Category: Sharbel YouTube

Running a single AI agent is great, but running an entire team of them is a completely different level. So, in this video, I’m going to show you how to run a company of agents. We’ll create a team together, give each agent a role, put a manager in charge, and let them run real work on their own. And the best part is I found a platform where we can do this with absolutely no code required. Let me show you how it works. Let’s start with why the one agent thing breaks. I mean, you’ve probably felt it if you’ve ever tried giving one agent hundreds of different jobs. You ask it to do research, writing, and design, and what ends up coming back is, let’s just say, mid. It’s the same reason you wouldn’t hire one person to be your researcher, your writer, and your designer. They’d be stretched too thin to be good at any of it. But, honestly, that’s the small wall. And there are different ways to solve for this, which I cover in some of my other videos. However, there is a bigger wall, and nobody really talks about it. Here’s the thing, I don’t just play with agents. I actually run a business with a team of nine people, and every AI agent harness I’ve used, whether it’s Hermes or Openclaw, they’re great single-player agents. I mean, they are built and designed to be used for one person. That’s not a bug, it’s a feature of those tools. So, the second you try to run that like an actual team, it starts falling apart. Your people can’t get in. They can’t see what the agent did. They can’t jump into the same thread. They can’t work alongside it. And most companies are not single-player. I mean, we’ve tried adding a Hermes agent to a Slack team channel, but permissions on there are a nightmare. And that was a big wall for me. I have this powerful agent and a team of nine people, but no way to put them all together in the same room. So, let’s fix it the way a real company would. And that’s exactly how I found HyperAgent. It’s the first setup I’ve used that’s actually built for teams instead of one person. So, my whole agency can work inside of it together. And luckily, this video is sponsored by HyperAgent. HyperAgent is a place to hire and run a whole team of AI agents in the cloud with no code required. Nothing to install. You just sign in, make an account, and you’re in. Oh, and you can now sign in to HyperAgent with the ChatGPT plan you already pay for. So, you don’t have to pay for any other AI subscription. And if you don’t have a ChatGPT plan, you can use my link below to get $100 in free credits to use on any AI model that you want. Okay, let’s hire our first agent. Building an agent in here is basically like writing a job posting. To hire our first agent, we’re going to click over here, new agent, and the first step is identity. We have to describe what the agent’s job is. And I’m going to start with our manager agent. He’s going to be in charge of managing and orchestrating all the other agents. So, let’s write, you are Matt, chief of staff, you run my company’s AI agent team, and you report only to me. Let me actually add, your job is to take what I ask for, break it into the right pieces, and hand each piece to the specialist who owns it. You don’t do the work yourself, you delegate it, you follow up on it, and you bring me back one clear answer. Always keep it short and concise. I hate long paragraphs of fluff. That’s it. From there, we’ll click on draft identity, and it will create its own system prompt. We, there you go. It just did. It gave itself a system prompt, a description, and we’re going to give it a name. I’m going to go for Max. And then, the next step is picking out its model. Since I connected my GPT subscription, I’ll go with my OpenAI, uh, not GPT-6 Astra. We don’t need something that strong. I’ll go with the latest. So, bigger models make better calls and cost more. So, match the model to the role. You can even set a budget cap for each and every one of them by toggling this on and then putting what your budget limit is. Since I’m on the subscription, I’m going to keep it as is. And then, when we’re setting up the other agents, you can point the grunt work at a cheaper model. Then, we go to tools and integrations. These are essentially what these you want to hand each agent. For example, our content agent that we’ll create later needs to have Instagram access. So, we’ll connect that. But, I don’t think our chief of staff necessarily needs that, so I’ll just go to the next one here, which is invocations. Here, you can choose where other places you want to be able to reach your agent. If you want to talk to it via Telegram or for it to talk to you via Slack or email, you can turn that on and set that up right here. I don’t need that necessarily. I want to be able to talk to it here, so I’ll just keep it as is. Oh, and another thing you can set here before I forget is who can also reach your agent. I which I’ve set this to anyone since we’re going to include my team when I set this up as well. Then, finally, the last step is knowledge. If you’d like to customize this, you can, but I’ll keep this as it is by default and click on create agent. And there you go. We