§ 01 · The Hook
The bait, then the rug-pull.
Most people install Claude Code, run a few prompts, and conclude they have seen what it can do. Nico has spent years building AI automation systems for thousands of business owners, and his argument is blunt: treating Claude Code like a chat interface is using a fraction of what it can actually do.
Where the time goes.
00:00 – 01:03
01 · Intro -- the five-level map
Host introduces the five-level framework and his background building AI automations for businesses. Sets up the diagnostic promise: where are you, and what is next?
01:03 – 04:26
02 · Level 1 -- Manual usage and basic prompting
Installing Claude Code, basic terminal usage, /usage and /plan-mode commands, and the dangerously-skip-permissions flag demonstrated with a live Anthropic news report task.
04:26 – 07:08
03 · Level 2 -- Persistent memory (CLAUDE.md)
/init command generates a CLAUDE.md from existing project files. Shows how to update it conversationally mid-session. Session amnesia is the ceiling of Level 1 and the entry point for Level 2.
07:08 – 14:53
04 · Level 3 -- Skills, agents and plugins
Skills as reusable SOPs invoked by name. Live split-screen: front-end design skill vs. plain prompt. Agent demo: keyword research agent with DataForSEO MCP expands its own search scope independently. Plugin demo: /build-website creates and deploys a full site to Cloudflare.
14:53 – 18:59
05 · Level 4 -- System integration (MCPs)
MCP stands for Model Context Protocol. Claude Code connects to Webflow, Notion, Canva, Figma. Live demo: video-to-blog skill with Webflow MCP scrapes YouTube, rewrites in author's voice, interlinks site sections, and publishes in one call.
18:59 – 23:00
06 · Level 5 -- Autonomous workflows (cron jobs)
Headless mode plus cron jobs let agents run on a schedule. macOS requires granting /usr/sbin/cron full disk access. Recommends building a custom command-center dashboard to track headless runs.
Visual structure at a glance.
Named ideas worth stealing.
00:00
model
The Five Levels of Claude Code
- Level 1: Manual prompting
- Level 2: Persistent memory (CLAUDE.md)
- Level 3: Skills, agents, plugins
- Level 4: MCP system integration
- Level 5: Autonomous cron workflows
A progression model for Claude Code capability -- each level unlocks qualitatively different behavior, from one-shot text to unattended business automation.
Steal for Any tutorial or course positioning Claude Code as a career or business skill
Lines you could clip.
00:15
"Most people install Claude Code, try a couple of prompts, and that's it. A glorified chatbot almost."
04:14
"When I exit this session, Claude forgets everything. What we did, all my businesses, anything I might have discussed. That's where you hit the ceiling of level one."
18:56
"Claude just went from a tool that lives in your terminal to a tool that interacts with the real world with any system that you use."
22:00
"If you learn how to master it as an autonomous system, you'll be ahead of 99% of people out there."
Things they pointed at.
14:53toolWebflow MCP
14:53toolNotion MCP
14:53toolDataForSEO MCP
14:53toolCanva MCP
How they asked for the click.
23:03
product
"If you wanna see how these five levels can completely automate your marketing, getting you more leads and sales, I recommend you check out our community."
Soft community pitch at the end after substantive tutorial content is complete. Not intrusive -- the video stands alone without the upsell.
Word for word.
HOOK opening / re-engagementCTA the pitch
00:00HOOKThere are five levels of using Claude code for productivity, and each level unlocks a completely different way of working. Most people install Claude code, try a couple of prompts, and that's it.
00:11HOOKA glorified chatbot almost, And they don't realize that they're only using a tiny fraction of what this incredible tool can actually do. So in this video, I'm gonna walk you through the five levels of using Claude code and how to level up from basic prompting to building complete systems that can automate complete parts of your work saving you hours of time.
00:31HOOKAnd regardless of whether you're just starting out with Claude code or you've been using it for a while, this will help you figure out where you're currently at and what the next level looks like you start working faster and more efficient with this tool. And if we haven't met before, my name is Nico, and I've spent the last few years building AI automations for businesses and marketing workflows.
