{
  "video_id": "RHZLJhp-JMM",
  "channel_slug": "techwithtim",
  "channel_handle": "Tech With Tim",
  "title": "Paperclip Is Insane - Full Tutorial",
  "duration_seconds": 2320,
  "url": "https://www.youtube.com/watch?v=RHZLJhp-JMM",
  "upload_date": "20260404",
  "transcript": "We've now entered the era of entire AI companies.\nForget individual AI agents.\nWe've had those now for what, one year?\nWe're talking about fully autonomous companies\nthat run using multiple\nAI agents\nthat are coordinated and chat with each other.\nNow, the way this is possible is using a new open\nsource framework called Paperclip.\nNow this has been getting a ton of hype over the past\nfew weeks.\nHere.\nIt's got thousands,\ntens of thousands of stars on GitHub.\nAnd in this video I'm going to show you how it works,\nwhat it is,\nand how you can set it up\nso that you can run your own AI company.\nI've been playing around with it\nnow for the past 48 hours.\nIt's pretty cool.\nIt's already accomplished\na lot more than I would just do on my own.\nUsing something like Claude Code.\nAnd the best part is this is completely free and open\nsource, and you can configure it however you want.\nOkay,\nso I think it makes the most amount of sense for me.\nJust show you a demo of what this is\nbecause it is a relatively new tool.\nAnd then I'm going to show you how you can set it up.\nAll right.\nSo in front of me you can see my AI company.\nLet me just go to the org chart here.\nRunning inside of paperclip.\nNow I'm running this on a virtual private server.\nWe'll get to that in one second.\nBut you can see the right now I have five AI agents.\nI have a CEO, a founding engineer, newsletter\nresearcher, newsletter reviewer and a newsletter\nwriter.\nNow, the only agent that I actually created\nor gave any instructions was the CEO.\nAnd then from there, the CEO actually asked me\nbecause I'm kind of like the board\nof directors in this situation,\nif you could hire multiple other AI agents.\nAnd it decided that because I'm\ntrying to start a newsletter here with this company,\nor at least improve my existing newsletter,\nthat it need to hire these different developers\nor these different agents.\nNow, these agents could then hire their own agents,\nand you could have much more complex org charts,\nas I've seen online.\nBut in my case, five is completely fine.\nNow, what I started with\nis simply just giving this company a goal.\nNow the goal that I gave it is that I wanted\nto increase the newsletter subscriptions and revenue\nby creating a paid ad strategy, increasing click\nand open rates, increasing number of subscribers.\nWill automating the creation of newsletters.\nRight now I have someone who I pay to do\nmy newsletters and helps me out with the process,\nand I would like to just fully automate that and\nhave this start improving it by looking at my stats.\nNow you can see right now\neven I wasn't even trying to demo this.\nA new task actually just started running.\nYou can see my newsletter\nresearcher is live right now\nand what happens is every\nyou know you can set the time.\nBut I think in my case, every ten minutes\nmy CEO will spin up,\nsee what it needs to work on, and then delegate\nvarious different tasks to these other AI agents.\nSo you can see ten minutes ago, eight minutes ago,\nsix minutes ago, you can see all of the runs.\nIf you're running this with an API,\nyou can see how much money you're spending\nand see all of the tasks\nthat they've been working on.\nSo in my case,\nI've already had them analyze all of my previous\nnewsletters, give me detailed\nreports on ways that I could improve.\nI had to go and actually create optimizations\nfor my main tech with him website.\nIt already pushed a PR here\nwith a bunch of changes last night,\nand then it also created its own new repo,\nwhere it's developing its own newsletter platform\nso it can handle everything and show it to me\nin an interactive dashboard.\nNow that is, for me,\nliterally just setting the goal for the company\nand then just giving a few manual issues here.\nSo kind of the way it works is you can give an issue\nand you can assign it to a particular agent.\nIf you assign it to the CEO,\nit will just delegated out.\nAnd then doing\na little bit of configuration with these agents\njust to make sure they run a little bit better.\nBut overall, I've just been sleeping.\nI wake up and all this work is done by my\nAI company, and I\nhave a million ideas on what I can do with this,\nbecause this is already incredible.\nTechnology does require a little bit of setup,\nbut it is super, super cool.\nSo what I want to do now is\nI want to show you exactly how to set this up,\nso you can see how we start from scratch,\nhow we hire multiple agents,\nand how you can start running this again.\nThis is new tech. It's open source.\nI'll leave all the links and resources\nin the description down below.\nAnd this allows you to facilitate multiple AI agents\nusing things like clawed code, open claw, codex,\nreally any model that you want and run it either\ndirectly on your own computer,\non a virtual private server, which I'll show you now.\nOkay,\nso the setup for paperclip is very straightforward.\nYou just need to run this command or whatever device\nyou want to install it on.\nNow you can run this on your own machine.\nAnd I have seen some people doing that.\nHowever I really would recommend against that.\nThere's a lot of different reasons,\none of which is security.\nFirst of all, if you're running this,\nyou have fully autonomous AI agents\nrunning on your own device with access to all of your\nfiles, your computer, your home network,\nwhere they have full route control of your computer\nand can effectively do anything that they want.\nThat means that they can go rogue,\nthey can go to the internet,\nthey can buy a course,\nthey can access your credit cards, etc., etc.\nit also means that this device\nneeds to be running 24 over seven\nif you want it at least to continually work,\nand that you risk anything.\nLike if a fire happens, for example,\nand you lose your machine, you lose all of the work\nthat these two agents are doing, assuming\nthey're not uploading it to the Claude.