The argument in one line.
Claude Code is a tool that lets non-technical business owners automate tasks by having AI take actions on their behalf, and these 28 core concepts—from prompts and context to MCP servers and browser automation—are the building blocks needed to deploy it effectively.
Read if. Skip if.
READ IF YOU ARE…
- A non-technical business owner running on automation who wants to understand what Claude Code can actually do without learning to code.
- Someone building multiple client projects or internal tools who needs to organize Claude Code workflows across different contexts.
- A founder or operator who uses ChatGPT but is confused about what Claude Code does differently and when to reach for it.
- A business person exploring no-code automation who wants a reference walkthrough of all major Claude Code concepts in one sitting.
SKIP IF…
- You already build software professionally or have deep experience with AI agents — this is intro-level framing.
- You're looking for hands-on tutorials on specific outcomes like building a particular app or workflow — this is concepts only, not implementation.
The full version, fast.
Claude Code is an AI coding assistant that takes actions on your behalf rather than just chatting, letting non-technical business owners build websites, apps, and automations through plain-English prompts inside a free coding workspace like VS Code or Antigravity. The lesson walks through 28 building blocks � prompts, context files, the claude.md instruction file, models, tokens, slash commands, memory, compacting, skills, hooks, MCP servers, APIs, subagents, parallel agent teams, plugins, browser automation via Playwright, extended thinking, checkpoints, GitHub backups, scheduled tasks, and the terminal � each framed as a tool you assemble into custom workflows. The practical move is to give Claude rich context, build reusable skills for anything you do twice, connect external apps through MCP, and reset conversations before tokens balloon.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:53
01 · What is Claude Code
AI that takes actions, not just talks. Chatbots tell you, Claude Code does it.
00:53 – 01:58
02 · IDEs and Coding Workspaces
VS Code, Antigravity (Google), Cursor. Pick one, stick with it. Claude lives inside as an extension.
01:58 – 03:06
03 · Projects and Files
File system is Google Drive for your project. Multiple clients = multiple projects in one workspace.
03:06 – 04:21
04 · Prompts
Be specific, give it a role. Generic input = AI slop. Specific input = results that actually fit.
04:21 – 06:19
05 · Context
Feed Claude your tone, humor, and vocab files. More context = more personal, more accurate outputs.
06:19 – 07:00
06 · References
Use @ symbol to pull specific files into the conversation mid-prompt. No copy-paste needed.
07:00 – 08:49
07 · Permissions
Five modes from Ask through Bypass. Start with Plan Mode on big builds, loosen after.
08:49 – 09:57
08 · Tools
Read, Glob, Grep, Write, Edit, Bash, Web Search. The hands of Claude. Zero config needed.
09:57 – 10:49
09 · Context Window
Short-term memory. Fills as conversation grows, quality drops past 50%. Watch the bar.
10:49 – 11:51
10 · CLAUDE.md
Your instruction file. Train Claude like a new employee with rules, preferences, and must-nots.
11:51 – 12:59
11 · Models
Haiku for fast/cheap tasks, Sonnet as daily driver, Opus for hard problems. Match brain to task.
12:59 – 15:23
12 · Tokens
Three-quarters of a word each. Costs compound exponentially as conversation history grows.
15:23 – 16:32
13 · Slash Commands
/context to check usage, /clear to reset. Prevents burning your weekly session limit mid-project.
16:32 – 17:45
14 · Memory
Persists across all conversations. Store facts about yourself and your business once, reference everywhere.
17:45 – 18:35
15 · Compacting
/compact at 50% context saves money and prevents quality drift that kills long sessions.
18:35 – 20:04
16 · Skills
Pre-written instruction sets for repeatable tasks. On-demand workflows triggered by a slash command.
20:04 – 21:36
17 · Hooks
Automated actions at conversation lifecycle points. Guard rails, triggers, and end-of-turn sounds.
21:36 – 22:30
18 · MCP Servers
Connect Claude to Airtable, Slack, Gmail, Notion. Install once, use forever.
22:30 – 24:14
19 · APIs
Backup when MCP does not exist. Store API keys in .env file for security on deployment.
24:14 – 25:51
20 · Sub-Agents
Pre-configured specialists. You give the order, they return the result. You do not see the process.
25:51 – 27:29
21 · Agent Teams
Manager agent delegates to employee agents in parallel. Faster, better context separation, better results.
27:29 – 28:11
22 · Plugins
Skills built by others, install in one click via the /plugins marketplace.
28:11 – 29:08
23 · Browser Automation
Playwright plugin lets Claude control the browser like a human. Logs in, downloads, uploads, automatically.
29:08 – 29:59
24 · Extended Thinking
Slow mode: Claude defines the problem, considers approaches, picks the best, then answers.
29:59 – 30:38
25 · Checkpoints
The undo button. Rewind code to any previous conversation state with one click.
30:38 – 32:19
26 · GitHub
Google Drive for code. Backup your project and the deployment gateway Vercel pulls from.
32:19 – 33:32
27 · Scheduled Tasks
Cron jobs. Claude as employee not assistant. Runs on any interval without you present.
33:32 – 35:06
28 · Terminal
The black box from the 1960s. Rarely needed because Claude can usually do it for you via the IDE.
Lines worth screenshotting.