00:50HOOKAnd inside our community, we've used these kind of systems to automate tasks for thousands of business owners and AI enthusiasts like yourself, helping turn hours of work into mere minutes. Without any further ado, let's get into it. Level one is manual usage and basic prompting.
01:07You're using Claude as you would in the normal chat interface. You should already know how to install it. I'll leave instructions for you in the video description if you haven't already done so.
01:15And by now, you know how to open Claude code by opening up a terminal and typing in Claude. Now we see our friend show up here, and you can get it to do some fairly basic tasks, like, for example, research the latest news on Anthropic and come back to me with a report.
01:31With that instruction, paste it. I hit enter, and we'll let Claude do its thing. And here we have our report.
01:35Now with here, we can also get Claude code to turn this into a nice design HTML report. So we'll ask it just that.
01:42Can you please turn this into a HTML report that I can see? At this level, inevitably, Claude will ask you for permissions. In a second, I'm gonna show you how to run it in a version where it won't ask you for this.
01:52We're just gonna say yes, allow this session. While that is doing its thing at this level, you should also understand some forward slash functionalities. So let's finish the report.
02:02It's opened it up in my Chrome browser here, but let's go to the forward slash commands that you really should know. One of the things that I think everyone using Claude code needs to know is the usage. I'm gonna type in usage, hit enter, and this tells me how much of my current allocated token usage I have used.
02:19So if you use it with a subscription, which I highly recommend that you do, it'll reset reset the usage every five hours. So for example, for now, I've only used 2% of my current allocated session.
02:29Another thing that you should know is you can tab through a couple of the settings. You can see here accept edit on. And if I hold shift and tab, I can cycle through a couple of the modes.
02:39One of the most useful tools here is plan mode, which you should use if you think you're gonna do something with a bit more complexity to it. You put plan mode, you tell Claude what you wanna do, and it'll think about the task at hand before it actually does it.
02:53Now the last functionality I believe everybody needs to know at this level is a setting called dangerously skip permissions. Obviously, there's a little bit of danger here, but it makes Claude code so much more efficient. Let me show you how that works.
03:05If I'm in the same session, I'm gonna exit out of this by going forward slash exit. That will end the current session, and I'm going to now start Claude code with this special permission by typing Claude dash dash
03:19dangerously skip permissions. Once you hit enter, you'll see now that you've got this new setting here bypass permission. So let's get it to do the same task that we asked before, to do research on Anthropic, to create a HTML report, and to open it in my browser, and you'll see that it won't ask for any permissions.
03:35Can you please do research on Anthropic for the latest news over the past seven days? Once you have that, turn it into a report. Make sure it's a HTML report and open it in my browser once you're done so I can have a look.
03:46With those instructions set, I'm gonna hit enter and let it do its thing. And here's the completed report in my browser without me having to allow any extra permissions to Claude. I use this setting all the time, and you'll see later on in this video how this setting gives Claude code so much freedom and power that it can build SEO optimized websites with one single prompt.
04:04So that's level one. You install it, you prompt it, and you control how much freedom Claude has. It works and it's useful, but there's a very big problem.
04:14When I exit this session, Claude forgets everything. What we did, all my businesses, and anything that I might have discussed with it during this session. That's where you hit the ceiling of level one, but this is where level two solves that problem.
04:26Level two is persistent memory. What if Claude code could remember what you're doing, what your goals are, how you like to write, and any preference that you have, not only as a user level, but even in specific projects that you're working on. This is all done with a Claude dot m d file.
04:41Looks a little bit like this. Let me show you what the file actually looks like. So I've got this site inside a project where I do the majority of my SEO work, and it just has an understanding of the metrics, what we're working on, the folder setup, and even what MCP integrations this folder has access to.
04:57More on MCP integrations in a second. And this allows Claude Code to be a lot more efficient. So what happens if you haven't built one yet?
05:04Well, you definitely don't need to type one out. I'm gonna show you a command that allows you to build this MD file in one single command. So let's say I'm working inside a folder called AI search consultant,
05:14and I already have a bunch of files in there, a website build, maybe I've got some keyword research or whatever it is that you're working on. I'm going to right click the folder here and open up a new folder at a terminal. I'm gonna activate Claude code by going to Claude,
05:29and now I've got Claude open in this folder. One of the easiest way to create an MD file is go forward slash I n I t, init. And I'm going to hit that, and you can see that it says initialize a new Claude MD folder with this codebase.