\nGenerally speaking, whenever you're\nrunning these types of autonomous agents,\nwhether it's OpenClaw or whether it's paperclip,\nyou really should run them in\nsome kind of virtual private server\nor at least your own dedicated machine.\nBut if you don't want to fork out\n$1000 or $1500 for a new device,\nand especially if you're not going to be running\na local LLMs, you can get away by spending\nas little as $7 per month to run this on a hosted\nvirtual private server with a one click setup,\nwhich is what I'm going to show you how to do\nand what I personally do for all of my AI agents.\nNow, the easiest way to get this set up right now\nis actually to use the one click to play\nfrom Hostinger.\nLiterally all you have to do is just select a plan.\nYou can go to KVM one, two, three depending on\nhow many agents you think you're going to be running.\nI'll leave a link to this page in the description.\nSimply press deploy from this page\nand then Hostinger will\nautomatically deploy paperclip for you\nin a secure Docker environment.\nAnd then all you have to do is pass\na few configuration values and you're good to go\nand everything will be set up.\nSo what I'm going to do here\nis deploy a new paperclip instance.\nAgain I'm going to go with the KVM two plan.\nThat's what I recommend at least for the amount\nof space and all the stuff that it gives you.\nI'm going to press on deploy from the site again\nlinked in the description.\nAnd then what we're going to do here\nis just go through, select our plan\nand then start provisioning the instance okay.\nSo once you get to this page here\nselect the duration of your plan.\nIf you go with 12 months or greater\nthen you will have the ability to use this tech\nwith Tim discount code\nbecause I am a partner with Hostinger.\nAnd this will give you an additional 10% off\nyour plan, as you can see here.\nSo anyways, go ahead,\npress on, continue, choose whatever plan you want,\nand then choose the daily Auto backup.\nIf you want this\nto be automatically backed up for you,\nyou can also choose the location\nthat you want to deploy this to.\nSo once you press deploy,\nyou should be brought to a page that looks like this\nwhere you're going to start\nconfiguring the paperclip instance\nnow in order to sign into your paperclip account,\nbecause we are going to be hosting this\nin the Claude.\nIn a secure environment,\nyou are going to need an email and a password.\nSo you can see that we have an admin name.\nYou should just leave this as admin, but\nyou can make it anything that you want for the admin\nemail.\nMake this your own email\nand then for the admin password I would recommend\njust going with the auto generated one here\nor generating a new one that is secure.\nDon't use a weak password\nbecause when you deploy this paperclip instance,\nit will be available over the internet.\nAnd that means that you want to have secure\nlogin credentials.\nSo anyways, what we're going to do here\nis just copy down the admin password.\nSo I'll just save that myself.\nYou can access it later.\nAnd then just put one of the four values here.\nSo anthropic API key OpenAI API key Gemini or cursor\nAPI key.\nYou need at least one of these installed\nif you want paperclip just to work by default\nwhen you start running it.\nAnd this is if you're going to use the API usage.\nBut the advantage of paperclip is that\nyou can actually use your Claude code subscription\nor your Codex subscription,\nor your Gemini subscription,\nor whatever it is that you have as a subscription,\nand you can run directly through that.\nNow I am going to do a disclaimer here.\nCheck the terms of services of those providers\nbecause some of them allow this.\nSome of them do not allow that.\nSo just be a little bit careful.\nBut effectively what's going to happen\nwhen you install paperclip is that it's going to be\norchestrating agents that are agnostic\nto whatever platform they're actually running on.\nSo you could do it in Codex, you can do it in Claude\nyou can do it in Gemini.\nYou could have your open Claude agent\nactually connect to this.\nAnd that's kind of the advantage of this platform\nis that you bring your own agent\nor your own API or Lem,\nand it will just run the exact same way.\nAnd you can actually have different agents\nusing different subscriptions or different tokens.\nSo what I'm going to do is I'm going to run this\nthrough my Codex subscription\nor my Claude code subscription,\nbut if you don't have one of those\nand you want an API key,\nthen go get the API key, put it here\nand you can set spending limits for your agents\nso they don't spend too much money.\nSo anyways, for now, I'm going to press on deploy\nwithout putting in one of those values.\nBut again, I recommend you put in at least one.\nYou can remove it\nafter we're going to wait for this to provision.\nAnd then I'm going to show you the exact setup.\nOkay.\nSo it's been a minute here in the instances\nnow provisioned.\nNow once it gets provisioned\nassuming you're using Hostinger.\nBut of course you can use anything that you want.\nYou should be brought to a page that looks like this,\nwhere you see a Docker manager\nnow hosting or deploys this in a Docker container,\nwhich is one of the most secure ways to do it,\nwhich is why they do it.\nAnd which allows for the one click deploy.\nAnd what that means is that if you want to start\nmanaging this, there is a few additional steps\njust to be aware of that,\nI'm going to run you through.\nIf you're running this on your own computer\nor a dedicated machine,\nwhat you're going to do is just open up\nthe port where paperclip is running.\nIt should show you that\nwhen you go through the installation steps\nand you'll be able to see a login screen.\nSo from here, if we go to Docker Manager\nand we look at where it says open\nunder the Docker projects\ninside of the paperclip container, we can press open\nand it should redirect us to a URL here\nwhere we have an authentication screen.\nNow from this screen\nthis is where you're going to put in your details.\nSo we're going to put in the email\nI did Symantec with Tim Dot net.