- Claude Code takes actions on your behalf rather than just talking to you — it builds files, runs code, browses the web, and makes changes autonomously on your computer.
- An IDE is just a folder system where Claude Code works — you do not need to write any code to use one, and VS Code and Anti Gravity are both free options.
- Giving Claude a role in your prompt — 'you are a web designer' or 'you are an expert investor' — tailors every response toward the specific expertise that matches your task.
- Context is the difference between AI slop and outputs you are genuinely happy with — the more Claude knows about you, your business, and your goals, the better it performs.
- Feeding Claude a sample of your writing and asking it to extract your humor, tone, and vocabulary into a dedicated file creates a persistent voice guide that carries across all future outputs.
- The @ symbol in a Claude Code prompt references a specific file directly, which is critical when multiple files have similar names and you need Claude to read exactly one of them.
- Permission modes — from ask-before-everything to full autonomy — control how much trust you give Claude to make decisions without checking in, with more autonomy meaning faster but less supervised output.
- Multiple projects can live in the same workspace so that switching between clients or contexts requires only selecting a different folder rather than closing and reopening anything.
- A one-page website built with Claude Code by a non-coder in the same empty folder produces the same artifact that Lovable produces inside a virtual machine — the only difference is who controls the infrastructure.
- CLAUDE.md is the instruction file Claude reads every time it enters a folder — writing your goals, constraints, and preferences here means you never have to re-explain the context.
- Specific prompts with a named location, a named audience, and a named goal produce dramatically better outputs than vague prompts like 'build me a website.'
- The most important Claude Code concept for non-coders is that every file type — HTML, JavaScript, markdown, CSV — is something Claude can create, edit, and organize on your behalf.
Twenty-Eight Claude Code Concepts in Plain Language — the Complete Reference for Non-Technical Builders
Claude Code reference
Jono Catliff's 36-minute walkthrough treats every Claude Code concept as infrastructure a business owner needs to understand, not features a developer needs to implement — and the plain-language framing makes the most counterintuitive concepts (context window, compaction, sub-agents) immediately usable.
01What is Claude Code
- Chatbots tell you what to do; Claude Code does it — the action capability is what separates an agent from a conversational assistant
- The framing for business owners: it is not a development tool, it is an execution layer
07Permissions
- Five modes from Ask through Bypass — start with Plan Mode on any build you care about, loosen permissions incrementally as you verify the agent's judgment
- Plan Mode forces a review step before any action — the most important permission setting for business owners unfamiliar with what the agent will do
09Context Window
- Short-term memory that fills as the conversation grows — quality degrades past 50%, making compaction a maintenance habit, not an occasional fix
- Watch the bar; run /compact at 50% to preserve quality and prevent the drift that kills long sessions
10CLAUDE.md
- The instruction file that trains the agent like a new employee — rules, preferences, business context, explicit must-nots
- Well-written CLAUDE.md is the single highest-leverage configuration step for consistent output quality
20Agent Teams
- Manager agent delegates to employee agents running in parallel — faster execution and better context separation than sequential single-agent work
- The parallel execution is what makes agent teams qualitatively faster, not just incrementally faster
27Scheduled Tasks
- Cron jobs for Claude — runs on any interval without you present, which is the distinction between AI that helps you work and AI that works while you sleep
- The scheduled task pattern is what converts Claude from an assistant to an autonomous team member with assigned recurring responsibilities
Terms worth knowing.
- Claude Code
- Anthropic's command-line coding tool that can read, write, and edit files on your computer and take actions on your behalf, rather than just chatting like a typical AI assistant.
- IDE (Integrated Development Environment)
- A desktop application where code lives and gets edited, with a file explorer, editor, and terminal in one window. Examples include VS Code, Cursor, and Google's Antigravity.
- VS Code
- Microsoft's free code editor, the most widely used IDE. Claude Code can run as an extension inside it.
- Antigravity
- Google's free IDE, a coding workspace similar to VS Code where Claude Code can be installed as an extension.
- Cursor
- A paid IDE built specifically around AI-assisted coding, competing with VS Code and Antigravity.
- File system
- The folders and files that make up a project on your computer. In an IDE, the file system is shown in a sidebar so you can navigate and organize your work.
- Workspace
- A container inside an IDE that can hold multiple projects at once, letting you switch between them without opening separate windows.
- Prompt
- Any message you send to an AI tool. The clearer and more specific the prompt, including any role you assign the AI, the better the result tends to be.
- Context
- The background information you give an AI about you, your business, and what you want, so its answers feel tailored rather than generic.
- .md file
- A plain-text Markdown file, commonly used to store instructions, notes, or configuration that both humans and AI tools can read.
- @ reference
- A shortcut inside Claude Code where typing the @ symbol lets you point Claude at a specific file by name, so it reads exactly the file you mean.
- Permission modes
- Settings in Claude Code that control how much Claude can do on its own. Modes range from asking before every edit, to accepting edits automatically, to fully bypassing prompts.
- Plan mode
- A Claude Code mode where the AI drafts a step-by-step plan first and waits for your approval before making any changes.
- Tools (in AI agents)
- The built-in capabilities an AI agent can use to act on the world, such as reading files, editing files, searching the web, or running commands.
- Bash command
- A text command run in a terminal to do things like create files, run programs, or move data. Claude Code can run these on your behalf.
