Modern Creator
Nick Puru | AI Automation · YouTube

I Automated 99% of My 3 Businesses With Claude Code

A live walkthrough of the manager-plus-specialist Claude agent system one founder built to run his content brand, his 20,000-member community, and his automation agency from one folder.

Posted
yesterday
Duration
Format
Demo
educational
Views
194
22 likes
Big Idea

The argument in one line.

A single manager agent that routes tasks to isolated, single-purpose specialist agents — each with its own rulebook, memory, and app connectors — can automate roughly 90% of the repetitive research, drafting, and outreach work across multiple businesses, while every consequential decision stays with a human.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You run more than one content or service business and are drowning in repetitive research, drafting, and outreach work across all of them.
  • You already use Claude Code or the Claude app and want a concrete blueprint for turning single-shot prompts into a standing team of agents.
  • You're comfortable working from plain-English instruction files and folders rather than a dashboard SaaS product.
  • You want to see an unedited demo of AI agents doing outreach, content research, and campaign writing, not a hypothetical pitch.
SKIP IF…
  • You're looking for a done-for-you SaaS tool — this is a self-built, DIY folder system, not a product you install.
  • You have no existing content, brand voice, or client documents to feed the agents — the system only works because it's stocked with real material.
TL;DR

The full version, fast.

Nick Puru runs three businesses — a content brand, a 20,000-member community, and an automation agency — through one folder-based system: a single manager Claude agent routes every task to a specialist sub-agent (LinkedIn, Twitter, YouTube, newsletter, outreach, go-to-market), each with its own rulebook, brand voice, and memory, connected to real tools like Gmail, Slack, and ClickUp via connectors. He demos the YouTube idea-research agent, an outreach agent that finds verified leads and drafts personalized emails, and a go-to-market agent that writes and pressure-tests ad campaigns against 13 AI customer personas before spending on ads. Agents fire three ways — manually, via webhook, or on a schedule — and the system covers roughly 90% of the grunt work, while every real decision, client call, and piece of published work stays human-reviewed.

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 →
Chapters

Where the time goes.

00:0000:52

01 · 99% of my business runs without me

Cold open: the promise is stated and the presenter commits to showing the live setup instead of talking about it.

00:5204:04

02 · The system: one manager, many specialists

The manager agent and its specialist folders (LinkedIn, Twitter/X, YouTube, newsletter, community, ads) are introduced; each specialist keeps isolated memory and never shares voice with another.

04:0404:57

03 · Connectors: plug your apps in

MCP is explained as plainly 'plugging your applications into Claude' — Gmail, Slack, ClickUp, Notion, calendar, and a call recorder are shown connected.

04:5705:34

04 · Plugins: don't build every agent

The Claude app's Plugins tab is shown — pre-built, often free, bundles of skills (sales, marketing, support, finance) that replace building an agent from scratch.

05:3406:56

05 · Demo 1: the concept command

The YouTube specialist's ideation skill is run live on a random topic — it checks saturation, scores the idea, writes the brief, and drops it onto the ClickUp review column.

06:5608:06

06 · It says no to saturated ideas

The agent can kill a video idea before production time is wasted; LinkedIn and Twitter agents share the same content brain but write in separate voices, graded weekly.

08:0609:39

07 · Morning AI news into Slack

A scheduled agent scans Reddit and followed sources overnight and drops a news summary into a dedicated Slack channel before the presenter wakes up.

09:3910:41

08 · Demo 2: outreach

The outreach agent is introduced as powering both the AI Accelerators community and the Reprise agency, and is described as the same system sold to students as a product.

10:4112:07

09 · Finding leads & drafting emails

The agent finds businesses matching the ICP, writes personalized emails in the presenter's voice and format, and only ever uses verified email addresses, never guesses.

12:0713:30

10 · Demo 3: the go-to-market agent

A campaign request for a new AI offer is typed in; the agent pulls from stored frameworks, past winners, and offers to draft ad scripts, landing page copy, and hooks.

13:3014:40

11 · Testing copy on 13 AI personas

A second agent runs new copy through 13 AI personas modeled on the real customer base, tearing it apart before a dollar is spent on ads.