can also give it a an image, which I would like to do that. I think Max is a wizard. I’d like a 16-bit avatar of him. And it’s generating it. And there we go. I love it. I’m going to choose to keep it, save it, and from there, the next step is to create a new thread with our agent and make sure that plan mode is on. It is on by default, but just make sure. That way, it’ll show you its steps before it touches anything for the first turn, for the first prompt we’re going to give it. And with that, let’s hire the rest of its team members. Now, we bring on the rest of the company. There are two ways we can do this, either by going through what we just did for every different AI team member, or by asking our chief of staff agent to do it for us. And that’s exactly what we’re going to do here. So, I’m going to send it. Before you can run anything, you need a team, so let’s hire them. Create these four agents. Number one, a prospector, who finds and vets new leads every day and hands the qualified ones to the setter. He needs web search and browser. Then, second, we need a setter, who lives in my inbox, replies to incoming leads, and drafts outreach in my voice. I already have a template to reply. We can set for it. Never send anything without my approval. Needs email access. Another tool we need to make sure this specific agent has. Then, is a client success manager, who keeps an eye on our active clients and their feedback and flags anyone who’s unhappy or gone quiet. He needs access to where our client conversations live. I already have a bot set up for that, or rather, a Telegram registry set up for that, so we need to link it to this. And then, a content creator, who tracks my Instagram performance and brings me content ideas and full drafts to film. Needs access to my Instagram, its analytics, and to VidIQ. The purpose of you and your team is to help me manage and scale my business. This is the first prompt I’m going with. I’m going to hit send. And just like that, it’ll open its first thread in plan mode for that first prompt only, where it’s going to start ideating and thinking and going over the first prompt that I sent it, what I like to call the startup prompt. And from there, we’re going to see what it’s going to come back with. So, here, it already asked the first question, which inbox should the setter work from? I use Gmail, so Gmail. Where do your active client conversations and feedback primarily live? Multiple places. Let me actually name them. Telegram plus Slack. Telegram plus Slack. I already have an API that I can give you later on to access this data. And what should I use to configure the setter’s voice and the content creator’s research workflow? I’m going to send that. Those are the three questions it asked me with their three replies. Sent me the team structure is clear. I’m turning it into an implementation plan that pre-preserves approval gates and accounts for the integration that still needs to be connected. So, starting to draft the implementation plan, and it asked me one more question, “How would you like me to proceed with this four-agent team plan?” Let’s read it. Yep, it seems to have understood well what I want, so I’m going to approve its plan and send that, saving all the things that I’ve given it to memory, and it’s going to continue to do that every time it learns something new. It’s very good, and we’re able to see what it’s saving. There we go. These are the four agents. All I need to do, I can look into each one if I want to. See prospector, there it is. Integrations, even set its schedule for it. I save all of those agents. I’m going to tell it that I saved. Here is the next step, which is connecting each agent to whatever tools they need. Here, really, it’s very customizable depending on what your own specific agents that you’re creating need. What you can do, and what we can see here, is all of them appear on the left bar. So, what I’m going to do is I’m going to click on each one and make sure that they have their capability set. For example, the content creator needs access to Instagram and to VidIQ, so I’m going to go ahead and connect everything that I need to connect. And there we go. Just a few minutes later, everything is set up. I’ve even given each of them a profile picture and all the capabilities. Like, the setter has access to Gmail, the content creator has access to Instagram. And by the way, there’s a whole ton of integrations. I doubt that you won’t be able to find whatever integration is that you want. So, now we’ve got a company. Let’s put them in the same room and start working. It’s called rooms, and you can access all of them over here. A room is basically the office. It’s like a Slack channel where you can add different agents and different team members to different groups. It’s essentially one shared workspace where your whole company works together. And three things happen here. Your specialist agents, in our case the prospector, the setter, the client success manager, and the content creator all run in parallel instead of one at a time. And then your actual human teammates can jump into that same thread and talk to them. Let’s