- Context window
- The amount of text an AI can keep in active memory during one conversation. Once it fills up, older messages get dropped or summarized and quality drops.
- CLAUDE.md
- A special instruction file Claude Code reads automatically at the start of a project. You use it to teach Claude how to behave, what rules to follow, and what to avoid.
- Model (AI model)
- A specific version of the underlying AI brain. Claude offers Haiku (fast and cheap), Sonnet (balanced), and Opus (most powerful but most expensive).
- Token
- The smallest unit AI models read and bill against, roughly three-quarters of a word. Every message in and out counts toward your usage limits.
- Input vs output tokens
- Input tokens are everything the AI has to read, including the whole running conversation. Output tokens are the words it writes back. Long chats balloon input tokens fast.
Things they pointed at.
Lines you could clip.
00:26
“ChatGPT talks to you, but Claude Code actually takes actions on your behalf.”
03:37
“Instead of saying hey build me a website, which is gonna give you generic AI slop, you can say build me a one-page website for a wedding DJ business in Toronto.”
10:59
“You hire an employee, and you need to train that employee on what the job is. The CLAUDE.md file is that instruction file.”
14:22
“That is why it is an exponential curve because the longer your conversations get, the more the tokens and memory expand.”
25:11
“With a sub agent, you order food, the kitchen delivers it. With skills, the chef cooks it on the table in front of you.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
analogy
00:00Cloud Code is difficult to learn because there's just so many different concepts that you need to learn. And it feels like every single day that goes by, there's a new concept that you have to understand.
00:11In today's video, we're gonna be breaking this down in the simplest way possible. I'm gonna be teaching this like I was explaining it to a 10 year old. Let's get into it right away.
00:20The first thing is is what actually is Claude code and what separates it from something like ChatGPT? Well, ChatGPT talks to you, but Claude code actually takes actions on your behalf.
00:32So you can message Claude saying, hey. Build me out a website, and it will literally build out the entire website for you. This whole thing was generated using Claude code for me.
00:42You can also generate blog posts for you. You can even generate web applications like a dashboard that runs your entire business. And all of these slides that I'm showing you was also generated using Claude Code.
00:55The next thing is is IDEs and coding workspaces. Now, technically, this isn't Claude Code, but this is almost a requirement for most people to use Claude Code. So I wanted to include it in there because it's one of the first things you're gonna hit when you go to use Claude Code.
01:10IDE stands for integrated development environment, and you do not need to know any software engineering or anything technical whatsoever. That's the whole reason you use Claude so that you don't have to do anything technical. There's three main IDEs or coding workspaces in the marketplace.
01:26Versus code, anti gravity, both of those are free, and you have Cursor, which is also paid. Now, essentially, you can head over to a site like antigravity.google and download the free desktop.
01:38And again, this is a Google product. And once you're inside, you'll you'll be in a coding workspace just like this. Now where Claude code comes into the picture is it's an extension that lives inside of your coding workspace.
01:51So on the left hand sidebar here, you can add plug ins, search for Claude code, and then start using it immediately from within Versus Code. Now the next thing is projects and files.
02:05Anytime you're building on a project with Claude code, you have something called a file system. System.
02:11K? You can think about this really just like Google Drive, but for your project. You have you can store whatever different files you want.
02:18You could have videos. You could have images. You could have text files.
02:21You can have folders to organize everything. But the cool thing is is that you can have multiple different projects inside of your coding workspace.
02:32So right now, I technically have one project here called Claude Code Concepts. Okay? And then I have another project called project two.
02:41We could go ahead and hit file, add to workspace, and then I can add project three over here. So let's say you have multiple clients.
02:49You could have a new project for every client from within the same, um, workspace so that you can move back and forth between all of them just by messaging Claude and keeping things nice and organized.
03:02And, of course, again, you can add any files that you want to this project. The next thing is prompts. So prompts you're probably already familiar with, but anytime you're messaging back and forth with Claude, that is considered a prompt.
03:17So when you open up Claude by hitting this nice little Claude button right over here, and let's say we say that we send a message, hey, that is a prompt, and then Claude code is going to reply back to us. Now the one key about prompting inside of Claude code is that you want to be as specific as possible, and you want to give Claude a role.
03:41So instead of saying, hey, build me a website, which is gonna give you generic AI slop. You can say, build me a one page land one one page website for a wedding DJ business in Toronto. It's lead generation focused and so on and so forth.
03:56So the more information you provide with, the better it will be delivering you the outcome you're looking for. When I say give it a role, in this instance, you'd say, hey, Claude. You're a web designer, and it will act according to that position.
04:10Again, maybe you wanted help with investments. You could say, hey. You're an expert investor or you're an expert social media manager or whatever the case may be, and it's going to understand exactly how it should tailor its responses to you.
04:24The next thing is context. Now when you're using Claude code, the distinction between getting bad responses back like AI slop and fantastic responses back that you're really happy with is the context you provide Claude code with.
04:38The more it knows about you, your business, what your outcomes the outcomes are that you're looking for, the better it's gonna be at delivering you results that you're happy with. So let's say, for example, you wanna generate something like a LinkedIn post over here.
04:53If I just tell Claude, hey. Generate me a LinkedIn post. It's gonna go ahead and do a really bad job.