14:4015:44

12 · The three triggers

Every agent in the system fires one of three ways: manual command, webhook (an incoming event like an email), or schedule (a recurring clock).

15:4416:56

13 · The truth about "99%"

The presenter clarifies the number: it's roughly 90% of the grunt work removed, not the thinking — every real decision, sales call, and piece of client work stays human.

16:5617:42

14 · When it breaks

A stale knowledge file once caused an agent to confidently write a script around months-out-of-date numbers; the fix was refreshing source research before every write.

17:4219:18

15 · Where to start

Closing advice: don't automate everything at once — pick the single most-hated daily task, usually email, automate that first, then let the next candidate reveal itself.

Atomic Insights

Lines worth screenshotting.

  • One manager agent that routes tasks to single-purpose specialist agents outperforms one AI trying to do everything, because each specialist only ever writes in its own voice.
  • Keeping brand voice separate is deliberate: a LinkedIn agent's writing must never leak into the Twitter agent's writing, so each specialist keeps isolated memory and rules.
  • MCP is just the technical name for plugging your existing apps into Claude — once connected, an agent stops being a chatbot and starts acting inside Gmail, Slack, and ClickUp.
  • Anthropic ships free, pre-built plugin bundles for sales, marketing, support, and finance that replace building a specialist agent from scratch.
  • A content-idea agent that can say no to a saturated topic saves more time than one that says yes to everything, by killing weak videos before a week is wasted.
  • A lead-outreach agent built with a hard rule to only use verified emails, never a guessed one, is what keeps automated cold email from tanking deliverability.
  • Before spending a dollar on ads, a second agent runs new copy through 13 AI personas modeled on the real customer base, so reactions arrive before any media spend.
  • Every agent in the system fires exactly one of three ways: manually typed, fired by an incoming webhook, or run on a schedule.
  • A single stale knowledge file caused an agent to confidently write a script around a months-out-of-date statistic — the fix was refreshing source research before every write.
  • '99% automated' in practice means about 90% of the grunt work — research, drafting, formatting, outreach — is offloaded, while which video to make and which email to approve stays human.
  • The same lead-finding-and-outreach agent that runs the founder's own agency pipeline is sold to his students as the product itself.
  • The recommended way to start is picking the single most-hated daily task — usually email — and automating only that first, rather than automating a whole business at once.
Takeaway

One routing agent, many single-job specialists, three triggers.

AGENT SYSTEM DESIGN

The lesson isn't that AI writes your content — it's that separating routing from execution, and knowing exactly what triggers each task, is what makes an automated system trustworthy enough to leave running.

