The argument in one line.
A single AI agent handling every task loses context and starts making mistakes, so the fix is a boss agent that delegates each narrow task — upload, transcribe, post, research — to its own single-purpose sub-agent.
Read if. Skip if.
READ IF YOU ARE…
- You're a solo creator or small-business owner already using Claude Code or a similar AI coding tool and want to automate repetitive content or ops tasks.
- You publish the same content across multiple platforms (Instagram, YouTube Shorts, TikTok, X) and are doing the repurposing and posting by hand.
- You've tried giving one AI assistant too many jobs at once and watched it lose track of details or make avoidable mistakes.
- You're comfortable connecting third-party services (Tailscale, Buffer, API keys) to get automation working, even if you're not a professional developer.
SKIP IF…
- You want a plug-and-play SaaS tool — this setup involves SSH, API keys, and a beta app with no public signup yet.
- You need production-grade reliability guarantees — the builder demonstrates only a first test run, not a proven long-running system.
The full version, fast.
The video argues that AI agents fail when one bot is asked to do everything at once — it loses context, forgets details, and breaks the project. The fix demonstrated is a "boss bot on top" pattern: one coordinating agent that spins up narrow, single-job sub-agents (uploader, transcriber, poster, content strategist) inside a Claude-Code-based builder app. The builder connects to the creator's Mac via Tailscale and SSH to pull raw video from an iCloud Drive folder on a schedule, and to Buffer to auto-post across Instagram, YouTube Shorts, and X using pasted API keys. A research sub-agent is instructed to only surface real, verifiable viral videos in the creator's niche rather than inventing examples. The whole pipeline, including scheduling a daily cron job, was set up in roughly twenty minutes.
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 – 00:45
01 · The pitch: hand grunt work to AI
Creator introduces himself as scaling a business to $100M and framing AI agents as a hiring replacement.
00:45 – 02:08
02 · Why one bot per task
Explains that overloading a single agent with every task causes it to lose context and make mistakes; raw Claude Code subscriptions cap at 50 bots and are tedious to extend.
02:08 – 04:32
03 · Spinning up the content-boss bot
Creates a new project and a "Content Boss" agent inside the Jarvis builder, then tells it in plain language to build an uploader, transcriber, X-poster, and plumbing-niche content strategist.
04:32 – 06:10
04 · Touring the existing bot fleet
While the boss bot works, the creator tours his live personal-brand and business-strategist agent trees (content poster, comment/DM reply, YouTube editor, clipper, thumbnail creator, web designer/image generator).
06:10 – 07:07
05 · The pipeline plan comes back
The boss bot returns an uploader/transcriber/poster/strategist pipeline design pulling from a MacBook iDrive folder and posting to Instagram, YouTube Shorts, and X.
07:07 – 09:03
06 · Connecting Tailscale and SSH to the Mac
Sets up Buffer access, then installs and configures Tailscale so the agent can SSH into the MacBook and reach an iCloud Drive content folder.
09:03 – 10:30
07 · Confirming the sync and real research
The bot confirms it can read the Mac's files and asks for account details; the content strategist is set loose to find real viral plumbing videos with no invented links.
10:30 – 12:16
08 · Wiring up Buffer to auto-post
The boss bot reports the pull pipeline is live and tested end-to-end with a daily 8am schedule; creator connects new channels in Buffer using pasted API keys.
12:16 – 13:17
09 · Wrap-up: twenty minutes, live
Full pipeline (cron job included) took about twenty minutes to assemble; closes with a CTA for comments and a beta-access tease for the Jarvis app.
Lines worth screenshotting.
- Feeding one AI agent every task — finances, content, scheduling — makes it lose context and start missing details; the fix is one job per agent.
- A coordinating "boss" agent that delegates to single-job sub-agents avoids the context-overload failure mode entirely.
- A raw Claude Code subscription caps out around 50 bots and requires manual setup for each new one, which discourages spinning up narrow agents.
- Tailscale plus SSH lets an agent pull files directly from a personal Mac's iCloud Drive folder without exposing the whole machine.
- A file-sync agent that only copies (never deletes or moves) source files avoids destroying the creator's original footage.
- Buffer's API key connection makes auto-posting to a new social account a roughly ten-second setup once the account exists.
- Explicitly instructing a research agent to skip unverifiable links, rather than guess, prevents it from fabricating viral video examples.
- A single scheduled pull (here, daily at 8am) is enough to turn a manual repurposing habit into a fully automated pipeline.
- Testing automation end-to-end — dropping a real test file and confirming it lands — is treated as the proof step before trusting the pipeline, not the build step itself.
- The entire agent fleet, including the daily cron schedule, was assembled in about twenty minutes once the builder app and integrations existed.