04:58It's gonna throw in a bunch of emojis that I don't want. It's gonna sound generic. It's not gonna have personality.
05:03It's not gonna sound like me whatsoever. But I could add in here files on my humor, on my tone, on my vocabulary, and then it can actually go ahead and understand what kind of jokes I like to make, how I sound in real life, and then it can actually write posts that sound exactly like me.
05:24Okay? So in here, just as a side note, Jono's humor is dry, self aware, and observational. It gives examples of what that looks like.
05:32I I swear, Claude knows my humor better than I know myself. And so these are kind of examples on how you can get the right results back from ThoughtCode. How you'd actually pull in things like humor and tone and vocabulary is you would go to something like your LinkedIn posts or your emails that you've sent off or wherever you have samples of writing.
05:56It could be call transcripts as well. It could be video transcripts. You will just take that, okay, that whole post, and then you would just paste it into Claude and be like, here is a sample LinkedIn post.
06:07Can you please pull up my humor, tone, and vocabulary and update all three files that you have on record? And now it's gonna go ahead, pull out all of the information from this, and then update all of these different files.
06:21Cool. The next thing is is references. So when you're dealing in Claude code, projects can often grow pretty fast and get pretty big.
06:30If you wanna reference a particular file, like for example, this tone file over here, we can drop in the and symbol. Okay? And when we drop in the and symbol, I'm just gonna move this over and so you guys can see it properly.
06:43Now we can start referencing particular files. So I can type in here tone dot m d, and now it's going to be able to read that particular file because I referenced.
06:55So you can actually see that it's gone ahead and read the exact file that I was referencing. This is especially, uh, useful if you have multiple files with the same name or very similar names. The next thing is permissions inside of Claude code.
07:09So when you're communicating back and forth with Claude, you can determine how much control you actually give to Claude code, and that comes down to this button right over here. Here are all the different modes. Let me explain this in a very simple way, and we're going to remove plan mode for now, and we're just gonna talk about these other four.
07:29The further down this goes, okay, the more control that you give to Claude to make autonomous decisions on your behalf. So when you say, hey. Ask for edits.
07:39Anytime Claude does anything like editing a file or connecting into an external application, which we'll talk about later on, it's gonna ask for your permission. When we talk accepting edits, which is the next layer down, okay, what's going on here is that it's going to edit files in your file structure over here automatically.
08:01But anytime you're doing anything kind of, like, outside of editing these exact files, So for example, connecting into Gmail or connecting into external applications or whatever the case may be, it's gonna ask for your permission. With auto mode, it's going to only ask for your permission if it thinks that it's gonna do something dangerous.
08:20And with bypass permissions at the bottom here. This is where it's just gonna do everything on its behalf, and sometimes it can make mistakes, overwrite things that you didn't wanna overwrite, but it does go a lot faster.
08:31And with plan mode here, the distinction is is that it thinks now and then it acts later. It will create an entire plan for you. And once you approve that plan, you're happy and it thinks through all of the different contingencies, then it will go ahead and build it out.
08:46So anytime you're building a large project, it's usually advisable to start with plan mode and then switch over to one of the other four categories after the planning is done. The next thing is tools inside of Claude code, and this is the hands part of Claude. It allows Claude to actually take actions on your behalf.
09:07So when you look at this conversation here, you can see that Claude has read, okay, a particular file, and then it did something called web patch. So it actually went to this URL, and it read this documentation here, this this page.
09:21Okay? Then it went and it fetched this web result, and it read this page as well, understood what's going on here. And then it wrote this article right over here called techstack.md.
09:34So Claude is taking actions on your behalf, and these are tools that it has access to. These are out of the box. There's no configuration required, but it has the ability to read a file.
09:44It has the ability to find a file by its name, which is called glob grep, which is search inside of files. Then it can write and edit. It can do bash commands, which we'll talk about a bit later on.
09:56It can search for a URL and search on Google and so on and so forth. The next thing is the context window inside Claude code, and this is Claude's short term memory.
10:07So if we just open up any one of these individual conversations, everything going on here is the context window. Okay?
10:16The more messages that you send in here, the more Claude has to remember. So Claude has a limit on how much it can remember just like a human would.
10:27So if you and I are having a conversation for ten years, you're not gonna remember all ten years. You might be able to remember a couple hours. Right?
10:34Same kinda deal with Claude. It's not gonna be able to remember what you said, like, ten years ago. It's only gonna be able to understand a certain length.
10:42And then after that, it's gonna fill up and the quality drops. And then you have to reset the conversation, which we're gonna talk about later in this video.
10:52The next thing is the Claude dot m d file, which is the instructions you tell Claude to behave the way you want it to behave. A good analogy for this is you hire an employee, and you need to train that employee on what the job is and how to do it properly.
11:10The it's the same kinda deal with the claude.md file. This is the instruction file that you give to Claude to tell it how it should behave, what it should do, what it shouldn't do. So let's say that we're writing a LinkedIn post one more time.
11:22A rule that you have here is, hey, don't let me publish a LinkedIn post unless we qualify it as a nine out of 10. I don't wanna send off a five out of 10. I don't wanna send off a six out of 10.
11:34It needs to be at least nine out of 10. Another rule could be, hey, before you create a file, make sure it doesn't exist yet. We might not wanna like, so for example, you don't wanna duplicate your humor file or your tone file.