call this one general chat and create the room and most importantly add our members like setter, the content creator, the client success manager, the prospector, and most importantly our operator Max. And I’m going to send hi Max, I’ve just set up this room. This is where we’ll all be working together. And Max answered hi Charbel, got it. I’ll use this room to coordinate with you and the team, keep updates, and pull in the right specialist when needed. Max, what else do you need so that we go live? I’m ready. And here you can see them start to show up. I sent the message to Max, but because the setter is involved, I’ll need the setter to draft an outbound template. That’s literally what I said. The setter is even typing. Like imagine this, the prospector drops a batch of leads into the room, the setter picks them up and drafts the replies, and then it tags me, the human, with something like hey, want me to send these? And nothing goes out the door until a person says so. Or depending on your permissions, I can set it to run automatically, but for the first drafts, I recommend that you keep human approval. And that right here is a team stand up running on its own. So, let’s give them a real job and watch what happens. This is amazing because I asked Max, our operator, to go ahead and start the workflow, and it came up with a plan, and then it started delegating. Here you can see there 12 replies. I sent one message, I received 12 replies, and that’s not them necessarily talking to me, that’s them communicating with one another. And we can see that. The setter sent that it found 15 inbound leads along with an email draft that it gave me, which I do want to change, but it’s good because this is where we get to set it up, and an outbound email template, which I absolutely love. This one I’m not going to change. I’m actually going to tell it. And it also sent us the client success manager report as well, which I’m going to blur for obvious reasons, sensitive client information. But I’ve told it that’s great. Let’s set it up so that the client success manager sends us a ping around anything that’s urgent every single day. Look at this. It’s amazing. Max is tagging the client success manager. Hey, don’t send anything. The client success manager, yeah, of course, everything is read-only. No client messages are being sent. Look at that. Even the content creator came in the loop and sent us his report essentially. And look at that. The prospector came back with eight founders found. And I’m going to ask Max every time the prospector finds a lead, they should have an email. If they don’t have an email, then we can’t send anything. And if they have an email, then that to the setter, which will send our outbound template. So, to recap, by just telling them to go, after we’ve outlined everything that we’ve outlined, the prospector found and vetted new leads and handed them to the setter who drafted replies. Then the content creator came in, read my Instagram numbers, and pulled together today’s content ideas with drafts ready in my voice. Look at this. Before you ask AI for 30 content ideas, open your last three sales calls along with the full script, what visuals should be on there, and a CTA and captions as well. We have one, two, three short-form video ideas. And we have our client success manager that scanned clients for us and flagged one that’s gone quiet. Then Max took all of this and packaged it for me very tightly and nicely so that I don’t need to go through everything I just went through. Tagged me only when it needed something from me for me to approve. Absolutely love this. What’s very impressive is I didn’t tell it how to find outbound leads. But looking into its data, it’s literally pulling in different sources for every single lead. No two sources are the same. And it’s going to find their LinkedIn. From their LinkedIn, it’s finding their emails, and it’s drafting the emails with the information it finds. Do you start to see the vision of how a company humans and AI starts to work and click together? And you think about this because the prospector goes and hunts for new leads, outbound leads. And he sends that to the setter, who then approaches those leads. Just like this. I click approve, and there we go. We’ve just sent our first outbound lead that we didn’t even have 5 minutes ago. That lets us set calls and close them. And then our content creator goes out and comes up with content ideas for us for Instagram. Which by the way, if you’re not following my Instagram, you absolutely should. I just started posting content on there, and I absolutely love it. There’s a ton of great free value on there. It’s Charbel XYZ. But imagine that, the content creator drafts you ideas for your Instagram. You post that content, and you get inbound leads messaging you on your email. The setter replies to them, also helps you close the deal, and then the client success manager helps you keep track of all your clients, making sure that all of them are happy and satisfied. That’s the entire vision of how HyperAgent can help inside of that organization. There we go. It just sent the first email. This is magical. All