16:24Nick Puru | AI Automation · Tutorial
The REAL Way to Use Claude Code (From The Creator Himself)
How the engineer who built Claude Code actually runs 15 parallel sessions — and the six-part system non-developers can copy today.
June 4thThe argument in one line.
Claude Code is a business operating system for non-coders: give it your SOPs as skills, your rules as a CLAUDE.md memory file, and your real tools via MCP connections, and your job shifts from executing repeatable work to directing a parallel team of specialized AI workers and reviewing their outputs.
Read if. Skip if.
READ IF YOU ARE…
- You run a service business or agency and spend hours on repeatable admin — proposals, follow-ups, client briefs — that look the same every time.
- You have team members executing the same tasks differently and want a single shared playbook everyone installs in one command.
- You have tried using AI chat for business tasks and found it frustrating that it resets every session and cannot touch your actual files.
- You are curious whether Claude Code is only for developers or whether non-technical operators can get real leverage from it.
SKIP IF…
- You want a technical deep-dive into token budgets, prompt engineering, or agentic architecture internals — this is a business introduction.
- You need a tool that works without organizing your files and writing down your SOPs first — the value here scales with your input quality.
The full version, fast.
Claude Code is not a coding tool — it is a business OS that runs your repeatable work through specialized AI workers. The five-layer stack is: a CLAUDE.md file that stores your business rules (read automatically each session), skill files that encode your SOPs, plugin bundles that install your whole system onto a new team member in one command, sub-agents that act as department specialists running in parallel without contaminating each other's memory, and MCP connections that give those workers live read-write access to Gmail, Notion, Drive, and your CRM. The honest closing matters: multi-agent coordination burns tokens fast, Anthropic's own Project Vend autonomous shop lost money, and the people who automate everything at once reliably give up. Start with one workflow, get it working three times, then expand.
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 – 01:02
01 · Cold open — the cost of the assumption
Businesses still paying humans for AI-completable work; Zapier and Anthropic as social proof.
01:02 – 01:52
02 · The reframe
One place, your business context, plus a team of AI workers you direct.
01:52 – 02:55
03 · How it works
Operator-in-the-middle mental model. Anthropic non-engineering teams as internal evidence.
02:55 – 04:44
04 · Setup — picking your door
Terminal vs desktop vs IDE; Cursor recommended; subscription vs API key pricing; pro plan at $20/month.
04:44 – 06:24
05 · Memory — CLAUDE.md
The business brain: write it once, read every session, onboard once like a new hire.
06:24 – 08:37
06 · Skills — saved SOPs
Email reply skill as live example; team sharing; auto-detection; skills only load when needed.
08:37 – 10:08
07 · Plugins — bundle and distribute
One installable box for the whole team; official + community marketplace; build your own private one.
10:08 – 13:15
08 · Sub-agents — AI departments
Specialized workers per job; separate memory; parallel execution; start with one not twenty.
13:15 – 15:35
09 · MCP — connecting real tools
Universal plug concept; Gmail, Drive, Slack, Notion; 500 calls analyzed as live use case; two to four tools max.
15:35 – 17:45
10 · Staying in control
Plan mode, permissions, undo. Human in the loop on money-out and client-facing outputs.
17:45 – 20:53
11 · Making it run on its own
Hooks, custom commands, schedules (local vs cloud). Shift from tool to autonomous system.
20:53 – 21:35
12 · Where it falls short
Project Vend lost money. Multi-agent coordination is still preview and burns heavy token usage.
21:35 – 22:52
13 · Cost and quality warnings
Million-dollar surprise bill story; garbage in, garbage out; golden rule: one workflow, three times, then expand.
22:52 – 24:57
14 · The real change — doing to deciding
PwC 10-week to 10-day case; Intercom 86% resolution rate; CTA to free guide and agency offer.
Lines worth screenshotting.
- Claude Code reads CLAUDE.md automatically at session start — one setup session replaces every re-explain-my-business prompt forever.
- A skill file is just three things: what the job is, when Claude should use it, and the steps — nothing technical about it.
- A plugin is a single installable box of skills, workers, and rules — a new team member gets your whole system in one command on day one.
- Sub-agents keep separate memory so they do not contaminate each other's context; that separation also lets several run at the same time.
- Anthropic's own guidance says one well-tuned worker beats ten you never refined — start with one department, not a full org chart.