0199% of my business runs without me
  • A promise stated up front and then proven live earns more credibility than a description of a result without a demonstration.
  • When an entire market is talking about a trend without showing it work, a concrete live walkthrough stands out by contrast.
02The system: one manager, many specialists
  • Splitting one broad job into narrow, single-purpose roles, each with its own memory and voice, produces more consistent output than one generalist handling everything.
  • A routing layer that only decides who owns a task, without doing the task itself, keeps a growing system simple to extend.
03Connectors: plug your apps in
  • A set of instructions is only as useful as the real software it can act inside — connecting to your existing tools turns planning into execution.
  • Native integrations mean an agent can read a lead list, send a message, or update a project board directly, without manual copy-pasting between tools.
04Plugins: don't build every agent
  • Checking whether a pre-built solution already exists before building a custom one from scratch saves most of the setup effort.
  • Free, publisher-vetted plugin bundles for common business functions are a faster starting point than a blank prompt.
05Demo 1: the concept command
  • Turning a repeatable multi-step process into a single named command removes the friction that stops a good process from actually getting used.
  • Delivering a finished, scored output directly into the exact workflow stage a team already works from eliminates a handoff step.
06It says no to saturated ideas
  • A system that can say no to a weak idea before resources are spent on it is more valuable than one that says yes to everything.
  • Screening every idea against a saturation and demand checklist before committing production time prevents wasted effort on content nobody is searching for.
07Morning AI news into Slack
  • Automating the passive monitoring of a fast-moving space, so relevant updates are waiting each morning, removes the need to constantly check for them.
  • The habit of staying current in a fast-moving field can itself be delegated, freeing attention for the judgment calls that actually require a person.
08Demo 2: outreach
  • The same tool used internally to run outreach can double as the product sold externally — building it for yourself first proves it works before you sell it.
  • Treating a manual, disliked task like cold outreach as a system to design, rather than a chore to grind through, is what makes it worth automating first.
09Finding leads & drafting emails
  • A hard rule against ever guessing unverified contact information protects outreach quality, even when it means finding fewer leads.
  • Personalizing outreach to a defined ideal customer profile, rather than mass-blasting generic contacts, is what keeps automated outreach from feeling automated.
10Demo 3: the go-to-market agent
  • Reusing a documented set of past frameworks, offers, and winning copy is what makes new campaign output sound consistent with an established voice instead of generic.
  • Running the outline and concept step before generating full assets avoids building deliverables nobody has agreed to yet.
11Testing copy on 13 AI personas
  • Simulating a customer panel to pressure-test new copy before spending on media catches weak messaging before it costs real money.
  • Feeding a testing agent real customer data and past conversations, not just the new copy, makes its pressure-testing responses more representative of real reactions.
12The three triggers
  • Nearly all recurring work falls into one of three trigger types: started by a person, fired by an external event, or run on a timer.
  • Sorting existing tasks into those three trigger categories is a practical way to find which ones are actually automatable.
13The truth about "99%"
  • A claimed automation percentage usually refers to removing grunt work — research, drafting, formatting — not to removing judgment or final decisions.
  • Keeping every consequential decision and client-facing interaction with a human is the actual point of automating the busywork, not a limitation of the system.
14When it breaks
  • An automated system that reads from a knowledge base is only as current as that knowledge base — a single stale fact gets repeated with total confidence.
  • Reviewing finished automated output rather than trusting it unsupervised is what catches an error like an outdated statistic before it reaches a client.
15Where to start
  • Trying to automate an entire business at once tends to cause burnout and abandonment; automating the single most-dreaded recurring task first is more sustainable.
  • Once one automated task is trusted and running, the next candidate for automation tends to become obvious without a separate planning exercise.
Glossary

Terms worth knowing.

MCP
Model Context Protocol — the connection standard that lets a Claude agent read and act inside real apps like Gmail, Slack, or a project board, instead of only chatting in a text box.
Manager agent
A coordinator agent that does no writing itself — it only reads an incoming request, decides which specialist agent owns that type of work, and routes the task to that specialist's folder.
Specialist agent
A single-purpose agent scoped to one channel or job (e.g. LinkedIn, YouTube, outreach) with its own rulebook, brand voice, and memory, kept deliberately separate from every other specialist.
Scheduled task
A Claude app feature that fires a given agent instruction automatically at a set time or interval, such as every morning or every 30 minutes, without a manual start.
Webhook trigger
An automation that fires the moment an external event happens, such as a specific email landing in an inbox, rather than on a timer or a manual command.
Dream 100
An outreach approach that identifies a short list of the most valuable prospects and pursues each one with high-effort, personalized outreach, such as a custom video, instead of mass email.
Plugin (Claude app)
A pre-built, installable bundle of skills and instructions, some published free, that gives an agent a whole set of capabilities without writing custom prompts from scratch.
Resources

Things they pointed at.

04:04toolGmail
04:04toolSlack
04:04toolClickUp
04:04toolNotion
08:06toolReddit
11:10toolInstantly
11:50toolApify
11:50toolSales Navigator
11:40toolLoom
Quotables

Lines you could clip.

00:00
99% of my business now runs almost completely without me.
cold-open thesis line, no setup neededTikTok hook↗ Tweet quote
07:20
If an idea has been done to death, if it's saturated, it just kills it before I waste a week.
concrete payoff of an automated gatekeeperIG reel cold open↗ Tweet quote
16:40
It took the busy work so I get the judgment back.
reframes automation as freedom, not replacementnewsletter pull-quote↗ Tweet quote
17:43
Don't try to automate everything at once. You will just simply burn out and you'll quit.
direct, actionable warning, quotable as-isTikTok hook↗ Tweet quote
The Script