11:47You only wanna create that file if it already doesn't exist. Okay? So that's kind of the claude.md file.
11:54It just teaches Claude on how it should behave. The next thing is models inside of Claude code. So you typically have three different models, Haiku, Sonnet, and Opus.
12:05Inside of Claude code, I'm gonna move this over just so you can see it properly. If you hit the slash command right down here, you can switch between models and you have access to, again, Sonnet, Haiku, and the default over here, which is Opus 4.7.
12:21The higher up you go, the better the models become. Opus is definitely the most powerful. Haiku is the least powerful, and Sonnet is somewhere in the middle.
12:31The distinction between them though is Haiku is fast and cheap, which means that, um, you're not gonna be using as much of your Claude limits while using Haiku. Okay?
12:44So every time you're using Claude, we're gonna talk about this a bit later, you have current session limits, weekly limits, model limits, all of that kind of stuff. Opus is gonna burn through it very quickly. Haiku is gonna burn through it way more slowly, and then Sonnet is somewhere in the middle.
12:58Obviously, there's trade offs between cost and power. The next thing is tokens, which is essentially the currency that Claude code uses.
13:08So when we go back to this usage page over here, you have certain limits on a session basis, on a weekly basis, across all models, across claw design, and so on and so forth. And the currency used to fill up that limit is tokens.
13:27Okay? And they can add up quite fast. The definition of a token is approximately three quarters of a word.
13:34So let's take this phrase for example here. Build me a landing page for a wedding DJ. Okay?
13:39This takes up 10 tokens. Every token is essentially represented by a different color. So we have build over here.
13:47That is one token. Then we use another token for the word me. Then we use another token for the word a, then landing page landing, then page, and so on and so forth.
13:57So these are all tokens that we're using when sending off a message. Now you might think, okay.
14:03Well, 10 tokens, not that bad. It's not a big deal. But the issue is actually less about the individual message you're sending.
14:10And the bigger issue is that anytime you're interacting with Claude code down here, okay, there's input and output tokens. And what that means is that the input tokens are the entire conversation. So if you've just had a monster two hour conversation here where you're going back and forth with Claude, that whole conversation is technically input tokens that you're still burning through.
14:38And when it outputs a response back, that's also tokens as well. So every individual message by itself, it's really not that expensive.
14:46But when you get to, like, 200 messages, all of a sudden, you have to remember not only that individual message over here, but then the last 200 messages as tokens as well.
14:59And so that's why it's an exponential curve because the longer your conversations get, the more the tokens and memory expand. And all of a sudden, when you're at the very end, it can balloon really quickly, and that's the danger zone where your tokens or credits can all of a sudden explode.
15:17And then before you know it, you're just out of that current session or your weekly limits or whatever the case may be. The solution here is to reset the conversation. Cool.
15:27And how we can do that and many other things is through slash commands. Okay? So inside Cloud Code, if we type in slash here, k, we have access to a lot of different commands.
15:41One of those commands could be saying context. And I feel like I'm screaming at Claude code by typing in all caps here.
15:48But when we type in context, it will tell us how much of the current conversation, how much of the context we've used.
15:58Remember, it only has a predefined set, um, memory.
16:03So it can't remember ten years of messages. It can only remember, let's say, one or two hours. And this will tell you how full that memory is.
16:12And the higher up this goes, the more you're spending. Another option is to type slash clear. And what this will do is it will reset the conversation so that you're not burning money on credits.
16:24This is really important just because a lot of people burn their credits, and then they're locked out for, like, a week. And, obviously, you don't want that to happen to you. Okay?
16:32There's a lot of other slash commands that you can use as well, but those are some of the main ones that I like to use. We're gonna be talking more about slash commands later on. The next thing is memory inside inside of of Cloud Cloud Code.
16:43Code. So So let let me me explain this as clearly as possible. You can see up top here, we have many different conversations open at the same time.
16:53Okay? We have four that you can see right here. The problem is is that these conversations are isolated.
17:01So if I say something in conversation a, conversation conversation c won't know what I said.
17:08It won't persist across the different conversations. So you're starting from the ground up with every single conversation you have. With memory, this is a way that you can save information that will persist across every single one of these conversations.
17:24Okay? So as an example here, I can type in something like update the system memory with my favorite food, which is sushi.
17:34And then all of a sudden, in the next chat window, I can say, hey. What's my favorite food? And it will remember that it's sushi.
17:41Okay? So you can remember things about yourself, about your business, about the way you like to write, or whatever the case may be, and it will persist across every single conversation that you have with Cloud Code. Now the next concept is called compacting, and this helps your longer conversations stay sharper.
17:59This is a slash command. And as we were talking about earlier, every single chat has a context window.
18:06When it hits a 100%, automatically, Claude is going to compress that conversation.
18:11How it does that is it just takes the whole conversation and summarizes it so it resets essentially your context window. Now you can wait for Claude to do this, or you can do it manually by typing in slash compact, and then it will compress the whole conversation down.
18:28The benefit of doing this manually is that when your context gets to about 50% full, Claude actually starts forgetting things, and, also, you start spending way more money through tokens. The next thing is skills inside a Claude code, and this is the thing that I use the most inside Claude.