- MCP connections give Claude Code read-write access to real tools; the practical limit is two to four trusted connectors before costs and noise climb.
- Plan mode makes Claude write its plan before touching anything — the correct default for any job that touches more than one file or sends anything out.
- Anthropic's Project Vend experiment — an autonomous shop Claude ran for a month — lost money. Autonomous general management is not ready.
- A company rolling out multi-agent teams at scale got a million-dollar-plus unexpected bill not because it broke but because nobody watched the usage.
- The shift this enables is not faster task execution — it is moving from executor to director, with AI handling the queue and humans approving the exceptions.
- Schedules can run locally while your computer is on or in the cloud via a GitHub repo or managed agents — local is the correct starting point.
- The garbage-in rule matters more here than in chat AI because a workflow that runs autonomously on bad data will produce bad outputs at scale with nobody in the loop to catch it.
The five layers that turn Claude Code into a business team.
WHAT TO LEARN
Claude Code becomes genuinely useful for non-coders only when you build all five layers — skip any one and you are still just chatting with a reset-every-session assistant.
01Cold open
- Businesses paying humans for AI-completable work are losing ground to competitors who automated months ago — the cost is compounding daily.
04Setup
- Choosing a subscription over an API key is the correct entry point for business use — flat predictable pricing lets you focus on the work, not the meter.
05Memory
- CLAUDE.md is the prerequisite for everything else: without a written record of how your business operates, every AI session starts from zero regardless of what tool you use.
06Skills
- A skill file is just a written SOP — three sections, no code — and Claude auto-detects when to use it, so you call it in plain English rather than navigating menus.
07Plugins
- A plugin turns your personal system into a team system — one install command replicates your exact setup across every seat without manual onboarding.
08Sub-agents
- Specialization prevents hallucination: one sub-agent per job type with its own bounded memory outperforms one agent given access to everything, because context overload reliably produces wrong answers.
09MCP connections
- MCP connections are where the tool becomes an operator — the difference between Claude talking about your email and actually opening it, drafting a reply, and filing the response.
- Two to four trusted connectors is the practical ceiling — piling on more tools increases token consumption and makes outputs harder to trace.
10Control
- Plan mode is the correct default for any job that touches real data: reading a plan costs seconds; unwinding an autonomous mistake costs hours.
11Automation layer
- Schedules running locally (while your computer is on) are the lower-friction starting point; cloud schedules require a code repository and are the right upgrade once one workflow is proven.
12Limitations
- Anthropic own Project Vend test — a month-long autonomous shop — lost money on pricing confusion, which sets a concrete ceiling on what fully unsupervised AI general management can currently do.
13Cost and quality
- The cost warning is structural, not incidental: multi-agent workflows consume tokens in proportion to the work, and organizations rolling them out wide without usage monitoring reliably face unexpected bills.
- The one-workflow rule is the difference between people who adopt this and people who give up: nail one repeatable job end-to-end before building the second.
Terms worth knowing.
- CLAUDE.md
- A plain-text file in your project folder that Claude Code reads automatically at the start of every session — the place where your business rules, tone, prices, and operating procedures live persistently across conversations.
- Skill
- A saved playbook file (essentially a standard operating procedure) that tells Claude Code exactly how to handle a specific recurring job — what it is, when to use it, and the steps to follow.
- Plugin
- A bundled package containing a set of skills, sub-agent workers, MCP connections, and rules that any team member can install in one command to get an identical Claude Code setup.
- Sub-agent
- A specialized AI worker with its own instructions and separate memory context, assigned to one department or job type and capable of running in parallel with other sub-agents.
- MCP (Model Context Protocol)
- A standardized connection layer that lets Claude Code read from and write to external tools — Gmail, Notion, Google Drive, Slack, a CRM — turning it from a chat interface into an active operator of your real business tools.
- Plan mode
- A Claude Code setting where the AI writes out its full plan of action before executing anything, letting you approve, edit, or cancel before any file is changed or message is sent.
- Hook (automation)
- A rule that fires automatically when a specified event occurs — for example, auto-checking every draft a worker produces before it ever reaches the operator.
- Project Vend
- An Anthropic internal experiment where Claude was given autonomous control of a small online shop for roughly a month; it lost money due to pricing confusion, establishing a concrete baseline for where fully autonomous general management currently fails.
Things they pointed at.
23:29linkPwC AI productivity study
23:45linkIntercom Claude agent (86% resolution)
Lines you could clip.