One job per bot beats one bot for every job.
WHAT TO LEARN
AI agents fail the same way overloaded employees do — cram in too many responsibilities and they lose the thread — so the fix is delegation, not a smarter single agent.
02Why one bot per task
- An AI agent given every task at once — finances, content, scheduling — will lose context and start missing details; splitting tasks across single-purpose agents avoids this failure mode entirely.
- A coordinating agent that spins up and manages narrower sub-agents scales further than trying to make one general-purpose assistant handle everything.
06Connecting Tailscale and SSH to the Mac
- Connecting an agent to a personal computer via a secure tunnel (Tailscale + SSH), scoped to read-only file access, is enough to automate a daily file-pickup task without exposing the whole machine.
- A sync process that only copies files — never deletes or moves the originals — removes the biggest risk of trusting automation with your source material.
07Confirming the sync and real research
- Instructing a research agent to skip anything it can't verify, rather than guess, is what keeps automated research trustworthy instead of just plausible-sounding.
- The real test of an automation isn't that it was built — it's a live end-to-end run: a real file dropped in, confirmed pulled, confirmed cleaned up.
09Wrap-up: twenty minutes, live
- Once the scaffolding (agent builder, file access, posting API) exists, wiring up a new automated pipeline is a matter of minutes, not days.
- Cross-platform posting tools that connect over an API key turn a manual multi-platform publishing chore into a single paste-and-go step per account.
Terms worth knowing.
- Sub-agent
- A narrowly-scoped AI agent assigned exactly one job (e.g., transcribing video) that reports up to a coordinating "boss" agent, rather than one agent trying to handle every task.
- Tailscale
- A mesh VPN service that lets devices — like a personal laptop and a remote agent — securely reach each other over the internet as if on the same private network.
- Buffer
- A social media scheduling tool that connects to platforms like Instagram, YouTube, and X via API key, letting a third-party system post to those accounts automatically.
- SSH
- A secure remote-access protocol that lets a program or agent log into and run commands on another computer over the network.
- Content-boss pattern
- An agent architecture where one top-level agent is in charge of a domain (e.g., content) and creates/manages narrower single-purpose agents underneath it.
Things they pointed at.
00:45toolClaude Code
00:45toolJarvis (agent-builder app, beta)
10:05toolApify
Lines you could clip.
01:00
“Anything you do with AI, you want to make sure you give each bot as little as possible to do.”
01:40
“You need to separate each individual task to one bot at a time.”
12:25
“The whole system is then set up, cron job, everything with this app. It literally took me, like, twenty minutes.”
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.
metaphor
00:00So if you are a business owner and you wanna scale your business, but it's difficult to find people to do different tasks, I have the solution for you. Quick introduction. My name is Shah.
00:09I'm trying to scale a business to a $100,000,000, and on the way, I am building automations with AI because AI is the future. And I wanna get as much as possible being done by so that I don't have to worry about it, so that I don't have to worry about hiring people for it and training them and seeing the human errors and then constantly fixing that.
00:29I can just work with AI. It's way smarter and it's actually able to do a lot more than you think. So I wanna show you exactly how I repurpose my content using AI and how it's able to then, you know, post across multiple platforms, do it automatically without me worrying and doing it consistently.
00:45So you're looking at my screen now, you can see I got this Java interface. This is basically just Claude code, uh, that I kinda plug my subscription into here so that I can access the tools that it has, which is the main tool I will show you is making these sub agents. Anything you do with AI, you wanna make sure you give each bot as little as possible to do.
01:08The problem with bots is that, and a lot of people get this wrong, is that they think they can just get one bot and feed it everything, do the finances, do this, do that, do that, do that, do my content, and then the bot loses context. It only can handle very little in its main files so that it can execute at a high level.
01:26Once you shoved too much in there, it forgets the details and then it has too much like it has too much to read.
01:33It was like, oh, it's kinda browsers through it. It's like, okay. This is kind of how I have to do it.
01:36And it forgets the details and it messes up the project. And everyone's like, oh, I hate AI. No.
01:41You need to separate each individual task to one bot at a time.
01:47And it's quite difficult when doing it with a normal closed subscription. People just have one project file. If you have the bots in Discord, I used to have all the bots over here.
01:54You can see each one of these are different sub agents, but you can only put 50 bots in here, and it's so difficult to make a new bot every time. You just had an idea. You wanna make a new bot.
02:05You have to go through all these steps. It's quite tedious. So this program over here, really solid because then I can just come in over here.
02:12For example, let's make a new project and let us call it the AI automation, uh, content automation.
02:24It'll make me a new project file, and then I can come in here and actually name, uh, .
02:34I like making little boss file files on top of it, uh, because these guys are super, uh, this this isn't in who's in charge of the whole content system. Right?