right, now that you’ve seen how this all works, you might be thinking, “I already run an agent. I’ve got Hermes or Open Claw or any other AI agent harness already. How is HyperAgent any different? Or how does it compare?” So, let me break it down. It’s not about who can run more agents because most of them can run a few. The real difference is who they’re built for. Almost every one of these is single player, which is great if you work alone. But I don’t. I’ve got a team of nine, and HyperAgent is the only one I found that’s actually multi-player, where my whole team works in the same space, in the same room, right alongside all of the agents. That’s the gap nobody else is filling. Hermes and Open Claw I genuinely love. I’m a big Hermes agent super fan myself. If you’ve noticed or taken a look at any of my channel, those tools are open-source. They run on your own machine. You own them completely. But, they’re built around a single always-on agent doing your jobs one after another. Grok Bot is multi-agent, starts at $100 a month and it’s hard to get a team inside. OpenAI’s dots is the same idea. It lives inside ChatGPT, but it’s also gated to only pro users, so also $100 a month. Meta’s Muse is a free single assistant for the everyday stuff, booking, shopping, email. HyperAgent is playing a different game. It’s the one built from the ground up for a team. Many specialists, uh manager over them, running in parallel in a shared room with you inside of it. It’s not a better single agent. That’s an org for your team and agents to work inside of it. So, here’s where I land. If you’re solo, a single agent takes you surprisingly far. So, start there. But, the moment you’ve got a team and clients and real work flowing, one agent in your own private window stops being enough. You need your whole company in the same room, your people and your agents together. That’s what I was missing for months, and it’s what made me add this tool into my toolbox. This was never about running one more agent. It’s about running your business like a company that keeps moving without you. From doing everything yourself to running the whole thing. So, tell me in the comments, what’s the first use case you’d use HyperAgent for? I want to see what everyone builds. And if you want to build your own team, HyperAgent is giving you $100 in free bonus credits when you sign up through the first link in the description. So, go hire your first agent and build it alongside me. If you enjoyed this video, then leave a like. Or, if you’re new to my channel, then subscribe because I’ve got a ton more content like this coming your way. Oh, and would you look at that? The algorithm gods really seem to think that you’re going to enjoy this next video as well. So, click it and I’ll see you there.

This Mac Mini runs all of my AI agents. It sits right here on my desk and it’s on all the time. I paid $700 for it last year. But the Mac Mini is only the body. Most of the AI models that do the actual thinking, I rent every single month from Anthropic and Open AAI. And those plans go from $20 a month to $200. OpenAI even has a new $500 subscription. So, I wanted to know, can a Mac Mini replace a $200 AI subscription? So, I actually tested it. And in this video, you’ll see what a Mac Mini can actually run, where it performs better than any subscription, at what price point it starts to make sense, and where it falls apart. Let’s get started. Before we answer the question, can a Mac Mini replace a $200 AI subscription? I think it’s important we understand what is it actually that we’re paying for. An AI agent is actually two parts. There’s the body. That’s the computer it lives on. It’s always on and it runs all our jobs. And then there’s the brain. That’s the AI model it sends everything to. So it can think. And right now my body is this Mac Mini and my brain is clawed and GPT running in someone else’s data center and I pay for every bit of thinking it does. And here’s something nobody mentions. When you pay $200 a month, you’re not paying for one thing. You’re paying for three different things. Number one is the model itself, the actual brain. On the cloth side, that’s Opus and Fable. on OpenAI. That’s GPT Astra for example. Number two, you’re paying for somebody else’s computer to do the reading because every time your agent asks the model anything, a whole data center has to read and process everything first. And number three is the allowance. In other words, how much you’re allowed to use before it cuts you off. But here’s the twist. A Mac Mini can technically do all three of those things. It can be the model. It can do the reading itself. And there’s no cap. So on paper, this little box replaces the whole subscription. However, can that one small box do a job that normally requires a data center full of GPUs? And to answer that question, we need to look at the most important spec on any machine. And it’s not the chip, it’s memory. To run a AI model on your computer, the whole model has to fit in your memory. In other words, your RAM. If it doesn’t fit, it doesn’t run. It’s as simple as that. And on a Mac, that memory is shared. The same 16 gigs