00:15
“I run my entire business on it and I've never written a single line of code.”
11:56
“A team of one good worker beats a mess of 10 that you've never tuned.”
22:18
“Just pick one workflow. Get it working really well three times in a row and then you can expand from there.”
22:56
“You stop being the person just grinding through the tasks and you become the person directing the work.”
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.
metaphoranalogystory
00:00Right now most businesses are still paying people to do work that AI could finish in seconds. And the ones who figured that out months ago, they're quietly pulling ahead of everyone still doing it by hand. So this video is about Cloud Code and it's probably not what you think.
00:11Now most people hear the name and picture a tool for some programmer, something that they will never touch. And they are wrong. And that one assumption, it is literally costing them hundreds of thousands if not millions of dollars.
00:21I run my entire business on it and I've never written a single line of code. This was never even about coding. It's about having a team of AI workers that actually are able to do work for you instead of just talking about it.
00:33And this scales. Zapier, the automation company, they now run more of these AI workers inside of their businesses than they have employees. Cloud, they use AI for 99% of what they do to enhance their product and even marketing.
00:43So here's the plan. I'm going to walk you through the exact setup that I use inside of my business and for our clients and every piece that matters for a business just like yours and nothing that you don't need. But with that being said, just using a normal AI chat through your browser, it has three problems that makes it completely useless for running real work.
01:00However, this setup, it fixes all three of those. And by the end of this video, you will know exactly how it works, how to set it up, and how to get your whole team using it.
01:07Now just a quick intro, if you don't know me already, my name is Nick. I've spent the last couple of years implementing AI systems inside of real businesses and teaching thousands of entrepreneurs how to sell AI effectively. But enough said, let's dive into it.
01:19So before we actually touch anything, let me just reframe what this thing even is because this one shift is going to be what makes the rest click for you. Now the simplest way to be thinking about this is we're just going to have one place that is holding everything about your business plus a team of AI workers that can actually do the tasks and use your real tools.
01:36So you're not going to be just chatting with it. You are literally handing it entire job. So here's the whole idea in just one picture.
01:43>> [snorts] >> So this is going to be you inside of the middle. You are going to be the operator, right? Now on one side, this is going to be your AI team right here.
01:52On the other side, this is just going to be all of your real tools. So this can consist of different things like your email, any of your documents, any of your files, and you just sit in the middle and you are directing it, right? And it's not just me saying this because the company that actually makes this, Anthropic, they have said that their own teams outside of engineering run real work on this as well.
02:11So, just to name a few different departments, their legal team, they built a working prototype of a phone system, their marketing team, they've spun up hundreds of different ad variations and different go-to-market systems. People who had just never written code or understood anything about computer science, they're building real tools just by describing in plain English what they want.
02:29So, that's going to be the entire shift. It's not a smarter search box or anything like that.
02:33It's just a place to run your business with a team. But now, let's just set it up and it's actually much easier than you might expect. So, jumping into the setup, this is actually the part that most people are scared of and it's honestly incredibly easy.
02:45So, you've actually got a few different ways that you can be running this. So, there's going to be the terminal, then there's going to be a desktop application, and you can also just run it right inside of the editors that people already use like VS Code. So, all of these are they're called IDEs.
02:59So, VS Code, Cursor, Anti-Gravity, it's the same Claude Code underneath. You just pick your door.
03:03Now, me, I often use Cursor. It's free and it just feels very normal to me. Not some like scary black terminal window.
03:09So, that's the one I'm going to be setting up here. So, first, you just go ahead and download Cursor. You can just grab it off of their site.
03:15It's completely free and you would install it just like any normal application. And then from there, you just simply open it up. Now, over on the left, there's going to be an extensions button.
03:23So, you click that, you search for Claude Code, and you hit install. It's that simple. That is the whole thing.
03:28It drops right into the editor for you. Next up, it'll just ask you to sign in and one quick tip here that actually saves you money is you can sign in with a paid Claude subscription, not with an API key, and this is what you would 100% wants to be using instead. Because with the subscription, you are going to be paying just one flat price and you stop thinking about it.
03:45With the API key, you have to be paying per use, which means that you're basically just watching a meter the whole time. So, for running a business, flat and predictable wins every single time. It's what I and my team use.
03:56Now, just to cover the plans real quick, Claude code is actually included on the pro plan, which is about $20 a month, and that's where I would start almost everybody. Now, above that, there's going to be max at $100 and a bigger one at $200 for people who live inside it all day, who are using it more.