18:44They're prewritten instructions to get tasks done. Now one example could be creating LinkedIn posts, but the reality is is that anything you do two or more times, you can turn into a skill and have Claude code automate it.
18:57I like to actually think about it like on demand workflows. I try and create a LinkedIn post every single day. Claude code writes the first draft for me, and it comes out very good because I've told Claude what a good post looks like, and then it can replicate it every single time.
19:12So now I just come into Claude and I say, hey. Write me a a LinkedIn post. It will pull in things like my tone and my humor and my voice and my, uh, good examples of what a a LinkedIn post looks like and everything else, and then it will create the post for me.
19:27Here's what it actually looks like inside of Claude code. When I say on demand, I literally mean that I can type in a slash command here, oops, inside Claude, like slash LinkedIn.
19:39And on demand, I can call Claude to create a LinkedIn post. Now these skills can easily be created in Claude.
19:48You can open up a new conversation and say, please create a skill for posting on LinkedIn, and it will go ahead and do that.
19:56And as you use these skills over time, it's not gonna be perfect out of the box. You just incrementally improve it until it becomes what want it to be. The next topic inside Cloud Code is called hooks, and these are automated actions that you can have Cloud Code do at some stage during the conversation.
20:15So if you think about, like, a traditional conversation in Cloud Code, there's multiple stages. You start the conversation, and you end the conversation, and then certain things happen in between.
20:25Like, you send a message, Claude sends a message back, and so on and so forth. You can set a task or an action to happen at any of those stages. So for example, I could say, if I say the word pizza at the beginning of conversation, don't reply back, block the request.
20:42So in here, I'm gonna type in pizza, and the conversation is just going to stop. Nothing's gonna happen. It's not gonna reply at all.
20:50And then I can say, hey, and the conversation will return, uh, back just as normal. And then I can say pizza again, and nothing's gonna happen, and so on and so forth.
21:00Now a better use case for this would be don't upload my information online if you think I'm exposing passwords or secret keys that I don't wanna expose. Okay? So that's another example.
21:12To finish up on hooks, we can also have something at the end of the conversation when Claude replies to us. So we could say, make a noise. Okay?
21:20And that noise that you just heard was a hook. You can come into Claude and say, please create a hook for me, k, that makes a sound every time you reply.
21:33And it will go ahead and build that out for you. It's really as simple as that. The next thing is MCP servers inside Cloud Code.
21:41MCP stands for model context protocol. It's how you connect Cloud Code to all of your favorite applications like Airtable, Slack, Gmail, Google Drive, Notion, and so on and so forth.
21:54How you set this up is you'd go to claud.ai/settings. You'd come over to connectors here, and you can browse through hundreds of different apps to connect into Claude code.
22:05Once it's connected in, you're good to go. You can start using it. You can also hit the configure here.
22:10Okay? And you can change the permissions Claude has. So if you don't want Claude to be able to write or delete emails, you could say needs approval or blocked, and it won't be able to take those actions automatically.
22:24So just as an example here, we can connect into Gmail, and every email that comes through, we can automatically label it as miscellaneous or personal or website or whatever the case may be.
22:37The next thing is APIs, which stands for application programming interfaces, and this is how you can also connect Claude into all of your favorite applications.
22:46It's essentially the same result as MCP. The distinction is is that MCP is relatively new. Like, it didn't exist five years ago, and it allows specifically AI agents to easily connect into your favorite applications.
23:03The problem is is that MCP is new, and there's not, like, uh, there's not every single application has a connection into Claude code. But you can still connect Claude into all of your other favorite applications even if an MCP server doesn't exist through APIs. Now practically speaking, all you need to do is tell Claude to connect in an application.
23:26Ideally, you'd start with MCP, and only if MCP doesn't exist, then as a backup, you'd use an API key. Okay?
23:33This is how applications communicated back and forth with each other for decades before MCP servers actually existed. The only thing that you need is a password to verify who you say you are.
23:48So let's say I log in to Gmail here. It's gonna ask me for a password to verify who I am. And very similarly, to connect Claude via an API into an external application, you need a password in order to verify who you say you are.
24:03And you wanna store that password in something called the dot e m v file. This is where all of your secret keys belong like this, for example. The reason why you wanna store it here is for safety reasons so that when you deploy your project live on the Internet, you don't have security vulnerabilities.
24:22The next concept is sub agents inside of Cloud Code, and these are preconfigured specialists that you can reuse.
24:29They're very similar in concept to skills.
24:34The distinction between the two, I'm gonna use an analogy here, is think about you heading to a restaurant. Okay? With a sub agent, you can kinda think about you ordering food, that food being completed in the kitchen, and then returned back to you.
24:48You gave the order. You got the food back, but everything in the middle, you didn't really see happen. K?
24:55With skills, it's like you order the food, the chef cooks it on the table in front of you. You see the whole process. You understand everything going on.
25:03You can interdoc, ask them to do certain things, and then you get the food in front of you. Okay? Now the benefit in doing this is that it's self contained, and it should require less context in that particular conversation.
25:17The benefit of skills is that it's in your chat, you get to watch every single step. So in this instance, I could say, read this particular agent. Okay?
25:25And we can see our agents in this folder right over here. And give me an overview of this project. It's going to find that particular agent.