\nAnd then the password that we filled in there.\nNow if you forgot the password no worries.\nWhat you can do is go to manage for the paperclip.\nProject or container again to get there you go\ndocker manage your projects, we go manage\nand if we scroll down, it will show us\nall of the variables that we have in the environment.\nJust wait for it to load.\nSo if I click environment you can see that\nwe have the admin password.\nWe can view it and then we can copy it\nand we can paste it in here okay.\nSo now what we're going to do is sign in.\nAnd from here let's close this.\nIt's going to ask us to start making our company\nnow before we can make the company\nwe need to make sure that we have some kind of agent\nthat we can actually use on our own computer.\nThe reason for that\nis that we need to connect some agent\nin order to connect some agent,\nwe need some kind of LLM provider.\nRight.\nSo if you entered one of those API\nkeys, you're good to go.\nAnd you can just select to use open AI\nor to use Claude or whatever API key you put in.\nBut if you didn't\nand you want to use your subscription for Codex\nClaude code, whatever,\nyou're going to have to sign into that account\non the virtual private server\nor on the machine that you're setting this up on.\nSo what I'm going to quickly show you how to do\nis SSH into the server and just sign in\nso that we connect Claude and we connect Codex\nor whatever it is that we want to connect.\nAnd then once it's connected,\nwe can go through this process.\nIf you already have the API key,\njust follow through this prompt.\nAgain, you can fast forward.\nI'll go through it and show you what to set up. Okay.\nSo what we're going to do here\nis go back to overview from our hosting page.\nNow what we want to do\nis we want to access the server as a root user\nso that we can install what we need\nto install and sign into the accounts.\nSo in order to do that we're going to copy this\ncommand right here which is SSH root add.\nAnd we're just going to open up a new terminal.\nNow this step should be the exact same\nif you're on Mac or Linux okay.\nSo we're going to open up a new terminal from here.\nWe're going to paste the SSH command.\nAnd we're going to hit enter.\nNow it's going to ask us\nif we want to connect to this host.\nWe're going to go ahead and just type yes.\nNow here it's going to prompt us for a password.\nSo what we can do is we can change the root password\ndirectly from Hostinger\nor whatever provider you're using.\nSo I'm going to go change\nI'm going to generate a new password.\nAnd I'm just going to make sure that I copy it\nto my clipboard.\nSo generate copy okay.\nAnd then confirm the reason you need to change\nthe root\npassword is by default\nit's set to something random that you do not know.\nSo you need to change it to something\nyou do knows that you're able to sign it now.\nOnce it's changed, it should be updated\nand we should be able to use that to sign in.\nHowever, if it doesn't work, just try the SSH\ncommand again and then paste the password.\nNow when you're pasting the password,\nnothing is going to appear in your terminal.\nIt's going to be completely blank.\nSo just it's fine.\nPaste it and then hit enter.\nI just want you to be aware of that.\nNow when I say ssh, essentially what this is doing\nis accessing\nthis remote computer\nand allowing us to run a few commit.\nOnce we run this, you won't need to do this again,\nbut it's just a good practice to be aware of\nand I know it can seem a little bit\nconfusing or intimidating, but trust me, it's not.\nYou're literally just accessing\nanother computer and running a few basic commands\nto sign in to your accounts.\nOkay, so now that we're inside of here,\nwhat we need to understand\nis that everything that's running for paperclip\nis running inside of a Docker container.\nThat means that if we want to access the data\nor we want to sign in to our account,\nwe need to do that inside of the running container.\nSo the way that we can do this is the following.\nWe can type docker ps.\nDocker ps is going to give us a status of all\nof the containers that are running on our machine.\nNow you'll notice that we have one\nthat says paperclip okay.\nAnd we have a container ID.\nSo we're going to copy the container ID\nby hitting Ctrl C on our keyboard\nor right clicking and pressing copy.\nAnd we're going to run the following command\nto get inside of the Docker container.\nAgain\nI promise you, once we do this one time we're done.\nAnd again, if you have an API key\nthat you'd rather use, you don't need to do this.\nOkay.\nSo from here we're going to type Docker exec dash it.\nWe're going to paste the container ID\nand we're going to type slash bin bash.\nYou don't have to worry too much about this.\nBut all you're doing is replacing this\nwith whatever your container ideas\nwhich will be different\non your virtual private server.\nFrom here we go ahead and press enter,\nand you're going to notice that it will change\nso that we're inside of.\nNow the container ID.\nSo let me clear just so we have more space in zoom.\nSo now we're inside of the Docker container.\nAll we need to do now is we just need to type clod.\nAnd then if we type Claude\nit will prompt us to log in.\nIf you want to log in with Claude.\nWhereas if you want to log in with Codex,\nI believe it's Codex\ndash log in or something like that.\nWe can try it. Codex dash, dash log in. Let's see.\nNot a valid value.\nLet's try dash log in. Okay.\nLet's just try Codex log in. Sorry.\nIt's codex log in, as you can see.\nAnd you'll notice that it gives us a link.\nBut it also says if we're on a remote\nor a headless machine, then we can use codex\nlog in, dash dash device, dash off.\nSo that's actually what we'd rather use.\nSo to get out of this\nwe're going to hit Ctrl C on our keyboard.\nThat's how you escape like an interactive window\nwhen you're inside of this SSH session.\nCtrl C same on Mac and Windows.\nAnd we can then paste this new command\nthat it told us Codex log in dash\ndash device device dash officer and then hit enter.\nNow when we do that it's going to say\nopen this link, sign in and then paste this code.