04:10So, this is what my team and I use. And as you might expect, free plan doesn't include Claude code. So, pro, it's your entry point for this and what I recommend for you.
04:18And that's the whole setup. Editor, extension, you sign in, and you're ready. That's it.
04:22And by the way, real quick, if you do want to follow along, I'm going to be putting a free guide inside of my school community with the exact doc and the templates, all the prompts that I'm about to show you, so you can just copy the whole thing. It's very easy. It's free.
04:33Link will be down below in the description. We have plenty of other guides and resources that we do not post on our YouTube channel inside of there, too. We've got about 20,000 members, so make sure to check that out if you're interested.
04:43But, let's get back to it. So, the first real piece, and honestly, the most important one, this is the memory. Now, this is what makes literally everything else work.
04:52So, there's going to be a file called Claude.md. Now, the easiest way to be thinking about this is this is just where your business effectively lives. So, it's just a plain text file where you write down everything about how you operate, and Claude, it reads it automatically at the start of every single conversation.
05:06So, instead of just re-explaining who you are every time, you just write it once. So, who you serve, how you talk, any of your prices, any of your rules, the way that you like things done, and from then on, it just knows. So, you can save anything important into the file by hand, and it also just quietly picks up little corrections that you make as you go on.
05:24So, instead of just resetting to zero every single morning, it keeps getting more like you. And the way that I describe it is you're just onboarding it once like a new hire instead of just re-explaining the job every single day. Now, here's mine.
05:36So, this is the real one for my agency. You can see it's not just code, it's actually just plain English. So, here's who you work with, here is our tone of voice, here's the stuff that we never do, and things you have to be cautious of.
05:47None of this is technical. It just reads like a note that you would write for a new team member. Now, one thing worth knowing is that you could have one of these at the top level for your whole business and you could have smaller ones inside of specific projects.
05:58They're the main rules, they apply everywhere, and then you have a specific client or projects can have its own extra notes on top of them. That's how I use it. >> [snorts] >> So, big rules, they're going to be going at the top.
06:09Your specific notes where they actually belong. Now, there's also a quick command called {slash} init, and this is just where Claude will look at a folder and write you a first draft of this file automatically. So, you don't even have to start from a completely blank page.
06:22So, that is the memory. Now, let's give it some skills. So, a skill is just a saved playbook.
06:26It's an SOP. It's the way you teach Claude to do a specific job in your way and how your business actually operates and how they go about those processes and just doing it once so it does the same way every time after that. So, if you've ever written an SOP, which is a standard operating procedure, so a specific way of doing a task, that is just exactly what this is, except Claude actually follows it step-by-step.
06:46Now, the magic with this is you don't even have to dig it up and run it. So, Claude it notices when a job matches just one of your skills and it pulls it up on its own. So, when I say something like draft a proposal, it goes off, it finds and sees that I have a skill for that and it follows my exact steps.
07:01So, I can either just in plain English call it like that or I can just get it a little bit more specific and say, you know, like {slash} draft proposal if I've created it with that name. Now, a couple of things that make these really good is they only load when they are actually needed. So, they don't slow everything down inherently.
07:16And you can also share them across your teams or collaboratively. So, this means that your whole team can be running the exact same playbook instead of everybody doing it their own way. So, when we are working with a team, we make sure that everybody's going to get access to the same things.
07:30Of course, if they're across different departments, then they're not going to need access to some things, but this is how we build this for operations. So, this is just a stack of skills that I've built up for my own business. Now, every one of these it's just a job that I was doing over and over again.
07:44I wrote it down once so that Claude can just handle it the same exact way every single time. It saved me hundreds of hours, probably every single month I would say.
07:52Now, let me just open one so you can actually see what's inside and notice how simple it actually is. So, let's go. It's really just three different things.
08:00What the job is, when Claude should be using it, and then the steps. Now, this one is my email reply skill. So, the steps are just how I want my emails to be sounding, how short to be keeping them, what to always include, and what never to touch without checking with me first.
08:15So, that's the whole thing. So, now when I ask for a reply, it comes back actually sounding like me and not just some generic robot every time, right? So, that's the idea.
08:23We want it to be in our own language and mimic our behaviors or our patterns in our speech. So, ultimately, skills are just how you bottle the way that you work or how your business operates. Now, what if you want to package up a whole bunch of these and hand them to your team in just one go?