05:44It's going to analyze all the files and folders within that folder and create an MD file for you. Now we should see in a minute a new file added in here called Claude dot MD.
05:55Now that it has a good understanding of the project, it'll see it'll create everything it needs to to create your claud m d folder file. Sorry. I'm gonna say yes, allow, and there is our claud m d file that you just saw was created right then.
06:08And here is the m d file that it generated. It gives us the architecture, the file, and all the instructions that it should have so it has a good understanding of everything, even the schema, the type of animations that we have for this whole project for the website. So what happens if I wanted to add more to this MD file?
06:25It is really easy. In the same terminal that we are having this conversation inside the folder, all you need to do is prompt it in a manner like this. Now please remember this or add to your MD file that whenever you write, do not use em dashes.
06:38But we're gonna put that in here, let's just have a look at the MD file whilst it updates that. And that was the last one here. It said never use em dashes in writing.
06:48Use colons, commas, parentheses, or separate sentences instead. So now Claude is consistent and understands the project that it's working on, but you still have one very big limitation, and that is that you're still doing everything one prompt at a time. What if one command could replace thirty minutes of you going back and forth with Claude?
07:06Well, enter level three. Level three is skills, agents, and plugins. And this is where Claude code stops being a simple AI chatbot and starts being almost an autonomous employee,
07:17but that'll come in level five. You can think of skills as set of instructions that you want it to follow the same time every single time when you invoke that skill, kind of like standard operating procedures. Now if you've recently installed Claude code, you should have some skills already available to you.
07:34And you can see by opening Claude code, going forward slash skills, and hitting enter. And you'll see here that it'll tell you all the plugin skills that you've got available to you, even one of my favorites, front end design skills.
07:47Now there is a marketplace also for skills. You can access that marketplace by going to forward slash plugins, and then you'll see the marketplace
07:58for skills. One word of warning, when you are installing plugins from anywhere, make sure it's from a high quality source. I recommend you just start with Anthropic skills.
08:08Because if you're just downloading skills from anywhere, any random GitHub repo, they might come with malicious prompt injection in it, and you just wanna be very wary of that.
08:18Now let me show you what a task can look like without a skill and the same task with a skill so you can understand the incredible difference. So I've got two Claude code open. On the left hand side, we've got a prompt saying, build a landing page for Yervamate,
08:33which is a tea that I drink all the time. Make it responsive. Use the front end design skill.
08:38Use HTML only. And at the end of the file, add the name with skill. On the other side, we're gonna have Claude code open as well, practically the same prompt.
08:47All we're gonna change is do not use the front end design skill. You can invoke skills, by the way, by just telling it, hey. Use this skill.
08:55So we're gonna hit enter, let it do its thing, and then see the incredible difference that using a skill for the right task makes. Okay.
09:03Now that they're both done, let's look at the difference. So this is the no skill. You can see here in the name, which isn't too bad.
09:11Feel the natural energy of South America. Yerba Mate, it's got a few little components here with a little animation, rooted in tradition, the benefits,
09:21the community. Not too bad. Now let's have a look with the front end design skill.
09:26You have tiny little elements that are moving here, which is really nice. Uh, the design already looks a lot more professional.
09:34You have moving components, and overall, it just looks a lot nicer. It's used the same color palette, is fair enough, but you can see here with one single skill
09:44evoked, the design is already so much better. You can also build your own skills quite easily, and you don't need to know the folder structure that is more effective.
09:55If you go to forward slash skills, you should see that natively there is a skill called skill development skill. So let's say you're doing some work with Claude. You are the for example, turning a YouTube video from the transcript into a blog post.
10:08You do that manually. You can then say at the end, hey. Can you invoke the skill development skill and turn this into a skill?
10:15Skillception. But you get what I mean. So you can build an arsenal of your own custom skills that will allow you to do your work a lot more efficiently.