\nAnd then it will automatically authenticate\nour device.\nAnd we're good to go.\nNow I don't want to do that.\nI'd rather use Claude code.\nSo for me\nwhat I'm going to do is just type Claude.And\nwhen I type Claude it's going to prompt me to log in.\nSo I'm going to go yes, dark mode is fine.\nAnd I want to use Claude Account with a subscription.\nSo I'm going to press enter again\ncheck the terms of service.\nI'm not exactly sure if they're allowing this or not,\nso it's up to you if you want to risk doing this.\nI'm just quickly doing it\nfor the demo that I will disable it.\nSo I'm going to copy this URL that it gave me here\nokay.\nAnd I'm going to paste it right here in my browser.\nNow once I paste it in my browser\nit is going to ask me to sign in.\nSo I'm going to authorize this\nbecause I'm already signed into my Claude account.\nAnd it's going to tell me to copy\nthis code back into my window.\nSo I'm going to copy the code that it gave me.\nGo ahead and hit enter.\nAnd then we should be good to go.\nAnd you can see that I've logged in successfully.\nAnd I have the Claude Max account,\nwhich gives me a ton of usage to be able to run this.\nNow again, you can use an API key.\nYou can use a subscription.\nIf you have a cheaper subscription,\nit's probably gonna run out quickly.\nAnd you're going to want a different API\nkey up to you.\nDo your research,\nbut you have to set it up with at least one of them,\nwhich is why I'm going through this process.\nOkay, so now that we have this, it says,\ndo you want to access this workspace?\nI'm going to say, yes, I trust this folder.\nAnd then it's opened up. Claude.\nNow I don't actually care about using Claude here.\nSo I'm just going to type slash exit.\nAnd now I'm out of here and I don't need\nto do anything on the virtual private server.\nOkay, so now that we have authenticated\nand we've at least added\neither one API key or one subscription or one,\nyou know, connector, right?\nLike Codex Claude, Gemini, open Claude,\nwhatever it is you want to use.\nNow we can continue. So now we're on the prompt.\nWe're creating the company now to make the company\ngive it a name.\nI'll just go with tech with Tim\nInk for right now.\nGiven a mission and a goal now, it's super important\nthat you're very specific\nwith this mission and goal.\nBut don't make it super long.\nYou don't need like a 10,000 word prompt.\nWhat you want to tell this company is like overall,\nwhat is it doing?\nLike the mission statement of,\nyou know, these top companies in the world.\nLike that's the kind of thing\nthat you're giving with a little bit more specifics.\nSo in my case, I'm just going to say something like,\nI would like you to optimize\nmy content pipeline\nand help me grow my YouTube channel.\nAnd even this goal is probably not specific enough.\nSo I would maybe reset and say, I want you to help me\ngrow my YouTube channel tech with Tim by automating\nprocesses, reviewing content ideas, creating scripts\nand, you know, reviewing previous performance.\nOkay. So that's like specific enough.\nWe can add a little bit more detail if we want.\nBut generally to start,\nyou don't need anything super large.\nNow from here we can connect Claude code or Codex,\nwhich are the two recommended\njust to start for our first agent.\nFor other agents,\nwe can connect them to other providers if we want.\nAbove,\nyou also can just use the same for all of them.\nNow what we'll do is\nwe can also look at more adapter types.\nSo you'll notice that we can connect to a cursor.\nFor example we can connect to pi to open code, to\nGemini to the process or to the open Claude gateway.\nBut if we do open Claude Gateway\nthen it's a little bit more complex in the config,\nwhich I'm not going to show you here.\nSo for now we're just going to go with Claude code.\nWe're going to leave our first agent as CEO.\nAnd for the working directory path,\nwhat you're going to want to do here\nis create a new folder somewhere on your machine.\nI'll show you how to do it\non the virtual private server.\nAnd then just put the path to that folder\nso we can work directly inside of there.\nSo if you're working on Mac, for example,\nmaybe you make a new folder on your desktop\nand then what you're going to have to do\nis give the path to that folder.\nNow if you go choose, it will show you\nhow to find the path on Mac, windows, or Linux.\nSo just follow the prompt there to find what it is.\nBut if we're running on a virtual private server\nso I'm inside of mine right now,\nwhat we can do is type LS.\nNotice\nthat we're inside of this kind of home directory\nwhere it has instances, instances,\nthen contains paperclip.\nSo what I can do is I can just make a new directory\nand I can just say paperclip\nI don't know, work or something.\nAll right.\nAnd then if I type ls you can see that\nwe now have this new folder paperclip work.\nI can CD into this paperclip work folder\nwhich stands for change directory.\nAnd by the way mkdir stands for make directory.\nSo I just made a new directory in my user's home\nfolder.\nYou don't have to worry about what that means\nif you don't know,\nbut you just type this mkdir paperclip work.\nLS will list the directory\nso you can see we have paperclip work.\nAnd then what\nI've done is I've changed into that directory.\nAnd now I type D which stands for Print\nworking directory.\nAnd it will give me the full path to where this is.\nSo I can just copy that\nand I can paste it right here.\nAnd I'm going a little bit fast\njust because there's like a lot of set up.\nI don't want to bore you guys too much,\nbut anyways, just make a new directory\nif you're on Mac or Windows a little bit easier.\nIf you're on VFS,\nyou're just going to run this command mkdir\nand then literally you can just use this exact path\nif you're working with hostinger\nand you'll be good to go.\nSo then we're going to go ahead and press on test\nnow for the adapter.\nIt's about to press next.\nBut we're going to do tests now.\nAnd when we test it\nit should give us a green highlight saying\nhey, all of this is working\nand we're connected to Claude code.\nIf we tried with codecs, for example. Right.