25:34It's going to do the job, and then it's going to return the results back to us. Again, this file structure over here is just boilerplate. You don't have to memorize any of this.
25:45You just ask Claude to build you an agent, and it's going to automatically build out this file structure for you. The next thing is agent teams inside of Claude code. The traditional way that Claude works is you might send a prompt like build me a five page website.
26:03Okay? And it'll receive that task, and it'll build the first page, then it'll build the second page, then it will build the third page.
26:12And this is sequential. You're not gonna start this third page until the second page is done, and you're not gonna start the second page until the first page is done. What that means is that it's very slow.
26:20With agent teams, what's happening is you have a manager agent that delegates responsibilities to employee agents.
26:29Okay? So what happens is if you're building a website page that has five pages, one agent or employee agent gets delegated the first page, another agent gets delegated the second page, and the third agent gets delegated the third page, and so on and so forth.
26:45What this means is that it's faster because it's not sequential. You're not waiting for each step be completed. It's doing all of the steps at the same time.
26:53It separates context, meaning you get better results as well. So that it's not one agent doing everything.
27:00It's three agents that specialize in a particular role doing that role. So you just you get better results in general. Now inside of Claude, this is more or less what it looks like.
27:12You can say spin up parallel agents. K? And just it works out of the box.
27:17Like, you don't to preconfigure anything or install anything. You just have to type something like build parallel agents. And you can see here, Claude has delegated responsibility to three separate agents, one for the home page, about page, and contact page.
27:32The next thing is plug ins inside of Cloud Code, and these are skills that you don't have to build yourself. So if you recall, we were talking about skills like generating LinkedIn posts. The problem with that is that you have to actually develop the skill yourself.
27:47You have to tell it what you want and also what you don't want. With a plug in, somebody else built it for you. Okay?
27:54They perfected it, and then you are essentially using their finished product. Okay? So inside Cloud Code, you can type in slash plugins, and you have access to hundreds of different plugins that other people have built, um, that you can use out of the box.
28:12A good example is this front end design plugin that allows you out of the box to build beautiful websites immediately. The next concept is browser automation, which allows Claude to control your Internet or your browser just like a human would.
28:27So I'm actually not touching my computer right now, but you can see Claude has taken control of my computer. It's logging into an application automatically, then it's downloading my invoices, and it's gonna upload that into my accounting software.
28:40It does this once a month for me so that I don't have to do this kind of stuff myself. Now this is actually different than Claude just searching a website for you. It's actually taking actions in the browser on your behalf.
28:54How you can do this is by heading over to the plug in store again and downloading Playwright.
29:01K? And what Playwright is is it's a type of code that specializes in browser automation.
29:07Once you have this installed, you can just tell Claude what you want it to do. It will take control of your Internet and do the task automatically for you. The next concept is extended thinking inside of Claude code.
29:19This gives Claude the ability to reason before acting. So you have two modes, fast mode extended thinking.
29:26In fast mode, you might get an answer back really, really quickly, but the answer might not be as well thought out as if you gave it the time to think about how it should best answer your question. With with when you have extended thinking enabled, it's gonna define the problem, consider approaches, pick the best, validate, and then give you a well thought out answer.
29:49In order to turn this on, you can hit the slash command down here, and you can enable, um, thinking right over here by toggling it on.
29:59The next concept inside Cloud Code is checkpoints, and this is a powerful one in case you ever make a mistake. It's essentially the equivalent of the undo button.
30:09Okay? And what it allows you to do is, yeah, just rewind the code to a previous state. So as you're having a conversation with Claude, okay, at some point in time, it's inevitable that mistakes are going to happen.
30:22And when mistakes do happen, you have this button right over here, this backwards arrow on every single message. You can move up the conversation until you find the right rewind section, click it, and then rewind the codes at that point in time, and then it's gonna undo all of the mistakes that happened. The next concept is GitHub, and this is not technically Cloud Code, but this is something that everyone should know how to use if you're actually using Cloud Code.
30:50You can think about it like Google Drive for code. You can upload your code online to have it safely stored somewhere. And there's two major benefits for this.
31:01The first one is is if you're deploying, like, a web application online, chances are you're gonna need to deploy that code to GitHub first. And then from GitHub, you would deploy it online through an application like Vercel, for example.
31:14Vercel just allows you to publish websites online so anyone can view them. But the more important thing, in my opinion, is having a backup of all of your projects.
31:24Because the thing is, if you lose your computer, it gets stolen, it breaks, or whatever the case may be, and you just spent six months building up this project and then you lose that project, that would be absolutely devastating. Having GitHub means that you can upload your code safely as a backup, and then when it comes time to, uh, if anything happens, you can always just pull the code from GitHub back into your computer.
31:51Now how you do this is you would sign up for a free account on GitHub. You'd hit the top right corner, go to repositories here, click a new repository, and then name it whatever you want, and make sure it's set to private.
32:05Create the repository. It's gonna give you a snippet of code right here. You can just copy this into Cloud Code, and it should push or upload all of your code from your project into GitHub.
32:20The next concept is scheduled tasks inside of Claude code. So let me give you an example before we break this down.
32:28Let's say you want Claude code to run your entire email inbox. Every single time emails come in, you wanna automatically label those emails, whether it's miscellaneous, personal, website, you name it.