runs MacOSS, your browser, your apps, and the AI model. If I open my activity monitor and click on memory, you can see MacOSS and my apps are already almost using 14 gigs of the 16 gigs we have. And I haven’t even loaded the single model yet. On top of that, MacOSS only lets the part of the chip that runs AI use a chunk of that memory. On 16 gigs, that’s somewhere between 10 to 12 gigs for the model. Now a model’s size is measured in parameters which is basically how many connections its brain has. More parameters usually means smarter but it also means much bigger. And those big 27 billion parameter models or you see 27B at the end of their name that you keep seeing in every YouTube video, they’re 16 to 18 gigs. So they don’t even fit. But what if I told you you can predict how fast a model will run before you even download it? Every time the model writes one token, which is about three quarters of a word, the chip has to read the whole model from memory every single time. So the speed is basically how fast your chip reads memory divided by how big the model is. Now, Apple says that my M4 Mac Mini reads memory at 120 GB a second. So, a model that’s around 7 gigs should land somewhere around 15 to 20 tokens a second. But let’s see if that’s actually even true. I already showed you how to install Olama in a previous video. So, if you want to watch it after this video, I’ll link it up here and also in this video’s description. Olama is basically the free app that runs AI models on your own computer. But what I will do is open a terminal page and type in Olama Run Quinn 3.59B verbose. The verbose part just makes it show you the speed at the end of every prompt and I will ask it to summarize the Mac Mini Wikipedia page for me. Then I will actually paste the same exact prompt to Claude as well so we can test them both side by side. Oh, and this reminds me, you can type in Olama PS and if it shows you 100% GPU just like it shows for us, then the model fit. So, we’re good to go. But it if it shows you CPU anywhere, then the model is splitting out of your memory and it’s going to crawl. But that being said, let’s run both of the tests and see how both perform. There you go. Claude has already started answering as soon as we put that prompt in. That wasn’t even 3 seconds of wait time.

while our Mac Mini took 2 minutes and 20 seconds to go through this prompt. And as expected, it went through 13 tokens per second. But that test was only on a small prompt, and my agent doesn’t just chat. And that’s where things start to get interesting. Almost every speed number you see online is measured on a short question like write me a poem. But that’s not how an agent talks to its brain. Every time my agent asks the model anything, it sends a lot more than just my message. It sends it its instructions, its memory, the list of every tool it can use and the whole conversation so far. And before the model can write a single word back, it has to read all of that first. Some people call that prefill, which is just a fancy word for the model reading everything before it answers. Now, fair warning before you try this yourself. On a Mac with less than 24 gigs of memory, Olama only gives the model about 4,000 tokens of memory by default. And that’s very tiny. An agent needs at least 64,000. So, what you do is you go into O Lama settings and drag that context bar. Otherwise, your agent forgets what it’s doing halfway through the job. So, to test this next question out, let’s give it a much longer prompt about the size my agent actually sends and start the timer. There you go. Claude has already answered. And as you can see, this prompt took a whole lot longer to process. And that right here is the trade with local models. The model itself is free to run, but you pay in waiting time instead. And you know what they say, time is money. So now the real question is, is the wait even a problem if the job runs while you’re not watching? So we know it runs and we know the agent has to wait longer on the big jobs. But is that weight actually a deal breakaker? Here’s the thing. For a job that runs on a schedule while I’m asleep and away from my desk, the weight doesn’t cost me anything. A morning brief can take 10 minutes instead of two. I’m not sitting there watching it and waiting for it. So, the weight isn’t really where it falls apart. The real limit is quality. And honestly, so far, we’ve only been scratching the surface because we’ve only looked at the small models that fit inside my little box. So, let’s zoom all the way out and look at every model side by side. This is the artificial analysis chart. It takes basically every AI model out there and scores how smart it is in one single number. And right at the top of this chart, you’ve got the best paid models, the clawed and open AI models scoring around 58. Now, here’s the part that surprised me because the best open models, and by open I mean the ones that are completely free to download, aren’t actually that far behind. They’re sitting at around 44 to 46. So, the free brains really are closing in on the paid ones. But, can you actually run one? Well, the best open model on