\nAnd then I try to press the test,\nit should fail us because we don't have any,\nconnectors set up for open AI.\nAnd let's see what it gives us here.\nAnd it says, hey, the open API key is not set, codec\nkey is installed,\nbut the authentication is not ready, etc., etc..\nOkay.\nSo we need to make sure that we have it connected\nlike we do already with Claude code.\nAnd if you're running on your own computer,\nyou're going to need to install one of these 2\nor 1 of these other CLI commands\nso that this connector passes before you go for it.\nOkay. We're now going to press next okay.\nSo we now move to the first task.\nNow for the first task\nyou don't need to change anything here.\nYou can just go ahead and press on next.\nWhat this is going to do\nis tell the CEO to hire its first employee\nand then just continue from there effectively.\nSo don't worry too much about the details.\nJust press next and then you can see it just asks\nif we want to launch.\nSo I'm going to go create an open issue.\nAnd what you're going to see now is what happens\nis we launch a new paperclip company.\nYou can see the company is on the left\nhand side here with the name.\nAnd we now have our agents, our projects,\nour work, our issues.\nAnd this is where it starts to become fun.\nAnd I can start showing you a lot of the setup and\nhow to get some actual useful results out of this.\nOkay, so first thing you need to understand\nabout paperclip once it's set up\nis that this isn't a chat bot,\nso you're not meant to come in here\nand ask it a question and work with it\nlike over a long period of time.\nWhat you're meant to do is create issues.\nThose issues will automatically get picked up\nby your agent, in this case the CEO.\nAnd then delegated to the various\ndifferent employees.\nSo it's really a different paradigm\nof working with AI agents.\nAnd don't expect to sit in here\nand do like 3 or 4 hours of work in the company.\nIt's meant so that the company runs itself,\nand you more present a high level objective,\nand the agents determine how they can solve those\nwithout you really needing to be involved.\nSo you shouldn't be sitting there saying, open a PR,\ndo this, do that, run this, install this command.\nYou should say more.\nHey, review\nall of my past YouTube data and generate a report,\nand then it will just figure out how to do that.\nYou don't need to walk it through,\nyou know, every single process.\nAnd for me it's been a little bit weird,\nkind of, you know, getting used to this new way\nof working with the agent.\nSo that's what I really want to focus on in this\nlater part of the video might be a little bit long,\nbut I want to give you some useful content\nso you can set this up properly.\nOkay.\nSo a lot of things to go through here on the\nUI right.\nEssentially this is a dashboard that shows you\nhow multiple agents are going to be working together.\nNow the first thing to look at is the CEO agent.\nNow the CEO agent is the first one that you create.\nThis is the one that you should really directly\nassign issues to, which we'll look at in a minute.\nAnd sometimes the CEO agent will ask you for approval\nto do certain tasks.\nSo you need to view yourself\nkind of as like the board of advisors, whereas\nthe CEO is the one who will go in and delegate\nthe different things that need to be done.\nNow the CEO runs something called a heartbeat.\nYou can see the heartbeat\nif you go into the configuration and you scroll down\nand you find this value right here\nthat says run heartbeat.\nNow you want to make sure the heartbeat is on,\nand the interval is the amount of time\nin which it's going to run.\nSo effectively, the way that this works\nis that by default,\nevery hour,\nthe CEO will be triggered to start executing.\nWhen it starts executing, it's going to look at\nany of the tasks that are in the pipeline.\nAnd then start working on those tasks.\nIf you don't have any tasks in the pipeline,\nit may not have anything to do.\nSo just keep that in mind.\nThe CEO runs every single hour\nor whatever time you set it.\nIn my case, I change this to 600 seconds\njust so it runs every ten minutes,\nbecause I didn't want to wait for it to run.\nYou know, every hour.\nBut if you don't want to use a lot of money,\nthen you can change the view.\nThe thing here, right, in terms of how much it runs.\nOkay. So anyways, that's the heartbeat.\nHopefully that makes sense.\nBut that's how this is like autonomously running.\nThe CEO runs a heartbeat.\nNone of the other agents do.\nAnd then when the CEO runs the heartbeat\nit can trigger other agents to run.\nSo now let's go ahead and press on save\nand you'll notice that we also have permissions\nwhich just can create agents.\nNow you can disable this or enable it\nI would recommend that you keep it enabled.\nAnd what will happen right away\nis that the CEO will create a new project\nvery likely, and start spinning up different tasks\nand assigning those to other agents.\nNow you'll notice that if I go here, it's\ngiving me a request here, and it says, hey,\nI want to hire my first agent,\nwhich is the founding engineer.\nNow this is just by default,\nwhat will happen automatically.\nSo just go ahead and approve that request.\nAnd then as soon as you do that,\nI can start assigning tasks to the founding engineer.\nSo now if you go into the org chart,\nyou can see you have your CEO\nand the founding engineer reports to the CEO.\nNow you can have the CEO\njust create as many agents as you want.\nYou also can manually create them\nby just pressing this plus button here.\nRight?\nI want to configure the agent myself,\nbut I recommend that if you want a new worker,\nnew employee, new agent, whatever,\nyou just instruct the CEO to hire them\nand give the CEO the instructions\nthat it can hire them whenever it wants.\nOkay, now you can see\nall this stuff is just happening in the background.\nYou know, as I'm trying to explain,\nit's a little bit overwhelming.\nSo I'm trying to go through it step by step.\nBut effectively the CEO is\nwho you're going to be interfacing with from the CEO.\nIt has a heartbeat.\nThat heartbeat runs at whatever increment you set.\nI just showed you how to change the increment.