10:23Okay. Let's get into agents, and this is where things get extremely interesting. Agents are essentially specialized AI assistants that handle specific types of tasks.
10:32You can create your own agents, by the way. We'll test out one of mine in a second. Second.
10:36But each agent runs its own context window, its own system prompt. It can use individual skills, access MCPs. I'll show you how to do that in a second.
10:44Think skills, but just a lot more autonomy. So let's test one out. I'm going to go back to my Claude and say,
10:51using the keyword research agent skill, I want you to do keyword research on the word AI search consultant. So while we let that do its thing, I wanna show you what that agent looks like. In a new chat, I'm gonna go forward slash agents.
11:07Again, this is gonna show me all the agents that I have, and it even gives you the option of creating your own agent. I've got the keyword research agent there. I'm gonna view the agent, and here are all the specifications
11:18and custom set of instructions that the agent has. You can even set it a color here. But more importantly, you can see that we've given you the instruction to use the data for SEO MCP.
11:28Agents with MCPs explode your productivity to another level, giving this agent access to real SEO data. Even if you don't do SEO,
11:38you'll understand the power of giving agents access to an MCP. And this is a little bit more different than a skill because the agent has a set of instructions. They can think about the responses.
11:48It's not just following a step by step program. You can see that it's called the data for SEO MCP successfully, and it's even analyzing more variations because it thinks it needs a little bit more data.
11:58And here's the incredible aspect of using an agent because it has an insight here that said why this is interesting. AI search consultant itself
12:08has zero search volume, meaning it's very, very new and it's emerging. It has some search volume but very little.
12:14So it said, well, what else can I search that's going to be about this keyword? And it did some more keyword research, AI SEO, AI search optimization, geo SEO, and it found all these other keywords, transactional or service keyword, geo consultant,
12:29AEO services, AI SEO consultant that has some search volume. So you can see that it can take kind of actions and it can think about the steps.
12:38If it didn't find data about this keyword, it's gonna think, well, okay. Let's find data about another keyword. Something that a skill wouldn't be able to do.
12:45It'll just say no search volume here. Tough luck. This one actually did the search, and it's a bit more proactive
12:51if you will. It's even given me some quick wins, what I should do, some insights
12:56to market research, and told me where it got the sources, not only from the MCP call from the data for SEO that I've connected it to it, but even did some additional keyword, and I'd even have to ask it to do that. Now last in level three, we have plugins, and this is the ability to combine plugins, agents, MCPs, and any other requirements
13:16to do incredible tasks for you. There's also a plugin marketplace, and this is where things start getting a little bit complex.
13:23But I figured the best thing for me to do is show you what it can do. Inside our community, I've built a plugin for the community that builds an SEO ready website and even places it in Cloudflare so it's ready to host and go. Let me show you how this works.
13:37If I go to my Claude, I'm gonna go forward slash build website, and I've got a plugin here already. So I'm going to run this, and it's gonna ask me a bunch of questions because that's what I've set the plugin to do.
13:48It's gonna ask what is this website about, how many pages it does it need, and all of the requirements for you to build this. This plugin is gonna ask you a total of seven questions. You can see the first one here, share your business details, name, tagline, phone number, email address, and service area.
14:03Now I'm gonna skip through this and show you what the end design of a website with this plugin can look like. And if you use the plugin and answer all the questions with the most detail as possible, you might get something like this. And I've done very, very little work here.
14:18I've just let it run, and it built the entire thing for me, even put it on Cloudflare and add the right schema. It even has little animations here.
14:26I've had to pick the color layaway, but it's kind of done everything. Let's go industries or services. So we've got GPT optimization,
14:34animations. And inside this plugin, for example, I have given it access to the Nano Banana Pro skill, which means it can grab images of me and put me in offices like this to generate the images as possible. I've done a detailed video tutorial on how to do this.
14:49I'll leave a link below, but I wanna show you the incredible use cases that a plug in can have. Number four is system integration, really MCPs.
14:57This is where Claude code stops just living in your system and starts interacting with the real world. MCP stands for model context protocol, and you don't really have to know how it works. You just have to know that through this, ClaudeCode can access and interact with practically anything.