02:45So I got my content boss, and now I can add little agents underneath him. K. You would see how easy it was to make a whole new agent.
02:51He comes with his basic understanding of how to run as a, like, a top tier boss, um, and it's prebuilt. Like, bot is not just fresh with clear MD files.
03:02They have some basis basic, uh, skills in here, which make it really easy to use. And of that is how to control my Chrome, how to do research, all these things that knows already.
03:12But yeah. So we're coming here off as a content boss.
03:16We can now add bots underneath him. And, uh, the best way to actually do this is actually talk to him.
03:24Hey. I wanna automate my, uh, content.
03:27Can you go through and build out a system of bots? I need to take the videos from my I drive on my computer, my Mac, uh, my MacBook, and then upload it to Instagram and YouTube shorts.
03:46I then also need to repurpose those videos. So I want you to download them, transcribe them, and understand what that video is about.
03:54And then I want a bot to then, obviously, post to x, but post it as threads and articles, you know, to x.
04:03Can you please build out the the the team of bots to be able to do this? And then I also want a content strategist who can do research for me and is able to give me feedback on what type of videos I should make, uh, on my niche.
04:18I am in the plumbing business right now. I wanna, you know, do more. I I mean, I'm in the pub I'm in the plumbing business.
04:26I need ideas and final videos for plumbing.
04:32Boom. Sent it to the bot, and now it will pull out this fleet. Let's see what it says.
04:39Now while he's thinking, I'm gonna show you exactly what I'm already doing. So I got my personal brand of the year. You can see I got the PB Boss on top, and I got content charges to post, a comment reply, a DM reply, a YouTube editor, the clipper who takes a YouTube videos and turns it into clips to post to YouTube and in the YouTube shorts and Instagram and TikTok, and I've got my thumbnail creator.
05:00So this is my whole content team. Um, and I've trained each individual bot to get the outcome that I want. You can even have more abstract ones like, uh, this over here.
05:10You can see I've got this whole new business model over here. Then I got the job as, you know, content manager. He gives me content ideas from the content strategist and then also handles of posting and everything.
05:21This video should be edited by this team over here and then, uh, posted across all platforms. And then we've got the business strategist. Housewing has all the Homosey books.
05:30And if you could see all three Homosey books in there and a bunch of the I think 700 of his videos, uh, transcribed and key lessons put in there. And then I got, like, the web designer, and then I got my image generator that runs through web designer, and this guy just helps me with building.
05:46So this is a whole team, and they're all connected throughout different APIs.
05:52It can control my Chrome. So really, really smooth. Let's see what this guy is saying.
05:58Can see it's updating who you are section, FloorMd. It's giving them the all their roles and training them exactly on what they need to do. Muted it.
06:06I should've actually had it talk back to me, but really nice outcome. So he's a content boss needing content automation team.
06:14He's got the uploader, the transcriber, the exposer, and the strategist who's doing plumbing niche research, file video ideas, feedback on how to make them. So we got the MacBook iDrive video folder, uploader watches for new videos, uploader then post it to Instagram and YouTube shorts, drops the videos to the to process folder, bang, transcriber, transcript, plus one paragraph summary, which then goes feeds to the poster, x to it, and x long form articles, bang, and then go to a strategist who runs in parallel, weekly plumbing research, and ranked video ideas plus feedback.
06:59So it needs the full access of video access. Doesn't run-in my MacBook. I can show you exactly how we can SSH into my MacBook right now, and then it can control everything.
07:10And then we've got Instagram, um, and it's buffer.
07:15So I'll show you what's buffer, how to set it up, um, and this is how I repost.
07:21I got all these random accounts here on buffer, and, uh, you can then connect the accounts and then post to them using the API. So you just give it the API key and then bang and post to Instagram.
07:34Takes, like, ten seconds to set up. I'll show you how I do it. And we got YouTube, which channel, and set set up the YouTube API.
07:45Or you can use log in Chrome, which is cool because then you can just run the Chrome, drive my Chrome tabs, and set it up. Now we don't actually need to connect it to the computer because we don't need to, uh, scrape anything or do anything.
07:58Uh, right now, the main thing is just giving it access to buffer, and then actually, we do need to connect it for the sake of our files. So if you wanna get it connected to the computer, you wanna go to Tailscale.
08:10Tailscale, set up an account over here. Once you're in Tailscale, you can come into the bottom over here and add your devices over here.
08:20So I go to Mac one. Bang. How to install onto Mac?
08:24All you gotta do is download this thing over here. Bang. I downloaded the macros.
08:31Package. You'll see here you get this type of thing. Click it.
08:38Continue. Install. And then, yeah, you can just install it.