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
00:0099% of my business now runs almost completely without me. Now you've been seeing it everywhere lately. Everyone is talking about cloud, AI agents, automating your whole business.
00:08It's all over your feed. But have you noticed something? Nobody actually shows you what it really looks like.
00:13It's all talk, so I'm not going to talk today. I'm going to open my laptop and show you the exact setup that does 99% of my work across three different businesses without me writing a single line of code.
00:24And it's not one giant AI doing everything. It's actually a little team of AI agents. We have one manager and a bunch of specialists under it, each one running a different part of the business.
00:33Now if you run any kind of business, this is the exact thing that gives you or your team your week back. So I'll walk you through the setup, run it live on all three businesses, and I'll be very straight with you about the 1% that I would never be handing off. If you don't know me already, my name is Nick.
00:45I've spent the last three years helping companies actually implement this stuff, and I've driven over $5,000,000 in bottom line revenue across all of our clients with it. So let me show you how this works. Okay.
00:53Now this part right here, it is the entire thing. Now what we have set up, it's not one AI trying to effectively do everything. We actually have a team.
01:00So we have here one manager agent, and then we have a bunch of specialist agents underneath this. So we have one that only handles LinkedIn, one that only handles my Twitter or X, one for my YouTube, one for my community, even one for my ads. Now each one, it's just an expert at exactly one job just like a real company.
01:18Now if you see all these folders right here, each one is a different specialist on my team. So this one is my LinkedIn expert. We then have my X or Twitter, YouTube, my newsletter, my community outreach, and they don't share brains across them, and that's on purpose.
01:34So my LinkedIn guy, they aren't going to be writing in my Twitter voice. My Twitter voice isn't going to be writing in an ads voice. Each one effectively stays in its lane, so each one is actually good.
01:45And this one right here called main, this is the manager. So I'll open its rule book just so you can actually see all of it, but it's all plain English. And right at the top, it says you are the coordinator.
01:54You're not going to be writing the post yourself. You route to the specialist whose folder owns that job. So when I actually give it a task, the manager reads it, figures out who actually owns that task, it hands it right to the expert.
02:05So I never have to just remember where anything is going to be living. Now inside of this, there is literally a table in here. If the request says LinkedIn post, send it to the LinkedIn agent.
02:13If it says tweet, send it to Twitter. If it's a YouTube idea, obviously, send it to YouTube. And that's the whole manager.
02:18It's just a switchboard. On top of that, every specialist has its own rule book and its own knowledge. So if you look at here, my community agent has the entire outreach playbook baked into it.
02:28My LinkedIn agent has the whole strategy in my past posts, all my content pillars, all that stuff. So when one of them actually does a job, it's using what I actually know, not just some random stuff on the Internet.
02:39Now it also even remembers things about my business over here. So I tell it something once, maybe it's going to be something about my offer or my pricing, who I'm actually targeting, and it just keeps it so I never have to be repeating myself.
02:51And this is where, you know, other harnesses like Hermes are incredibly powerful, but I just built it myself. And then the rest of these are the tray ins and the out trays. So research comes in, finished work is going to be landing here.
03:04So we have these scripts, we have posts, we have outreach emails, we have even landing pages, we have reports, and the actual output of three businesses is all sitting in plain text files in one place. So that effectively is the whole system.
03:18There's no app that I have to log in to. All of this is completely no code. And we have a manager, we have a team of specialists, and we have the text files.
03:26And this exact same setup runs all three of my businesses, so my content, my community, and my agency. Now beyond that, we could be deploying this on an operating system, which is a very popular thing that we've been seeing as of late, where it's just going to be an interface where it's a little bit easier to manage, but I prefer to just utilize and interact through everything inside of my IDE.
03:46Now more than that, all of this can be living inside of the cloud application, and that's really just gonna be for a couple of things like connecting your other applications or grabbing prebuilt skills. Like the app, it's honestly the easier way to be utilizing all of this. And especially if you are a little bit more beginner, that's where I would recommend.
04:02Now just having a team of agents full of instructions, it can't actually do anything on its own natively. It needs to be reaching applications that you already use, so your tech stack. And that's what connectors are.
04:12So the actual technical name for this is MCP, but you could just completely forget that jargon. All it means is just plugging your applications into Claude. So here are mine.
04:21I actually have Gmail. I have Slack, my project board in ClickUp. I have Notion, my calendar, even my call recorder.
04:27So the more that you plug in, the more that your team can actually do. So watch how simple this is. I can simply just tell Claude to drop a message in Slack and it just posts it.
04:36So it's not gonna be pretending. It just runs off. It's gonna be processing.
04:40And then in a second, it actually runs it. That's the whole idea with this. You connect applications that you and your team live in every single day.
04:47And once your agents can reach your stuff, they stop being a chatbot and they start doing the work for you. Now one more piece to this, and it's actually a shortcut. You might be thinking to yourself, like, do I really have to build every one of these agents myself?