\nI recommend\nmaking it a little bit lower than one hour.\nAnd what the CEO will do is create\nand pick up any tasks that are in the pipeline,\nand then delegate those to other agents\nor do some of the work itself.\nOkay, you can manually trigger the CEO\nto start working by pressing this run\nheartbeat button at the beginning.\nThat's probably a good idea, just because\nyou're going to be giving a lot of issues and tasks,\nand then you can assign a task to the CEO\nby just directly pressing\nthis assign task button, you can see that\nit will automatically select the CEO.\nYou can select a project\nbecause you can have different projects.\nIf you want to separate\ndifferent things that it's doing.\nAnd then you can give it a description\nas well as a priority.\nIf you want it to do something right away,\nyou just put the critical priority and the next time\nthe heartbeat runs, it will just work right away.\nAnd actually, I believe if you assign the issue to it\nand you create the issue, it should just\nautomatically trigger\nbased on the fact that you assigned something.\nRight now, you can add labels, you can upload\nfiles and data, all of this kind of stuff.\nRight.\nAnd there's other options\nlike you can put start, date and date etc..\nOkay. So that's kind of the CEO.\nAnd if you just want to like chat with the CEO\nyou just come here,\npress assign\ntask, say hey, do this thing and assign it.\nI'm going to show you examples in a minute,\nbut I'm trying to go through\nlike the high level outline here,\nwhich is a little bit confusing.\nOkay. So that's that.\nYou can look at all of the runs\nthat have been completed.\nYou can see all of the stats\nin terms of the input tokens and whatnot.\nIf you want to monitor\nwhat your agents are doing a little bit better.\nAnd then you can also set a budget.\nSo if you are using an API key,\nyou probably want to set a monthly budget here.\nJust to make sure\nthat you're not going to spend too much.\nAnd you can do that in the cost setting right here,\nwhere\nyou can see how much you're spending\nand you can customize all of these different values.\nYou know, as you see. Right.\nSo set budgets, all that kind of stuff.\nBut if you're running on a subscription,\nyou don't really need to do that.\nAll right.\nSo that is that. Now you also have the dashboard.\nIf you go to the dashboard you're going\nbe able to see all of the different agents\nand what they're working on.\nAnd these are effectively different colored\ncode Codex whatever agent you're running sessions.\nSo the founding engineer is running a session\nright now.\nIt's waiting for another session\nto start for another session to start\nthe CEO has its own session running.\nSo it's like allowing you to paralyze and view\nall of these\ndifferent sessions running on your company.\nOkay.\nYou also can see the success rate issues,\nall of this kind of stuff, you know, explore it.\nYou guys will figure it out more.\nBut I'm going to go into a few optimizations.\nAll right.\nNow we also have the inbox.\nThe inbox is stuff that it wants you to be aware of.\nTypically if you create an issue\nthen it will kind of put some things into your inbox\nsaying,\nhey, you know, this issue you created, it's done,\nor it's waiting for your review or,\nyou know, we want to hire a new agent or whatever,\nso you can see anything on red,\nyou can see all and you can see recent.\nI'm just going to mark them all as red right now.\nBut this is not going to be\nall of the status of the company.\nIt's only something\nthe CEO thinks that you're going to need to look at.\nAll right. So let me go to goals now for goals.\nThis is the probably the most important part\nof the company, right?\nWe created a goal by default\nwhen we kind of spun up this agent.\nBut you want to be pretty specific\nwith the goals that you have.\nYou can create various subgoals for a main goal\nif you want to guide the company\na little bit better,\nand then you can just create multiple goals as well.\nNow, I personally have not made that many goals.\nI find just having like 3 or 4 of them is enough,\nand then just being a little bit\nmore specific in terms of exactly what you want it to\ndo, is really helpful.\nSo at the beginning,\nyou know, if I'm talking about my YouTube channel,\nI'm going to have to give this some access.\nRight?\nSo something that I would do\nis add a subgoal and say connect to YouTube\nanalytics.\nRight. And then I would tell it, hey, here's\nmy credentials.\nHey, here's all my details.\nHere's how you can connect to Google Analytics.\nUse this key and connect. Boom. Right.\nAnd then I would just provide some key.\nAnd then it would automatically go and connect.\nAnd the other company that I showed you,\nI had it connect to my ConvertKit.\nSo I gave it a ConvertKit API\nkey in the issue directly here I said use this API\nkey, connect to my ConvertKit and start integrating\nthat to be, you know, part of our platform.\nI then made a GitHub account for my agent,\nwhich I connected in the, terminal here.\nSo I just like signed in with a new GitHub account.\nAnd then I told it, hey,\nyou are signed in with this GitHub account,\nstart creating new GitHub repos and,\nyou know, push PR to this repository.\nAnd then I gave just this GitHub account\nfor the agent access\nto my main website repository, where it can make PR\nbut it can't commit to the main branch.\nOkay.\nSo things like that, like you want to do your initial\nset up on what you want the company to do.\nOkay.\nSo a few other things that we need to go through.\nSo if we go to issues here, this is where\nyou're going to be working a lot of the time.\nAnd you can switch between this like list view\nand this Kanban view.\nNow if you look at this Kanban view,\nyou can see that we have like a backlog\nto do in progress, in review, blocked, done, etc..\nNow the two categories that you're probably going\nto want to be looking at is the in review\nand the blocked.\nNow sometimes the agent will want to do something,\nbut it's blocked.\nFor example,\nif I ask it to go push something to GitHub,\nbut it's not connected to GitHub or it doesn't have\na GitHub account, it's going to create\na blocked ticket and it's going to show me\nin that ticket why it's blocked.