08:36Well, that's the next piece. So, this next one, it's a big deal if you've got a team, and this is the plugins. So, a plugin, it's basically just a box where you take a bunch of your skills, your workers, your connections, your rules, and you bundle them all into just one package.
08:49And then, anyone in your team can actually install that whole package in simply just one command and instantly have the exact same setup as you do. So, think about what that solves. So, normally, getting a new team member set up the right way, it might take forever for you.
09:03With this, you hand them just one box, they install it, and they're running your whole system on day one. So, you get the same playbooks, you get the same standards, you get the same everything. And there's actually marketplaces for these.
09:14It's kind of like an app store. So, there's an official one from Anthropic, there's a community one, and you can also host a private one just for your own company. So, you can grab ready-made ones, or you can just build your own, and you can keep it internal.
09:26So, look at this. You don't have to build every one of these yourself. There are loads of these ready-made packs out there right now for all kinds of different work.
09:33Like, Anthropic built a whole one just for small businesses, like the everyday finance and sales and admin jobs already done for you. I actually made a full video breaking that down already, so I'll link it if you want the full deep dive, but that's just one of them. So, you could browse what's out there, pull the ones that you want, and what is most applicable to in your business, and your departments, and jobs, and processes, and they just drop straight into your Claude code setup.
09:56Now, the reason that I am putting this right after the skills is that it's the natural next step. So, first, you write your playbooks, and then you box them up so your whole company runs off the same box. And that's just a plugin.
10:06Now, let's talk about the workers themselves. Okay, now this is the one that makes the whole AI team idea real. So, this is sub-agents.
10:12Now, a sub-agent, this is just one AI worker that you've given a specific job and a specific set of instructions. So, the idea is you write it once, and after that, when the work just comes up, Claude hands the job to that worker on its own, so you don't have to go call it specifically. So, here is how I picture it, and this is the diagram in the document, but that's you at the top, and under you, these are your departments.
10:34So, proposal worker, we have a content worker, we have an operations worker, we even have a research worker. So, each one is just a specialist that only knows its job, and it does an incredibly well. So, we're not creating just one ultimate sub-agent where it has access to how to do everything, because in that case, we're going to experience context bloat and just feed it too much information, it's going to get confused, it's going to hallucinate, and just give you wrong answers.
10:57It's inevitable. So, that's why we split it off into different specialized sub-agents. Now, there's two particular things that actually makes this powerful.
11:04To actually expand on that a little bit further, each worker, it just keeps its own separate memory, so they don't trip up over each other, clog up the main conversation. And because they're actually separate, several of them can work at the same time.
11:15So, if you're not doing every task one by one anymore, you're running departments at the same time, all synchronously. Now, these are my actual departments sitting right here. So, each one's a worker that I have set up for a specific part of a business, and a worker, it's just a file.
11:30So, let me just open one up to give you an example. So, at the top it just says its name and a brief short description of what it does. And that description is what Claude uses to decide when to send it to work.
11:39And then below that, this is just the instructions. And that's the whole worker. Now, one thing worth saying, and it comes straight from Anthropic's own guidance, don't go overboard building 20 workers on day one.
11:48You want to just be starting with one, and then from there you can just add a specialist only when you've actually got work that is separate enough to need its own lane. So, verbatim they say a team of one good worker beats a mess of 10 that you've never tuned. And I completely agree with that statement.
12:02So, with that being said, that is your departments. And next is what happens when you actually just let them all run at once. All right, so at this point you've got a few workers, and now you get to do the thing that actually feels like running a company, which is running them in parallel.
12:14So, here's the picture. You hand out the work. One worker is maybe drafting the proposal.
12:18Another is going to be pulling the research. Another one is going to be running the follow-up. And they're all running and going at the same time.
12:24And then the results come back and get pulled together. So, that is the operator seat.
12:28So, we have a few different things moving at once with you pointing the way. So, there's actually a newer version of this where you can just ask Claude to coordinate a whole group of workers on a bigger job, where they're going to be sharing a task list and pass work between all of each other. So, it's genuinely impressive when you see a big job just get chopped up and handled all at once.
12:48But, I'm going to be straight with you here because this is exactly where people get oversold. This bigger team stuff, it's still early. It's still a preview, and it burns through a lot more usage.
12:58So, the person who created Claude Code said it very plainly that these teams use a lot of tokens. So, save it when a job is actually genuinely big and worth it, not just because it looks cool or it's idealistic. So, that's how the team scales.