08:44I already have it installed. Once you get it open, you can come in over here, and it will give you this window over here that you can start.
08:53It'll tell you exactly what you need to do. And then you'll get this little this you'll get this setup with your MacBook Air connected, whatever.
09:02Screenshot this and just send it to the bot. So I just screenshot, bang.
09:08The strategy is already came back with brief. It's fire. But let's send the tail scale.
09:14It can come on over here, control v, and then so let me know how we can set up tail scale or, I mean, SSH into my computer.
09:24I believe I already have tail scale set up. I just want you to SSH into it to then be able to pull from the files, go to my iCloud drive folder, and then get the make a new file in there for content.
09:39And then I will upload my videos into there every day. And you need to pull from there every day and upload to my account. K.
09:48And then while it's thinking, let's see what it may about the content strategist. So there's some interesting ideas. I just wanna make sure that it used research to come up with these ideas.
09:57It needs to down Luke used Applefy to connect it up.
10:02So you can check the Applefy.
10:06K. Connect up up Applefy to it. Give it the your API key from there.
10:11Yeah. And then they just need to search on x, uh, TikTok, Instagram to find viral videos for you. And then you know you have solid ideas that you can work off of.
10:20So let's ask it. How did the content surges come up with those content ideas? Can you send me the links of the videos that you found?
10:28The viral videos? Done.
10:30I'm in, and the video pipeline's front half is live and tested. Here's what I set up. I s x h eight into your MacBook through Tailscale, and it works.
10:36I already confirmed I can read your files. I made you a folder for this. Open Finder, go to iCloud Drive, and you'll see a folder called plumbing content.
10:43Drop your plumbing job videos in there. That's the only thing you have to do each day. I tested the whole thing end to end.
10:48I put a test video in that folder, pulled it down to the fleet, confirmed it landed, then deleted the test. So the pull is real, not just built. I also scheduled it to run automatically every day at 8AM.
10:56It grabs whatever new videos you dropped and stages them. Only It copies it never touches or deletes your originals. One important note.
11:01Right now, can pull your videos, but I can't post them yet. To actually upload to Instagram and YouTube shorts and to post threads and articles to x, I need you to tell me the accounts. Specifically, three things.
11:09One, your Instagram handle for the plumbing content. Two, your YouTube channel. Three, your max handle and whether you have x premium because long form articles need it.
11:17And the strategist still wants two things to sharpen your video ideas, your city for local hashtags, and your booking link or phone number for the call to action. So next move is yours. Drop a few real videos in that plumbing content folder to test it live and send me those account details and your city.
11:30I've got the strategist now finding real videos right now. It's searching the web for actual viral plumbing clips, and it'll come back with real links, the creator, the platform, rough view counts, and why each one worked. It should land in a few minutes.
11:40I'll send you the links as soon as it's back. And to be clear, I told it not to invent any links. If it can't verify a video is real, it leaves it out.
11:47Beautiful. Beautiful. Beautiful.
11:49So just like that, we have Buffer. We just go set it up. It added the fucking thing directly to my folder.
11:55You saw that. It's in it is plumbing content. Bang.
12:00I can go on my phone right now. And if I go into I can take my videos that I make TikToks with, upload it directly to my drive in my files folder, and it will show up here.
12:12And I can pull it from here and upload it wherever. So smooth as shit.
12:17In terms of the posting, I already have my account set up, so I'm gonna, like, go and do that. But because you have all these accounts here, you can just go in, press new, connect a new channel, and then go ahead and set it up.
12:30Simple as shit. And once it's in, you say, I wanna post it. I wanna give it the app the API key.
12:37You can find the API key over here, bang. API keys. Hopefully, mine doesn't show.
12:42And then you can just copy it and send it to the bot, and it'd be able to post directly to any account so that you have on that system. The whole system is then set up, cron job, everything with this app.
12:55It literally took me, like, twenty minutes. How long is this video? So, yeah, if you like that, if you wanna learn more about AI, if you want to have any questions about the details of certain things, if you're having an issue setting it up, definitely leave a comment below.
13:08And if you want the link to Jarvis and the stickers app, then let me know in the comments. It is still in beta, but I'll see if I can get a link for you guys.
Named ideas worth stealing.
02:08model
Content-boss-on-top pattern
- Boss agent (coordinator)
- Uploader (watches for new video, posts to IG/Shorts)
- Transcriber (transcript + summary)
- Poster (X threads + long-form articles)
- Content strategist (niche research + video ideas)
A single coordinating agent creates and manages narrow, single-job sub-agents rather than one agent trying to do everything.
Steal forany solo-operator automation stack (content, ops, customer support) built on Claude Code or a similar agent framework
How they asked for the click.
MENTIONED ON CAMERA