04:57And you absolutely do not need to. So over in the application, there's actually a plugins tab. Now if you navigate to this, the plug in, it's just gonna be a bundle of skills somebody already built and has packaged up.
05:07So instead of making each one yourself, prompting your agents, and providing all this information, you just install one plug in and you get a whole set at once. Now a bunch of these, they're actually made by Anthropic themselves. So this is the company behind Claude, and they're actually completely free.
05:19So look here. We have one for sales. We have one for marketing, customer support.
05:23We even have finance. So hypothetically, you run a small business. You grab the one that fits your particular business.
05:29You install it and instantly you have all those skills without writing one yourself, having to prompt anything. So before you build any skills or do anything from scratch, just check if a plug in already exists. Half the time, it does.
05:40Alright. Now let's get into some demos. So let me just run one live so it actually clicks for you.
05:44First business is going to be my content setup. So this channel, my LinkedIn, my Twitter, my whole media brand effectively, there's always a next video, a next post.
05:53So here's the real problem that it solves. I can't just film whatever I feel like. Every idea has to get researched first so I don't burn a whole week on a video nobody's going to be searching for.
06:03That used to eat half my day. So what I've done is I just built my YouTube agent a skill for exactly this and I just turned it into a simple command. In this part, it's actually the whole trick, so it's just gonna be plain English.
06:14But look at the steps here. First, go check how many people already made a video on this, and then check it against my checklist. Is it worth my time?
06:22Are people actually searching for this type of concept? And then write the full brief. So give me the angle, the three title options, the thumbnail, and then the last step, drop it onto my ClickUp board in the review column so I can just come take a look at it.
06:35So I'll just type one word, and then I give it the idea right here. I'm just gonna give it something random about local AI models, and then we could watch it run. So look at this.
06:42It's doing exactly what those steps had said. It's checking how saturated this topic actually is. It is then going to be scoring it, and then it wrote out the brief.
06:52And the payoff with this is it didn't just hand me text in a chat. It dropped a finished, a scored idea straight onto my project board in the review stage exactly where I work from and where my team also works from as well. So, ultimately, it was able to read my research, follow the skill step by step, and it used my ClickUp connector to just drop the result where I actually needed to go.
07:11So the manager, then the specialist, and then the connector all working together in one go. And one of the best things about this system is it actually tells me no.
07:18So if an idea has been done to death, if it's saturated, it just kills it before I waste a week or if I waste any resources paying my team to work on a video that isn't going to do well at all. And then on the flip side of things, so my LinkedIn agent, my Twitter agent, it each runs off of the same content brain in their own voice.
07:34And once a week, a separate analyst agent, it just grades every post that I put out, tells me what worked, and then it just places bets on what to actually try next. So my whole content operation, it quite literally runs like a science experiment now, I barely touch it. By the way, the exact skills and all the resources and guides of everything is going to be inside of my free school community.
07:52Link will be down below in the description. Sure to check that out. And one of the more important pieces is actually staying on top of the news, and this one's a bit different.
07:58It runs completely without me even being there. So, of course, I make content about AI. I need to know the second that something big drops like a new model, a new tool.
08:06Just because being early, it's half the battle, but I can't spend all day refreshing Twitter and just Reddit, so I actually automated it. So see this channel right here? Every single morning before I am even up, my setup has already gone through the AI news.
08:19So it's going through Reddit, the sources that I actually follow, the releases, and it just dropped a summary of what actually matters right here in Slack. I never touched it. So instead of chasing the news, it's waiting for me when I actually wake up.
08:31I scan it, and if something is worth a video, I run that concept command right in here I'm already had. And with this, you can take almost anything in here, and I can put it on a timer. So inside of the Claude app, there's actually a schedule.
08:42So I tell it, run this task at this time every single day. And that new scan, it fires every single morning all on its own. And as long as my computer is actually on, I wake up to the work already done.
08:53Now there's even a version of this that runs in the cloud so it works when my laptop is closed, but the simple one is plenty to start. And that's the real trick with this.
09:01So all of the repetitive stuff that really never ends, like watching the news, for example, I hand it off completely and it just runs for me. That's the whole job, all gone. So on top of this, I have it run every 30 because, you know, Cloud and Thropic and all these other AI harnesses are being released at the speed of light.
09:16So I constantly just have to be in the nose. So with this, I have a twenty four hour notice where I'm going to just get posted some sort of news every single morning before I wake up and I'm also going to get all of the most important news. For example, if Fable five or any other big release drops throughout the day, I'm going to get notified.
09:35We're just going to do a quick search every 30 minutes for the most important things like that. Now moving on to the third thing, this is my outreach. And this powers two of my businesses all at once, so pay attention to this.