13:09But, none of this is going to be mattering if it cannot touch your actual tools. So, let's connect it to your real business. So, so far everything has been inside of Claude Code.
13:18But, to actually run a business it needs to reach inside of the tools that you are already using. So, things like your emails, your documents, your files, your customer records, and your CRM. That is what this piece does.
13:28So, it's got a technical name, and it's just called MCP, model context protocol, but the idea with it is dead simple. So, let me explain that. So, you just think of this as one universal plug.
13:38So, on its own, Claude can talk about your email, but it can't open it. So, this is where you connect the plug, and now it can actually read your email. It can draft replies.
13:46It can pull a file. It can update a record. So, the picture with this is Claude coded in the middle, and your real tools are plugged in around it.
13:54So, we have Gmail, Google Drive, Slack, Notion, your accounting, maybe your meeting notes. Each one, it becomes something that your team can actually use. So, this is what mine looks like, and these are the tools that I've actually plugged in.
14:05And once they're actually connected, I can just say something very simple, like pull the notes from yesterday's call, and then turn them into a draft, and it just does it, completely end-to-end across two different tools. I have used this so many times, and it's one of the highest leverage things that I've ever done, is pulling every single call that I've had.
14:21I've had probably about 500 calls over the past 3 years inside of my AI business, on sales calls, and meetings with any of my students, and just finding like what the most important things are. So, like what is the top niche of businesses that we've spoken to? What are the top problems?
14:35So, being able to literally pull from hundreds of meetings, and getting all of the data from all of that. Granted, something like that would be and has been about an hour task, and it takes quite a while, but you can just get an idea of what you can be using it for inside of your business, or whatever you are doing. So, this is literally where everybody's saying, you know, you can run your company with AI, stops becoming a slogan, and it starts getting very real, because now one instruction can move through several different of your tools all at once.
15:01Now, just two things to keep in mind here, though. One, more is not better here. So, you can plug in the two, three, four tools that you actually live in the most, and I would just stop there.
15:10Now, if you keep piling on too many tools, it gets pretty noisy, and you'll notice that sometimes your tokens are going to get eaten up a little bit more. But, number two, stick to the official well-maintained connections.
15:21So, Anthropic, they are very clear that they do not personally vet every third-party connector out there. So, just use the trusted ones, not some random thing that you found. I would keep it a little bit safer with this.
15:29That That being said, your team now has hands. Next, let's make sure that you just stay in control of what it is actually doing. So, here's the question that everybody has at this point, and it's a pretty fair one.
15:39If this thing can get into my files and my email and actually do stuff, how do I stop it from doing something that I didn't want it to do? There's three simple things that you can do to keep you in control and prevent anything detrimental. So, first is plan mode.
15:52So, this one's for any of the bigger jobs. So, say you ask it to clean up a messy folder just full of client files or just go sort through your whole inbox. Now, you don't want it to just dive in and move 100 things around.
16:03So, this is where you turn on planning mode and instead of touching anything, it stops. It writes out exactly what it is planning to do first, and then in this case, you just read it over. If the plan looks right, you can hit go, you can approve it, or you can suggest edits.
16:17Because if it doesn't do anything or if it doesn't create a plan that you do not like, you just tell it what you want it to change. Nothing happens until you are happy with it.
16:24Number two is permissions. So, out of the box, it's actually very careful where it'll look at the things on its own, but the second that it wants to actually change a file or send something out, it stops and it asks you first. So, you can see right here.
16:36It is asking you. You can let it do that one thing or you can just tell it to, you know, trust it and let it run. But either way, you are going to be the one deciding.
16:44Now, one thing that I typically do inside of mine is I just skip permissions and often just let it go haywire and do whatever it wants. Because 99% of the time, it's not going to be doing anything detrimental.
16:54But in this case, if you are, you know, having to adhere to compliancy or different security concerns, then you will want it to be a little bit more cautious and not just allow it to skip the permissions. Now, number three is the undo. So, there's a rewind button right here that takes you back a step if something does go wrong.
17:10So, you're never going to be stuck with a mess. So, hypothetically, if we had messed up right here and we don't want to commit with this or if we just have to like revert what we just did, we could just go back. It's that simple.