15:12And you saw a little bit of it just before where we asked ClaudeCode to interact with the data for SEO MCP, which allows it to call real world SEO data. It's great, but a little bit boring.
15:24With MCPs, you can do pretty incredible things like, for example, what if Claude Code can not only write your blog post, but upload it in your website for you, do your on-site SEO for you, and even interact with the tools that you currently use on a day to day basis, like Notion, maybe even Slack or Canva or whatever it is that you have.
15:42Anthropic have a list of popular MCPs that it recommends you install. You can see here Canva, Figma, monday.com,
15:50Gamma, one and of my personal favorites, Webflow, because that is what I've built my website on. Let me show you the incredible things that you can do once you've installed the MCP. And by the way, connecting these MCPs is very, very easy.
16:01Anthropic even give you the command that you need to give it. For example, you just give Claude this command, m c p add m c p web flow, and it'll install everything. You might have to authenticate it, and it'll just pop up your browser, and you'll have to approve it, but you only really need to approve it once.
16:18Anyway, let me show you how this works. If I go to my Claude code and type in forward slash m c p, you see here all the MCPs I've got connected.
16:26Some of them I need to kinda re authenticate, but here's the one that I'm really interested in is the Notion MCP. So let's test it out.
16:35I have a workflow currently that can take a YouTube video of mine, rewrite it in a style that I like with an agent, and put it into my Webflow website. So let's do just that. Here's a YouTube video that I recently generated.
16:47I'm gonna copy the URL here. And I'll in the meantime, I'm also gonna show you the latest blog post so you see that it's doing this itself. Last one that I published was about I built three agents that run my business while I sleep.
17:00Right? So I'm going to use the video to blog skill. So I'm going to say run the video to blog
17:09skill agent with this video.
17:15So let me show you how this might work. I'm gonna grab a link to this YouTube video, and I'm also gonna show you that the last blog was created on the March 4, it was about building three AI agents that run my business while I sleep. So I'm gonna say use the
17:30video to blog skill with this video. Boom.
17:37Now that skill already has access to the data for SEO MCP, and you can see it's loaded the skill. It's going to turn the transcript into a blog post that I like in my tone of voice, interlink sections of my website, and even upload it to my Webflow website.
17:53You can see it's running already. I've also got it on dangerously skip permissions, so it's not gonna ask me for anything. Now all we gotta do is go here and wait.
18:02K? After it worked for about four minutes, let's see here. That was the last blog.
18:06Let's refresh the website, and hopefully, should have another one right here. There's the Kimi code one that I asked it to write. Now think about what happened there.
18:15We asked it to scrape data from a YouTube video. It then rewrote that in a way that I wanted to, which was in a bit of a content capsule technique.
18:26Let's have a bit of a look at the blog post. So it's got a too long didn't read section. I wanted to do that.
18:32It's internally linking to other sections of my website, which is just really important for SEO, and it's even externally linking to other parts of the Internet backing up the statements here.
18:44It's even got an FAQ section at the bottom and all the resources where it got it from. So it's doing so many things there at once with one single call, and that is the power of connecting all this with an MCP.
18:56So Claude just went from a tool that lives in your terminal to a tool that interacts with the real world with any system that you use. But here's the one problem that we still have. We had to tell it what to do.
19:08We grabbed the link from the YouTube video and say, hey. Do your thing. Turn this to a blog post.
19:12But what if we didn't have to tell it to do that? What if Cloches ran things on its own? Well, welcome to level five.
19:19Level five is about autonomous workflows. This is where you stop using Claude code as a tool and you start using it as a autonomous system that runs by itself. And everything that we've worked on so far comes together.
19:32The memory from level two, the agents, the plugins, and the skills from level three, and the MCPs from level four combined together to create this incredible tool that will honestly 10 x your work and almost run certain parts of your businesses completely autonomously. And it looks like this on the screen. Terribly boring, but this has changed the way that I work with Claude and my work in general.
19:54For example, remember the agent that took the video, the YouTube video, and turned it into a blog post? Well, that guy actually runs automatically
20:02every Tuesday and Thursday at 9AM. How do I do this? Well, this is possible with two main things.
