when you just give Claude a question. Level three is giving Claude hands. Imagine having the world's best, most capable builder on your building site, but it doesn't have a single tool. Wouldn't be very effective. So the idea here is that we can actually connect Claude with our world, with Gmail, with our calendar. So it can triage and draft emails. It can run our week and it can do anything we want to inside of Claude. For example, the way the best way to do that is to come down here, click on this plus like so, and you want to go ahead and click on connectors. And on connectors, come down and just simply click on add connectors. And here you can search for anything. For example, I can come down and find Gmail and I can connect that. I can connect to Data Girl. I connect to even my CRM. And we connect all these wonderful things together within Claude. For example, if I type in Google Drive and I click on this plus button here, for example, it will open up in the window and then we connect directly. So let's take Fire Coral for example, which is the best service that exists by actually grabbing information from webpages. It is fantastic. We can search, we can scrape, we can do anything. Now, I've connected this to Claude. So for example, if I come back over here, I can give it a command like this. "Hey there, head over to glider.com and grab me all their brand information and a summary of what that website does using Firecrawl." Now, Claude will natively understand which plugins to use based on the questions that you ask it and you don't even necessarily need to specify it. And as you can see, it's used Firecrawl to go ahead and grab that information. It's got the primary code, the background text. I can even say, "Hey there, great. Go and grab me their logo." And it gets all this information for me. Glider is a thing that I'm using to dictate right now and it can grab all this information using this. And Firecrawl's great cuz it saves us like 80% of our token cost cuz it doesn't grab all the HTML. It only grabs the specific information that we actually need. And just like that, it's downloaded the exact logo. And as you can see here on the right-hand side, it's pulled that from the website. And if this all sounds like Spanish or you want to know more, I'm going to put a link down below for the full Claude Code Masterclass that will take you through foundation setup, building websites, power features, memory system Hermes, all the stuff that I have never shared actually on my YouTube channel. It'll take you from zero to hero and you'll also get the full agentic operating system. I'll put a link down below so you can use that to level up. Now, this takes us very nicely onto the next level, which is level four. And this is about getting maximum performance out of Claude for the lowest price possible. And this is something most people miss. Now, level four can get you five to 10 times more output out of Claude for the same unit of time. Most people use Claude like this on the right-hand side, which is single layer traffic. They ask it a question, then another question, and when that's done, and so on and so forth. But, we can ask Claude to spin up sub agents, which is like using five lanes on a motorway instead of one, which means that the five sub agents can all do different tasks at the exact same time. For example, you could say something like, "Hey there, I'm trying to find the best AI company to partner with, OpenAI and Anthropic, Grok, or maybe even DeepSeek. I would like to spin up multiple sub agents, each of which to go ahead and ruthlessly research all of them and then come back to me with your findings. And then Claude will spin up a sub agent for each of these four things and then bring the results back to me. And as you can see, Claude comes back and asks us clarifying questions for us to be specific about. So, I go for a commercial deal. Let's go for raw model quality. And again, you can just answer these questions. And we can take this even further by coming down and using Ultra Code, which has a specific part of code which lets you do incredible things. For example, I could say, "Hey there, I'm 34 years old. I would love to know where to live based on my own specific circumstances. I would like you to do an adversarial debate. Have a loads of agents debate what would be the best thing and best location for me to live in." And what Claude will do is come back up with a plan based on everything it knows about you and actually have many agents debate each other. And as you can see, this is an example of what Ultra Code looks like. You can see the research phase, the verification phase, and it will synthesize and you can see what every single AI agent is doing. And you can trigger this by selecting Ultra Code here or just give a command to Claude. And the other thing that we can do to make Claude way cheaper and actually more intelligent is help it understand different files and code bases by using something called Graphify. Now, Graphify creates kind of these beautiful relational graphs between different code bases. Code bases just being some fancy speak for basically all the files that create any of the apps like this Claude Code operating system itself. Now, to do that, you can search this beautiful website called GitHub. You can click on this, click copy, drop that in Claude, and use it for any kind of code base you want to. And what it effectively means is that it gets the answers way quicker because it has a map and understands where everything is. For example, where is the login section? Where do I keep my files? Everything like that. And it all means that you basically get way better answers way faster. Meaning that you save so many more tokens and you can use Claude for way longer and it's just a couple of clicks to set up. But speed and efficiency are one thing, but if you're not using skills, you're not leveraging the power of Claude. Most people don't even know what this is. And the idea here is that we can give it any skill for any task, which is just a predefined recipe and way of doing things that makes Claude's abilities so much stronger. You can build beautiful websites that look like this and are like actually fantastic and stunning. You can build interactive websites, which I did actually in one shot and you can see the level this image generation, video generation. It is gorgeous. Even apps and systems that look exactly like this. To get them, all you're ever going to do is come up to Google and just type in the thing that you want. Say, "Hey, I want a front-end design skill." And you can just type in GitHub at the end of it and you can see all these things here. Or you can just ask Claude what are the best ones and you can find them. So, this one, for example, is from Anthropic, has basically 153,000 stars. All you do is come down to code, click on copy, and then say to Claude, "Hey, I would like to download and turn this into a skill." And the other way that you can do, once you give it that repo, is you can work with Claude to do a particular action. And when that's done, if it's something that you'll need to do again, just say to Claude, "Hey, turn this into a skill." And then you can activate that by doing {forward slash} and then you see all some existing skills that I've got here. Or just describe it and Claude will activate it for you automatically. And there are so many different skills. I cover lots on my YouTube channel. Effectively, you're limited only by your imagination. But your imagination can get you into trouble unless you understand how to leverage the sixth level, which is using Claude in a safe and compliant way. So, the idea here is that we give Claude the keys to the kingdom, but we do so safely. So, when we connect the different services, some of the stuff that we don't want to do is essentially give it more access than it needs. For example, in Claude itself, if I come down to manage my Gmail and I come come over here to connectors, I can see I've got these here. If I manage connectors and in the connectors we can select the permissions for things. For example, Gmail, I can see what I will allow it to do by itself and what requires approval. Meaning Claude just can't write and send emails for you without your express permission. And then generally follow the principle of least access which we only give it as much access as it actually needs. But access is one thing and it leads us very nicely onto the seventh level that if you don't get, you are not getting the full capabilities of Claude and that is running by itself and using a Claude code operating system. And the Claude code operating system tells us exactly how much money we've spent, not just within Claude but any other apps like Code X and anti-gravity. And crucially, it will dream for you. So based on every conversation you've had with Claude and all the different chats, it will find recommendations for you. What are skills that you can be using that you're not currently? How can I actually go ahead and save you money? And dynamically dreams and works for you where you're not even thinking about it. You can see everything you've got connections to mission control for goals, even your memory system which I covered in other videos. But effectively all the different systems that you have that are available so you can actually go in and check on these. And it just enables you to take your Claude to a completely different level. Now understanding the operating system is one thing but if you don't have it running and working for you, you are leaving too much value on the table. So the next thing we have to do is learn exactly how to get this set up which we're going to cover in this video right here.
Jobs for this video
| Stage | Status | Attempts | Last error | Updated |
|---|---|---|---|---|
| summarize | done | 0 | — | 2026-06-22 22:02:28.195508+00:00 |
| transcript | done | 0 | — | 2026-06-22 22:00:46.385806+00:00 |
| metadata | done | 0 | — | 2026-06-22 22:00:29.234511+00:00 |