17:20Now, getting into the actual point of all of this, so the goal it isn't to just change your whole business to a robot and walk away. It's to put a capable team in front of you with you checking the stuff that really matters. So, for anything important, money going out or something getting sent to a client, >> [snorts] >> you still should be signing off.
17:38You should still be having that human in the loop. And it's not a limitation, that's just how you run this properly. So, that is the control piece.
17:45So, now let's make it run without you baby-sitting it every single second. So, so far at this point, you're still the one kicking everything off, but this next piece, it's about making the routine stuff happen on its own. So, first, this is going to be the hooks.
17:57Now, a hook, this is just going to be a rule that runs by itself when something happens. So, every time a worker actually finishes a draft, it's automatically going to get checked before it ever reaches me.
18:06So, you set that up once and then from then on, it just happens every time without you thinking about it or having to provide any additional instructions to make that happen. Now, number two is your own commands. Now, these are just going to be your shortcuts for the stuff that you do over and over again.
18:21So, it's almost like a skill, and I've actually got a whole bunch of them set up now, and each one, it's just a little job that I run quite often, and I just saved it so I can fire it off by name. So, you can see like this command right here.
18:32Like instead of typing out the same long request every single time, I just type the command in one word and the whole thing kicks off. It's just like a shortcut on your phone or on your computer. Now, number three, one of the most important pieces is the schedules.
18:44So, instead of you kicking off a job every time, like you set it to run on a timer. So, for example, every morning at 8:00 in the morning, just go build my competitor brief, and it's done before I sit down. So, there's two actual versions of this, and the difference matters where one runs you right on your own computer.
19:01So, this is going to be able to see your files, and there's no extra setup, but it only goes while your computer is on and the application is actually open. Now, the other one, it runs up in the cloud. So, even when your laptop is shut off, and that one's definitely more hands-off, but it's a more technical setup where your work has to be sitting in an online code repository like GitHub or using Claude's new feature or platform called managed agents.
19:23So, if you're just starting out, the one on your computer is going to be the easier way in, but I do have other videos on how to set that up. So, if you're interested, go make sure to check that out. But, that right there, that's the shift from a tool that you use to a system that runs completely without you.
19:38So, all of the boring, repeating parts of your week, the reports, the weekly check-ins, the roundups, like those can just happen on their own, and you only have to step in when something needs you. So, that effectively is automation, and it's not even chained to your desk anymore. But, that said, I do want to cover one last thing, and it's going to be more about where this is all heading.
19:56So, with this, like you can send a job off to just run in the background and just carry on with something else completely different, and there's actually a dashboard that pulls all of this together. So, you can see, like this shows me every job that is running, which ones are finished, which ones are just waiting on me. So, I can have three or four different things going at just once and just glance in on them, like checking in on a team.
20:16And the newest one, it's actually called computer use, it's still pretty rough, but Claude can actually take over your screen and click through an application for you for the tools that do not plug in any other way. Now, none of this is day one stuff, I just want you to see that it keeps going beyond this.
20:30It's not just stuck on your desk, and there's so many features and so much that you can be exploring to ultimately increase leverage inside of your business. And another quick thing so that you're not actually confused out there, there's going to be a separate application from Anthropic called co-work, and this is just built for non-coding work, so it is very good, but it's its own thing.
20:47So, everything I'm showing you today, this is Claude Code, and I just want you to know the difference so you don't go in circles here. But, okay, I've shown you the whole system. Now, this is the part where it actually falls short because that matters just as much, right?
20:59So, this isn't just simply going to be a magic button that runs your whole company while you sleep. Anthropic themselves, they ran an experiment on exactly that. They actually called it Project vend where they just had Claude run a little shop on its own for about a month or so and it lost money.
21:15It gave things away. It got confused on its pricing and the honest lesson from Anthropic's own write-up is that this is all fantastic at specific well-defined jobs and it's not ready to be left fully alone as a general manager, which is exactly how I'm telling you to be using this.
21:28Have real jobs with specific defined functions where you are going to be reviewing all of the important ones, all the important things. So, two quick things. First is going to be the cost because the AI workers, they actually do do very much, but they use more under the hood than a quick chat is going to and most of the heavier plans bill by usage.
21:45So, there was actually a recent story about a big company rolling this out wide and getting surprised by the bill. It was about over a million dollars or something insane and it's not because it was broken, but because they didn't watch the usage.
21:56So, the move is very simple. Just start small, keep an eye on it, just understand what is happening and scale what's clearly worth it for you. Number two is garbage in, garbage out.