32:40You can have Claude at some interval, whether it's every minute, every fifteen minutes, every hour, every day, do a particular task for you.
32:49And this is a scheduled task. So what I'm telling Claude is every fifteen minutes, I want you to monitor my Gmail inbox and categorize everything.
32:58Create draft emails if necessary, forward emails, if it's an accounting email, and so on and so forth. So how we can do that is we can literally just come in to Claude, and we can say, schedule a task for every fifteen minutes.
33:12Okay? And what this will do is it will create a script on your computer that's technically outside of Claude.
33:21So it's on your computer, and every time interval that passes, whether it's fifteen minutes or hour, it will do the task that you have in mind, like sorting your email inbox.
33:32The next thing is the terminal inside Cloud Code. This is the last concept. This is, um, your traditional nineteen sixties terminal.
33:41I'm gonna actually close this and open the app one more time. So you can interact with Claude code completely from within this black box. All you need to do is type in Claude here.
33:53You need to trust the folder, and now you have access to Claude just in a different, um, in a different setting. So I can say, hey.
34:02It's gonna reply back to me. We have the slash commands over here. We can type in context, for example.
34:08It's gonna give me a response back. And you can do essentially everything that you could do inside antigravity just inside the terminal here.
34:16Personally, I don't like using this that much because, essentially, you can run everything in a way better user interface from within, um, from within Cloud Code.
34:27And almost all the time, you need to use the terminal. You can actually just ask Claude to do it on your behalf. Like, you don't have to open it up.
34:36You just tell Claude, hey. Do this thing for me, and it does the thing for you. If you ever need the terminal inside something like antigravity, Versus code, cursor, you can always come up come up to the top and toggle it.
34:49And now we have access to the terminal over here so we can do anything that we want from within the terminal. There's legitimately and it's rare, but there are some instances where Claude needs you to enter into the terminal.
35:00It can walk you through how to do it step by step when you're messaging it back and forth. You can open up the terminal, take the actions that are required, and then you're all set. That's it for this video, guys.
35:11Thank you for watching. I hope you found value in this. If you did, make sure to hit that subscribe button and that like button.
35:16I also have a free community where I give all of my YouTube blueprints away for free to automate your business and your life, and I also have a paid community where there's two transformations. The first one is for those of you looking to create your own AI automation agency, your business. I'll show you in the shortest time frame possible how you can find, close, and fulfill your first deal within thirty days or less, and there's hundreds of people that have made this work, even those without freelancing and business experience before.
35:40And the second transformation is for those of you who are business owners, I'll give you the exact blueprints that allowed me to scale to seven figures, automate 80% of my business using tools like Cloud Code. Now, obviously, you know, as you go through these transformations, it's not like a linear path. You're gonna have setbacks, and there are seven calls every single week where we can jump on together and talk back and forth to get you unstuck.
36:01And if you don't wanna have to implement any of this stuff yourself, I have an agency that can help you out with that entirely. So thanks guys for watching, and I'll see you
The Hook
The bait, then the rug-pull.
Jono Catliff opens with the exact frustration his audience carries into every search: too many concepts, moving too fast. He promises to cut through all of it by teaching like you're 10. What follows is 36 minutes of the most structurally disciplined Claude Code education on YouTube, with every concept pre-packaged into its own shareable slide.
Named ideas worth stealing.
07:00list
The 5 Permission Modes
- Ask for edits
- Accept edits (auto file edit)
- Auto mode
- Bypass permissions
- Plan mode
The further down the list, the more autonomous Claude becomes. Plan mode = think first, act later.
Steal forExplaining AI autonomy levels to clients or in onboarding docs
24:14analogy
Skills vs Sub-Agents Restaurant Analogy
Skills = chef cooks in front of you, visible and interruptible. Sub-agents = kitchen delivers the dish, black-box and self-contained.
Steal forAny explainer on agent architecture for non-technical audiences
12:59concept
Token Cost Curve
Token costs are exponential not linear. Every new message adds the entire conversation history as input tokens.
Steal forContent on managing Claude costs, timing your /compact correctly
04:21model
Context File Stack
- Humor file
- Tone file
- Vocabulary file
- Sample writing
Build a personal context stack in your project. Feed Claude samples of your own writing to extract and store your voice.
Steal forCLAUDE.md setup tutorials, personal brand AI voice workflows
How they asked for the click.
VERBAL ASK
35:06product
“I also have a free community where I give all of my YouTube blueprints away for free... and a paid community where there are two transformations.”
Double CTA: free Skool group with blueprints, paid community with two tracks (agency path + business automation). Delivered conversationally over 60 seconds. Soft social proof. No hard sell.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
- Instagram ↗
- TikTok ↗
- LinkedIn ↗
- X ↗
- Full crash course on Claude Code ↗
- Full crash course on Claude CoWork ↗
- Full crash course on Claude Code SEO ↗
- Book a time ↗
- Get help with your business ↗
- n8n ↗
- Make.com ↗
- Go High Level ↗
- Apify ↗
- Skool ↗
- Zapier ↗
- PandaDoc ↗
- Apollo ↗
- ManyChat ↗
- Vapi ↗
- PhantomBuster ↗
- ClickUp ↗
- ElevenLabs ↗
- Upwork ↗
- Instantly.ai ↗
- Airtable ↗
Visual structure at a glance.