that entire chart is over a trillion parameters big. Remember, a parameter is just one connection in the model’s brain. And a trillion of them is massive. And remember the rule from earlier, the whole thing has to fit in your memory. Even after you crush that model down to a fraction of its size, it still needs around 236 gigs of memory just to load. The biggest Mac Mini that money can buy is 64 gigs. And that costs about $2,700 or $3,500 if you’re in Europe like me. and we’d still need something nearly four times that size just to load it. So to run the good open brain at home, a Mac Mini isn’t even an option. You’d need a Mac Studio with $256 gigs of memory. That is about $9,500. So let’s actually do the math on that. $9,500 against a $200 a month subscription. That machine takes about four years just to pay for itself. And even after those four years, it’s slower than the subscription. And because you had to squeeze the model down to make it fit, it isn’t even as smart as that chart says it is. So chasing a frontier brain at home cost you 10,000 grand in 4 years, and you still lose on both speed and smarts. Which brings us back to our actual Mac Mini and the small model that does fit on it. Because odds are when you run that little model locally, you’re not replacing $200 of clawed opus. Nor should you be expecting to do that. You’re replacing what that exact same model costs to just rent online. And that’s almost nothing. About 10 cents for a million tokens to be exact. So even if your agent runs all day, you’re looking at a few dollars a month, not $200. That small brain lives way further down the same chart. So on the hard stuff, it’s going to lose, but on a clean, simple job, it’s absolutely fine. So the money was never the reason to do this. The real reason is something a subscription can’t give you at any price. See, cheaper isn’t even the real story here because there are three things this little box does that no subscription can at any price. The first one is privacy. When the brain runs on this Mac, your data can never leave this room. Your inbox, your calendar, your documents, none of it gets sent to a company. And that was the number one reason people say they want to run a local model in the first place. The second is that there’s no cap. There’s no 5h hour limit. There’s no weekly limit. There’s no message telling you to come back later. It just runs all day and every day as much as you want. And the third is the running cost, which is basically nothing. Apple says this Mac Mini uses 4 watts sitting idle and 65 watts flat out. Here in the Netherlands, electricity is around 26 cents per kilowatt hour. So sitting idle all month, that’s about.74. And even if it ran flat out 24 hours a day, the whole month, that’s about €12 a month. So here’s where we land. On the hard stuff, the paid brain wins. And it’s not even close. But on the private jobs, the repetitive jobs, and the things that run while you sleep, the free one isn’t just cheaper, it’s actually better. Okay, so say you want to do this. What do you actually buy? The new base Mac Mini, the M6, is $900. It’s faster than mine, but it still has 16 gigs of memory. So, it runs the exact same size models I just tested. Nothing bigger fits. The one that actually changes things is the M6 with 32 gigs. That’s $1,300. And that’s where the bigger, smarter models finally fit. the ones that don’t even load on mine. And the jump from 16 to 32 gigs is $400, which is two month of a $200 plan. And then there’s the M5 Pro chip with 64 gigs from $2,700. It reads memory about twice as fast, so everything runs quicker. But that’s for people making local AI their actual hobby. So, if you’re buying one just for this, 32 gigs is the floor. So, back to the question. Can a Mac Mini replace a $200 AI subscription? No, not on its own. And honestly, it was never really trying to. But here’s the smart money move, and nobody selling you a Mac Mini will tell you this one. Before you buy anything, just spend one week reading your own usage. If you’re not actually hitting your limit, drop to the 20 or the $100 plan today. You’ll have access to the same exact models you already pay for, but you’ll have more money back in your pocket with zero hardware or zero new things to change. Because what this Mac Mini actually is isn’t a replacement. It’s the best home for always on agents that you own like Hermes agent or OpenClaw. The brain that lives here handles the private stuff and the jobs that run on a schedule. And when a job is too hard for it, it hands it off for the paid brain for that one task. So tell me in the comments, would you trust a free local brain with your inbox or are you keeping the subscription? And I put together a free sheet with which model to run on which Mac, plus the exact settings I used. It’s the first link in the description. And if you enjoyed this video, then leave a like. Or if you’re new to my channel, then subscribe because I’ve got a ton more content like this coming your way. Oh, and would you look at that? The algorithm gods seem to think that you’re really going to enjoy this next video as well. So, click it and I’ll see you there.