22:05So, if your files are just complete mess and your notes are chaos, your team's output will be a mess, too. So, the businesses that actually win with this, they're going to be the ones who feed it clean, organized information and they're prompting it correctly. So, just do a little tidy up of your stuff.
22:18It's going to be paying off very much for you. And the golden rule on all of this is just pick one workflow. Get it working really well three times in a row and then you can expand from there.
22:27So, the people who actually just try to automate everything at once, they often give up. Things don't work. Their team doesn't actually use it and they just, you know, go back to manually doing things and complaining that AI doesn't work.
22:38We see it time and time again. But, the people who nail one thing first are the ones who actually end up running their whole operation on it and start, you know, scaling practically. So, that is the honest picture with all this, but here's what actually changes for you when this is in place.
22:52So, when all of this is actually set up, the real change isn't just that you get a faster tool. It's that your job changes. and checking the outputs.
23:04That's simple. So, you're moving from doing to deciding, and that's the whole point of running a team. You just happen to have built this one out of files and instructions instead of hiring for every single seat of every process that you need.
23:16And none of this is wishful thinking. Like Anthropic, they put out real numbers on their own company. A huge share of their work now runs directly through Claude just using AI, and they measured a real jump in how much their people get done, including our engineers shipping a lot more every single day.
23:30Now, an accounting and consulting giant, PwC, said they took a process that used to take 10 weeks down to 10 days. A support company, Intercom, they built an AI agent on this that resolves the majority of customer questions all on its own. And one last thing, this isn't going to be replacing the thinking, it's just taking the busy work off your plate so you got room to actually run the business and focus on a higher revenue-generating tasks.
23:51So, you just stop being the bottleneck for every little nuance monotonous task. So, that's the whole thing. You give it a memory so it knows your business.
23:58You then write the playbook so it works in your way. You set up all of the workers so it has a team. And you can just plug in your real tools so you're going to actually do the work for you.
24:06And with that, you're going to be staying in control of the parts that are truly matter. So, none of this is about coding, it's just about building yourself a small team that handles the work that you've already been drowning in so you can get back to the parts that only you can do.
24:18Or if you want to get out of the business and, you know, focus on time with your family or whatever else it may be. So, if you do want to actually build this, I put the whole thing in a free guide in our community. The document from this video, the example playbooks, the worker files, literally all of it.
24:31It's ready to copy, it's free. The link will be down below in the description. And if you'd rather not build it yourself, if you'd rather just we just set this up for your team and hand it over, that's literally what we do inside of my agency.
24:41There's a link for that down below in the description. You can book a free AI strategy call with my team. We'll show you how we can increase leverage in your bottom line this year.
24:49So, either way, you've got the full picture now. Pick the one piece that would save you the most time this week and start there. Well, that being said, thank you guys for watching and I'll see you in the next video.
The Hook
The bait, then the rug-pull.
Most Claude Code tutorials are for developers. This one is addressed to the business owner who has never opened a terminal, does not intend to, and is quietly starting to wonder if that assumption is costing them something real.
Named ideas worth stealing.
04:44list
The Five-Layer Claude Code Business Stack
- Memory (CLAUDE.md)
- Skills (saved SOPs)
- Plugins (bundled packages)
- Sub-agents (department workers)
- MCP connections (real tool integrations)
The layered architecture for turning Claude Code into a full business operating system, introduced progressively across the video.
Steal forAny onboarding doc or workshop for non-technical business owners adopting Claude Code
15:35list
Three Control Levers
- Plan mode — write before touch
- Permissions — approve each action
- Undo — rewind a step
The three built-in safeguards that let non-coders delegate without fear of irreversible mistakes.
Steal forRisk objection handling in any AI adoption pitch
17:45list
Three Automation Layers
- Hooks — event-triggered auto-rules
- Commands — named shortcuts for recurring jobs
- Schedules — timer-based job execution (local or cloud)
The progression from manual delegation to fully autonomous background execution.
Steal forSOW scope descriptions for AI automation agency clients
How they asked for the click.
VERBAL ASK
24:27link
“I put the whole thing in a free guide in our community. The document from this video, the example playbooks, the worker files, literally all of it. It's ready to copy, it's free.”
Dual CTA: free guide (Skool community) for DIY, paid agency offer (reprisesai.com) for done-with-you. Positioned as either/or, not upsell. End-card appears with agency branding.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