\nAnd tell me that it requires manual intervention.\nSo let me just scroll down here and you can see\nlike when you look in any one of these issues.\nSo I just clicked into one of them right at the end.\nIt gives you kind of like a summary of what was done.\nIf we versus I think I clicked into the wrong one.\nOkay. Let's just go this so sorry.\nLet's click into one of these.\nAnd you can see at the end it tells us, hey done.\nAnd then it tells you like exactly what it did.\nIf this was in a blocked ticket\nthen it would tell you why it was blocked\nand what you need to do to unblock it.\nOkay, now you can change this.\nSo if you do unblock it, you can change it to go back\nto to do back to backlog.\nYou can update any of this stuff, right.\nAdding new labels, adding new assigns.\nAnd you can really just assign the issues\nto who you want.\nNow in review typically means\nthat it's waiting for you\nto approve something before you move forward.\nSo if there's something that's in review.\nSo for example, let's drag\nthis one over to in review.\nRight. You'll see that it's not doing it right now.\nBut this should change to purple.\nAnd when it changes to purple you able to go down\nhere and see okay this is why it's in review.\nThis is what it needs for me.\nAnd you can actually just directly leave\na comment here so you can say something like approved\nand then comment directly on the task. Right.\nAnd then when you do that the next heartbeat,\nit should see this comment\nand it should be able to move it forward\nbecause it's approved.\nYou also again can manually do it.\nAnd you can put it\nas for example, back to in progress.\nAnd then it will pick it up and start\nworking on it again okay.\nSo hopefully you get the idea.\nBut that's kind of how you work with the issues.\nNow again, when do you want it to do something.\nYou don't have any chat interface.\nYou just have to make an issue.\nSo you have to say, go do this thing.\nI want you to set this up.\nI want this to happen and you just make these issues.\nAnd if you want them to happen immediately,\nyou make them as critical.\nAnd if you want it to trigger right away,\nyou trigger it to someone.\nSo you trigger it or assign it to the CEO,\nyou assign it to the founding engineer,\nand as soon as you do that, the agent will\nautomatically pick it up and just run it right away.\nIf you don't assign it to someone\nand you leave no assignee,\nthen it will go into the task list\nand the next heartbeat.\nThe CEO will pick it up and assign it\nas it needs to be.\nSo let's just do a few more examples.\nSo you kind of get the flow of how to work with this.\nAnd then we'll move on to some optimizations.\nSo for example, if I want to hire new employees\nI'm going to make a new task.\nAnd I would say hire\nI don't know, seven new employees.\nOkay.\nAnd then I'm just going to say, I would like you\nto hire seven new employees to start delegating\nvarious tasks related to optimization\nviews, thumbnails, video generation topics,\nresearch, competitor analysis, sponsorships, etc..\nAgain, I'm being a little bit random\nbecause it's just for the video.\nI want to put the priority as critical,\nand I'm going to assign it to the CEO.\nOkay.\nWe'll put it here in this project\nand create the new issue.\nAnd when we create it, it goes to to do.\nAnd you'll notice if we go to the CEO,\nwe'll see that it's just picked it up right away.\nWe can see what the task is okay.\nAnd you can see it says hire seven new employees.\nWe can look at it and we can start\nwatching in live time what it's doing.\nOkay. We can open up the run.\nWe can leave a comment. Right.\nWe can kind of direct you.\nWe can attach files,\nwe can see the cost, but it's going down.\nIt's doing this.\nNow if we go to the inbox, if we just wait a second,\nit should start popping up requests\nfor hiring new agents.\nAnd we'll start to see this org chart\nkind of get built out with new employees.\nSo I would recommend when you start working with this\nset up the structure,\nhave the CEO kind of hire who it needs to hire,\nget the agents in place,\nand then start adding a few different issues\nto kind of direct them on overall what it needs\nto work on, and then you'll have a better\nkind of foundation for it to do something effective.\nAgain.\nFor me, I was connecting GitHub, I was connecting\nConvertKit, I was connecting a few API's.\nThere's some stuff that you do manually need to do.\nLike they can't do everything on their own,\nbut once they have the tools and the connections\nand they kind of know\nand are aligned on what they're supposed to do,\nI saw that I got pretty good results,\nespecially from my other company here, right\nwhere it's been running now for like two days,\nand it actually has produced like meaningful outputs.\nSo if we go back to the, inbox here.\nOkay, we're still waiting for the hires.\nOnce it pops.\nWas up\nthere, I'll be right back and we'll kind of continue.\nOkay.\nSo it started submitting the requests.\nWe can see\nthe agents are kind of popping up on the side here.\nAnd notice it's asking the higher agent,\nhigher agent, higher agent.\nSo I'm just going to approve all of them\none by one here okay.\nAnd we'll get all these new agents\nset up in our org chart.\nAnd we can kind of see them,\nyou know, starting to run okay.\nNow I just want to say I was going through this\nI know it's a little bit all over the place.\nYou know I'm new to using this as well.\nIt's like pretty much brand new tech.\nYou really do need to play around with it\nto get some good results.\nLike it takes a bit of time up front,\nbut once you set it up,\nit's pretty crazy that it just runs on its own\nand you just kind of kind of check check in on it\nand see all of the work being completed.\nSo you can see we've got sponsorship\nmanager, thumbnail designer.\nIf we go to the org\nit's probably easier to see it. Right.\nAnd we can kind of\nlook through and see everything that we've set up.\nAnd really there's no limit to the number of agents\nor the depth of the hierarchy that we can create.\nBut to be honest, do you need 500 agents?