Jeff is the most hyped AI model of the year and it’s not what they’re telling you. They’re calling it the next big model, a GPT killer. Some are even saying it’s about to change AI coding forever, but it’s none of that. It can’t write, it can’t code, it can’t hold a conversation. I was able to beat it at a game of checkers without even trying. So, no, this is not a smarter Claude, it’s not a GPT killer. What it is is a completely different kind of AI. And after this video, you will change the way you use every model you already pay for. Because here’s what they skip in the hype. Jeff doesn’t think, it decides in about 200 milliseconds and the output is basically free. The guy who built it is an ex-OpenAI researcher who worked on the tech behind ChatGPT. So, in this video, I’m going to show you what Jeff actually is, the 10 ways I’d actually use it and the places I would not use it just yet. Let’s get started. Okay, so what is Jeff? The easiest way to think about it is fast brain versus slow brain. Your slow brain is when you sit and reason through a problem. That’s Claude, that’s GPT, that’s every model you have ever used. They think, they write, then they explain. Jeff is your fast brain, the part of your brain that looks at a shirt and just knows that it’s blue. You don’t have to think about what color it is, you just come to that awareness in the snap of a finger. That’s Jeff for you in a nutshell. You hand it messy text and it hands back one of three things, a yes or no answer, a pick from your list or a score. And every answer comes with a confidence number. So, you know when it’s sure and when it’s guessing. And here’s the part that blows me away about all of this. Up until now, AI models have been racing towards which can be the smartest. And the smarter a model gets, the more time it can take to complete a simple task. I mean, try asking GPT-6 Astra to make thousands of tiny decisions, and you’re going to be waiting minutes and paying for every word it thinks. Jev, on the other hand, can’t ramble, it can’t hallucinate a broken format, and it answers in about 200 milliseconds. Type-safe, the company behind Jev, says it’s up to 193 times faster and over 400 times cheaper than a normal model. And its output tokens are free. So, this isn’t a model you chat with. It’s more like a smart if statement you drop into your workflow. And that one small shift just unlocked a world of possibilities. So, let me show you the 10 places you can use Jev to your advantage. The first place I would point Jev is at a pile I don’t want to read, like your inbox, for example, is an obvious choice. That’s because you can hand it thousands of emails and ask it three things at once, like this automation I created. I gave it all of my emails and asked it, “Is this a receipt? Is this a brand deal? Or is this a scam?” And in a click of a button, you can see how quick this is. It can sort through every single thing. There you go, done in 4 seconds, cost us $0.007, and it was able to sort through 69 emails per second. Nice. And this isn’t just for email. I mean, you can point it at your support tickets, your DMs, your YouTube comments, any stack of text you’d normally sort by hand or have an expensive AI sort by hand. It will do it better and faster than Claude, GPT, or any model. The second one is a bit of the same idea, but here’s something else that’s now possible thanks to Jeff’s speed of execution. You can have it create real-time filters on your browser feed, and it will tag every post as a page loads and break them down for you into different categories like breaking news, actual signal, or slop, for example. Like, look at this right now, likely human, 84%, likely human, likely human. We can actually, with this plugin, enable only AI slop. I only want to see AI slop, and we refresh, there you go. We’ll only be able to see AI slop on our feed, and we can do the opposite. We can filter out AI slop from our feed. And I mean, you can do this on Twitter, you can do this on LinkedIn, on Reddit. This is just this open-source tool, even other websites, but the sky is the limit here. And you can even have it filter out, quote unquote, slop from your timeline in real time. This is a tool someone open-sourced, by the way, and it flags AI slop on X, Reddit, LinkedIn, wherever. As always, if you want the link to all the tools and all the resources I share in this video, I’ll drop them all down on the first link in the description below. Moving on to use case number three, here’s where it can start helping you make money. If you have a lead pipeline or a CRM, say, for example, a lead fills out a form, Jeff can score it instantly on how good of a fit it is for you. So, here you can see with this lead, it’s a three out of three. With