09:45So my second business, it's a community. So it's called the AI accelerators. If you're not already inside of it, highly recommend to go check that out.
09:50We have about 20,000 people in there. Then my third business, this is Reprise, my automation agency. So both of them may live or die in the same thing, is just reaching out to the right companies and starting conversations.
09:59And I will be honest, I hate doing that manually. It is just a complete waste of time doing that manually. It's the kind of thing that you put off for weeks as well.
10:07So I decided that I don't have to do this. I have an agent for this. And here's the wild part is it is the exact same client acquisition system that I hand to my students.
10:15So what runs my agency's outreach is the product. So look what this can actually do. So first, it can find leads.
10:22It can write personalized emails. It can build the sequence. It can even write the proposal, and it's just a whole operating system for landing clients, and it's all in plain English.
10:30I didn't have to write code for any of it. So I could throw something like this in, find leads in my niche, and then draft the cold outreach emails. So, obviously, just one line and watch the tools that it is actually pulling in.
10:41So first, it is going to be finding the actual businesses that fits my ICP. So it's not just getting any random names, it's getting the ones that match who I actually sell to. And then from there, it writes each email personalized to them in my voice using my real format.
10:56And on top of that, I built a hard rule into this agent where it's only ever going to use a real verified email, and it'll never just guess one. So here's the result that it came back with.
11:06It's all just sitting in my Gmail, finished emails, in my tone, ready to go, and I never opened a single app myself. And the agent just did it through the connectors that I set up at the start. Now beyond this, I have it actually import everything into instantly, which is my mass email campaign.
11:20And then usually what I do is something called a Dream 100 where if there is whale clients that I really want to be working with, that's where I use this system where I'll also create a Loom video and send it off to them. But in any case, this system is capable of scraping the leads from Appify, scraping leads from Sales Navigator where I plug it into something called Vane, but that'll be a completely separate video.
11:40Don't wanna dive too much into that. But it'll literally find all of the leads, do the research on them, create the email, plug them all into instantly ready to send.
11:49And with this whole management system, so the lead finder, the email writer, the the tracker that my students use to see their own progress is inside of Claude, and I built this in just a few hours. That used to be an entire software product or having to be, like, paying somebody $50,000 just to set this up.
12:04And this fourth one is how I launch anything new. So whether that's going to be a new offer, a new campaign for Reprise, or just a new angle for my community, every launch, it needs the same pile of assets which is generally going to be ad scripts, ad copy, landing pages, the hooks, all the funnels, all that stuff.
12:22So the old way of doing this, that quite literally was a week with a copy writer or a whole day of just me staring at a blank document with every single campaign. So I happened to just build an agent for it. And the reason that its work sounds like me and not just some generic template is that it's got my whole go to market brain all baked in.
12:40So all of my frameworks, my past winners, my offers, all of my YouTube videos, so how I just generally talk. Let's go over here and navigate to our go to market agent.
12:50And I'm just gonna type in I have a new campaign for this AI offer. I need you to write the ad video scripts in the landing page angle. So from here, I just tell it to campaign and it pulls from quite literally everything that it already knows about my businesses and it writes the whole set for me.
13:03So it'll come up with the ad scripts, the landing page, the hooks, and in a couple of minutes, it'll be completely done. But typically, wanna be a little bit more specific about what we wanna start with and what we need at the front. So generally, we wanna come up with our outline and the concepts and what our plan actually is before we get into building all the ads and all that stuff.
13:20So we don't wanna move too fast and, you know, just jump into having it build things that we don't need quite yet. And my favorite part about this is that before I spend a single dollar on ads, I don't guess whether the copy actually works. I actually have a second agent run it through a focus group.
13:33So we have 13 AI personas that are exactly my target customer and they just literally tear the copy apart. We then have a copywriter agent rewriting all of this and the scoring agent is going to be picking the winner. So I get real reactions before I ever pay for a click.
13:48And that's the whole point with this. It's not going to be spitting out some random marketing. It's actually using my real playbook and then it pressure tests all of this against my real customers before it costs me anything.
13:57So we give it any customer data, any past conversations, past interactions, it does research on the internet going through Reddit, different blogs and, you know, really anything that it can find online, we pass to it. And that right there, that used to be a full team and now it's just a few minutes and I'm launching campaigns that look like an agency built them for $20,000 because in a way, one actually did.
14:17So what I just walked you through, the idea research, the news feed inside of Slack, the LinkedIn and Twitter posts, the outreach, the campaign launches. Those are honestly just a few agents running my entire business. There's a bunch more doing just little jobs in the background constantly.
14:31But the part that I want you to take away is underneath all of this, they are all the exact same idea. So everyone is just a plain English file, a specialist with a job like I showed you at the start. So now what makes them actually run?