\nProbably not. Right?\nLike having seven, eight, nine, ten,\nmaybe 15 or so is going to be really more\nthan enough to get what you need done.\nAnd it's going to be hard to monitor\neven this number that I have right here.\nWhere you really get good is when you start creating\nkind of like feedback loops with the agents.\nSo you have like sub teams\nand then those teams have like reviewers.\nSo one thing I found that worked\nwell is having like a newsletter researcher,\na newsletter drafter, a newsletter editor\nand then a newsletter reviewer.\nSo it's kind of this cycle\nwhere they cycle the information to each other.\nAnd then that way these agents\nare checking their own work, and then finally it gets\nsubmitted to the CEO and then me for final review\nbefore we actually go live or something.\nOkay, so we've created these agent teams.\nThese are super cool.\nHowever, we really need to start fine\ntuning what these agents actually do.\nAnd the way that we can do that is by creating this\nprompt template and by specifying some capabilities.\nSo for the CEO specifically,\nthis is probably the most important, kind of prompt\nthat you're going to write where you need\nto instruct it, what it actually has to do.\nSo it knows on every heartbeat,\nevery time it spins up. Okay.\nThis is my process. I need to follow this.\nIf you don't do that, what I found\nis that a lot of times the CEO just says, hey,\nI don't have anything to do.\nAnd it's not proactive, it's more reactive.\nSo I'm going to show you the prompts that I have\nfor my current organization.\nOf course, you're going to have to adjust them.\nAnd what I did is I actually had Claude\njust write the prompts for me based on what I wanted.\nSo I went to Claude. I said, hey,\nthis is what I'm looking for.\nYou know, write me a prompt.\nAnd then I was able to generate that.\nSo if we go to my other company and I go to my CEO,\nI'll show you in configuration\nthat this is the long system prompt that I gave it.\nSo effectively what I said is, hey,\nI want you to check the status of every employee.\nI want you to assign work to every idle agent.\nSuper important.\nSo it's actually proactive and it's doing more stuff.\nI want you to review all the completed work, draft\nand queue the emails\nand then review metrics and adjust the strategy\nbecause that's what the CEO should be doing.\nAt least in this organization.\nAnd then I gave it a really strict protocol on what\nI want it to do.\nYou want every heartbeat.\nPhase one stats check.\nPhase two review the completed work.\nPhase three\nright when you give these agents repeatable\nworkflows, this is almost like a skill file, right?\nIt starts to become a lot more useful. Okay.\nAnd then I didn't do anything\nwith the capabilities here or the permissions,\nbut you can mess with those if you want.\nAnd then for all of these other agents\nI gave them same thing, just a slimmer\nsystem prompt\nso they understand what they need to do.\nOkay, so you're a researcher agent.\nThis is what you do right? This is the output format.\nThese are the rules that you have\nfor the founding agent.\nSame thing right okay.\nFounding engineers are a git workflow.\nYou need to commit stuff to get. Don't push to main.\nMake sure you make a pull request newsletter\nreviewer.\nRight. Same thing. And you go through\nand you can see.\nAnd if we go back to my dashboard or where is it,\nlet's go to issues\nand we go to the Kanban, you can see that\nthere's a bunch of stuff that they've now completed.\nWe had some canceled tasks previously.\nWe had some blocked ones I wanted unblocked them.\nWe have some stuff that's in progress\nand it's running pretty effectively.\nSo that's kind of how you set up the company\nand how you get the better results.\nFocus on these prompts,\ngive them a process and specifically for the CEO,\nyou want to instructed to be proactive and assign\nnew stuff.\nOkay.\nNow of course\nthere's a lot more that I can go through here right?\nEffectively this is new tech.\nI'm not an expert in it,\nbut I did figure out enough\nto get it working to a point where I'm pretty excited\nby what it's going to be capable of doing.\nYou can make new companies as well.\nBy the way,\nit's worth noting on the left hand side here,\nit has have as many of them as you want running.\nWhat you should under understand,\nsorry, is that this is effectively\nan orchestration platform around the existing LLMs.\nYou bring in whatever agent you want, and this is\njust allowing them to communicate more effectively.\nAnd it's setting up this team like structure,\nwhich is effectively just different Claude code\nor codex sessions that stay persistent\nand can kind of go and message each other and,\nyou know, work on this dashboard\nwhere they send effectively\nAPI requests to create different issues and tasks.\nYou could remake this thing yourself in a few days\nif you wanted to custom to your own situation,\nbut obviously it's pretty well set up.\nAnd I mean, it works pretty effectively\nfor managing a lot of agents.\nAt the same time,\nthis is going to be overkill for simple tasks\nif you just want, you know, a simple website spun up,\ngo and ask Claude Code to do it.\nBut if you want to actually run something\nautonomously in the background and mess around\nwith these four AI organizations,\nI think this is a really cool first step.\nAnd I'm excited to see how the technology evolves\nbecause it's definitely not 100% there yet,\nbut very soon I think this will actually be\nsomething viable, and we may legitimately\nhave AI company teams that go out and do things\ncompletely on their own without need\nfor your intense oversight,\nbecause right now I am finding myself going in here\nmaking a lot of issues,\ngiving some individual instructions,\nwhich I would obviously like to get away from.\nSo anyways, guys, let me know what you think of this\nin the comments down below,\nand I look forward to seeing you in another video.",
  "transcript_chars": 47944,
  "ingested_at": "2026-05-21T19:01:11.059600+00:00",
  "source": "retry-no-transcript",
  "yt_meta": {
    "view_count": 102703,
    "like_count": 2315,
    "channel_id": "UC4JX40jDee_tINbkjycV4Sg"
  }
}