20:08One is the ability to run Claude in what's called headless mode. And essentially what this means is that Claude can just run without you interacting it with it in a terminal, on the application, anything. It just runs in the back end system without you even realizing.
20:23And two is about cron jobs. This is essentially you being able to schedule certain tasks. Like the video to blog post agent, we tell it to run that as a Cron job every Tuesday and Thursday at 9AM.
20:37And it's literally that easy. The agent that you run once, you say within that same conversation, hey. I want you to run this as a cron job every and you pick the time.
20:47But there is one very, very important component that you need to set up in order for this whole thing to work, and that's providing the cron jobs with certain permissions. You want to go into your system settings and open this up.
20:58Next, we want to go to privacy and security, and we want to go to full disk access. Right?
21:05You can see that I've got the cron tab here enabled, but you won't have that. And the next thing you need to do is press the plus bar. And without doing anything else, it's very important that you press command shift g.
21:15Let's do that now. Command shift g. And now you should see a little search bar here.
21:20In this search bar, you're gonna type in this exact thing, and I'll leave it in the video description below. It's forward slash USRForward slash S BINForwardSlashCRON.
21:33And that allows you to add this tab that you won't have to the full system access. So I've already done that. I'm not gonna do that again.
21:40I don't wanna mess with the system, but double click on that, and you will see this cron tab enabled. The last thing you wanna do is make sure that it's got a little blue sign there, making sure that it's enabled, and that is it. Now you won't run into any issues with the cron jobs.
21:52Now I hope this has caused a bit of a shift in the way that you think about using Claude code. Yes. You can use it as a tool, but you can also use it as a autonomous
22:01system. And if you learn how to master it as an autonomous system, you'll be ahead of 99 of people out there. But before you go and do this whole thing, one of the last things that I want to leave you with is think about creating yourself a bit of a command center, a dashboard, because it's very difficult to see what has worked and what hasn't when you're running things in a headless manner through cron jobs.
22:23So I've built myself my own custom command center where I can see how many sessions I'm running in the terminal, and I can also go to the agents. And I can see the cron jobs that are scheduled from which project, the last time they run, and their status. Now I will leave a prompt in the video description below of what I used to create this dashboard.
22:44CTABut because the way that you use Claude will be different from the way that I use Claude, I think you need to create your own dashboard that meets your needs and therefore build it yourself, but I'll leave the instructions below. And if you wanna see how these five levels can completely automate your marketing, getting you more leads and sales, and even seen in the AI search engines,
23:05CTAI recommend you check out our community where we have all of these workflows and more available to you that are constantly updated, and we provide a lot of support through live q and a's, an onboarding session with myself, and continuously updated tools, skills, and more so you learn to be more effective and more efficient with these AI tools.
23:22CTAIf you found value in this video, make sure you hit the like button and subscribe for more. Thanks for
— full transcript
§ 05 · For Joe
Five steps from chatbot to autonomous business system.
WHAT TO LEARN
Claude Code is not one tool -- it is five progressively more powerful configurations, and most users stop at the first one.
- A CLAUDE.md file eliminates session amnesia: create one with /init and Claude will understand your project architecture, preferences, and available tools every time it opens.
- Skills are reusable instruction sets you invoke by name -- the same task with the right skill produces measurably better output than a raw prompt, as the side-by-side landing page demo shows.
- Agents differ from skills because they can reason between steps and change course: when a keyword has zero search volume, an agent looks for related keywords rather than stopping.
- MCPs turn Claude from a text generator into an actor -- connecting it to Webflow, Notion, or Canva means it can publish, update, and manage content in external systems without you touching a browser.
- Headless mode plus cron jobs means a workflow you tested manually can be scheduled to run at 9AM every Tuesday without any terminal window open or human present.
- macOS blocks cron jobs from file system access by default -- granting /usr/sbin/cron full disk access in System Settings is a required one-time fix that most tutorials omit.
- Once you run headless agents on a schedule, a monitoring dashboard is not optional -- you need somewhere to see what ran, what succeeded, and what failed.
- The dangerously-skip-permissions flag is what makes long autonomous runs practical: without it, every file write stops and waits for manual approval.