14:42Well, there's actually only three ways. And once you've got these three, the whole thing is going to be clicking for you. Number one, it's going to be manuals.
14:49So I just trigger it myself like when I type that concept or command or kickoff campaign. So some things I just want to start myself. Number two is going to be a webhook.
14:58So something comes in and that kicks off the work all on its own. And the clearest example that I can give you is an email. So a certain kind of email lands and the second it does, an automation fires and it drops the reply before I've even opened it.
15:09And maybe there's gonna be some filters on what type of email it is, maybe it's a query inside of a subject line, whatever. Number three is the schedule. So that's the morning news summary that I gave you earlier.
15:19It runs on a clock, same time every single day or whenever you really specify you need it to be running, and you don't have to be babysitting it and just running it manually. It'll just run all automatically for you. And that's the whole picture with this.
15:29We have a team of simple agents and three ways to be running them. And once you start seeing your own work in those three buckets, you'll know exactly how to automate any piece of it. Now this next piece, do want to be honest because that 99% number, it does get thrown a lot and it is usually nonsense.
15:44So here's what I actually mean when I say 99% of the boring tasks are automated. It's not necessarily 99% of the thinking.
15:51It's 90% of the grunt work. So the research, the drafting, the formatting, the outreach, that is the part that it takes right off of my plate. But I still pick which videos to be making.
16:00I still am going to be approving every single email. I still get on the phone with the clients myself. My sales team, they still take those sales calls.
16:07The decisions, they're effectively still mine, and I'm the one on the camera. That is not a limitation. That's the whole point.
16:14So it took the busy work so I get the judgment back. And I wanna be really clear about one thing because this is where people typically lie to you. My agency, Reprise AI, we actually build automations for other companies and the work, the real client builds, it's not just done by Claude.
16:30It does have a big hand and a big play into that, but I have a team of human engineers who do all of that. So Claude just runs everything that gets us the client and it markets us like the content, the outreach, the go to market stuff, but the actual building for a majority of it, the thing that a client is gonna be paying for, it has a human on every step.
16:46And I'd never typically would ship that unsupervised and neither should you. And with that, it's not completely magic, it does break at times.
16:53So all my agents, they just read from a knowledge folder. They're brain effectively. And not too long ago, I forgot to update it and it was just quietly working off of old numbers for a specific job at hand.
17:02So it wrote me a script built around a stat that was just months out of date and it sounded completely confident about it. And I did not catch this until I actually read it back. And the fix is very simple.
17:11Now it just refreshes that research first, so every time before it writes a single word. But that right there, it just taught me the real rule is that this is only ever as good as what you feed it you still have to look at what it gives you. So you do still need that human in loop for, I would say, the most crucial, very important things.
17:27So no, I'm not telling you to set this up and just walk away forever. I check it myself. But once the pieces are actually right, it runs for you and I'm just reviewing finished work instead of just doing everything from scratch.
17:37So if you're in any kind of business and you're thinking this is a lot, here's the thing that you need to do right now. Don't try to automate everything at once. You will just simply burn out and you'll quit.
17:46Just pick the one thing that you do every single day that you hate. Now for almost everybody, that's email. So start there, automate your emails first, and once it's running and you trust it, you'll naturally see the next thing and then the next thing.
17:57But for anything more beyond that, check out our previous videos on how to automate your entire business. I'll have a link for that down below in the description. But with that being said, that's really all this is.
18:05You find the most annoying task, something that's very annoying, monotonous, and takes your team a lot of time to actually perform, and then you turn that into an agent. And then a few agents, it becomes a team, you give them a manager to actually route the work, you connect a couple of your applications and it just starts running end to end for you and your team.
18:22You put it on schedule, and now it's working while you sleep. It's the same few pieces, just all stacked up. And that's how one folder actually ends up running one entire business and then three for you.
18:31Now I showed you this in the simplest way that I could. There's a lot more that you can be doing with this, way more you can be automating, but none of it is hard. It's the same handful of pieces just repeated.
18:41And what you saw today, that's just the foundation. Everything else is just building on top of it. So if you wanna actually learn Claude properly from the ground up, I made a full beginner guide that goes way deeper inside of my community.
18:52Look at me down below in the description. And like I said, the exact agents and the skills from this video will also be inside of there. And if your business genuinely needs built and you'd rather not take the time to learn this and do it yourself and you just want it done right, that's the work that we do at Reprise.
19:04So link will be down below to book in a call with our team. We'll jump into an audit, show you where and how we can increase your bottom line, and drive leverage inside of your team. So we're gonna call down below if you're interested and looking to grow in 2026.
19:16Thank you guys for watching. I'll see you in the next video.
The Hook

The bait, then the rug-pull.

Nobody actually shows what "AI runs my business" looks like on screen, so this one just opens the laptop: a single manager agent, a folder of narrow specialists, and three real businesses running off the same system.

Frameworks

Named ideas worth stealing.

00:52model

Manager + Specialist Agent Architecture

  1. Manager agent (router)
  2. LinkedIn specialist
  3. Twitter/X specialist
  4. YouTube specialist
  5. Newsletter specialist
  6. Community outreach specialist
  7. Ads specialist

One coordinator agent reads every incoming task, decides which single-purpose specialist folder owns it, and hands it off — each specialist keeps its own rulebook, brand voice, and memory, isolated from the others.

Steal forany multi-channel content or client operation with more than one recurring writing job
14:40list

The Three Trigger Types

  1. Manual — you type the command yourself
  2. Webhook — an external event fires it, e.g. an email lands
  3. Schedule — it runs on a clock, e.g. every morning

Every automated task in the system starts one of exactly three ways; sorting existing work into these three buckets reveals what's actually automatable.

Steal forauditing which recurring tasks in any business are ready to hand off to an agent
09:39list

One System, Three Businesses

  1. Content (YouTube, LinkedIn, Twitter, newsletter)
  2. Community (AI Accelerators, ~20,000 members)
  3. Agency (Reprise AI, automation consulting)

The same manager-plus-specialist folder system runs all three businesses at once — the outreach agent alone powers both the community and the agency pipeline.

Steal forfounders running more than one revenue line who want one shared operating system instead of three separate ones
CTA Breakdown

How they asked for the click.

VERBAL ASK
17:42link
Book a call with our team... link down below to book in a call with our team

Closes with two parallel CTAs — a free community link for viewers who want to build it themselves, and a booked audit call with the paid agency (Reprise) for viewers who want it done for them.

Storyboard

Visual structure at a glance.

open
hookopen00:00
agent folders
valueagent folders01:32
concept command
valueconcept command06:35
drafted outreach email
valuedrafted outreach email11:02
go-to-market campaign
valuego-to-market campaign12:27
community CTA
ctacommunity CTA18:51
Frame Gallery

Visual moments.

Watch next

More from this channel + related breakdowns.

Chat about this