this one, it’s a zero out of three. And based on how much of a fit or which type of lead it is, you know, you can customize this as much as you want, and you can have template replies for each type of lead. For example, template A, say for example, it’s a great fit of a lead, it’s a three out of three, and they’re in the AI niche, then it picks template A. Hi Maya, this is exactly the kind of build I do. Grab a 20-minute scoping call here. Say it’s not a fit, for example, it’s going to go with template D, which is a no. Hi Aisha, thank you for thinking of me, I’m not the right fit for this one. See how quick this is? And we can keep going. If it’s a spam, it can just not answer at all. If it’s a type B, so it’s a marketing lead, for example, it can send them our marketing template. I mean, literally, what it can do is how creative you can get. Those are its limits, essentially, right? So, it can score leads for how well of a fit they are. It can see what type of lead that is in instances, and then it can decide which template is best for them in instances, literally in 300 milliseconds. There are so many ways you can expand and build on top of this. Now, here’s a fun one. Up to here, Jeff’s been sorting data, but you can drop it inside an agent as the thing that picks the next move. Someone built a flight booker where Jeff reads the page and chooses the next click every single step of the way. And it booked a flight in about 7 seconds. A normal model doing this is slow and pricey because it thinks out loud on every single click. But Jeff, on the other hand, it just moves. So, you can have Jeff control your computer for you much faster than any other model. I’ll also include this free tool inside the first link in the description. All right, this next one is cool. It’s a single text box, and as you type, Jeff figures out what you actually mean, and the box morphs into the right thing. If you type a date, you get a date picker. If you type a color, you get a color swatch. And it all happens in real time because of how fast Jev is in deciding and executing on every single keystroke. This is an interface you could not build before. Use case number six is you can have Jev power your website search bar. Instead of users having to type the exact keyword to find something on your website, you can have Jev on the back end guess what they are looking for and populate the answers in real time. Previously, this was possible, but it needed whole embeddings pipelines and a vector database. It was expensive and it was a pain to set up, but this skips all of that. It grabs the keyword hits first, then Jev reranks them by what you actually meant. And because Jev is fast and it can’t be talked out of its format, it makes for a perfect bouncer. And no, I’m not talking about a bouncer at the club. I’m of course talking about it being a bouncer for your AI agents. Say for example, you have a research automation for your AI agent. For example, an automation where every day you want Claude to fetch the top AI news. And the way Claude does it typically is it casts a wide net. It looks for things that might look like the answer you’re looking for and it summarizes all that data back to you. Well, instead of Claude being the one who decides what to scrape and what to discard, you can wire Jev in a way for it to be the primary bouncer on your sources. Meaning, the information gets to Jev first and it gets to quickly scan things like is this safe, is this useful, are we on goal or on target, is this the answer essentially that we’re looking for. So, if we test this automation out, we click go and it casts the wide net and look at this. Damn, this is quick. There you go. It scraped through 173 resources and it decided to hand 33 to Claude. I don’t know if you saw just how quick it was doing this, but then we can hand it to Claude and let Claude do the the work itself if you need the high thinking. And just look at the difference between how fast Jeff was and how much time Claude is taking to read just 33 items. Come on, Claude. Jeff was way faster than you. There you go. Claude finished its run, but all in all, um, with this automation, I tested one with just Claude and one with Jeff. And now that we put Jeff in the middle, in front of Claude, it spent 78%fewer tokens essentially. So, without the bouncer, we spent 29,000 tokens on this request. With Jeff as the bouncer, we spent 6,000. The junk gets caught at the door before your expensive model ever sees it. The next use case is for anyone who runs ads. You can feed Jeff a stack of your competitors or your niches ad copies and have it break down every single one of them. It can look for things like what’s the offer, what’s the angle, what’s the selling point, is it urgency, fear, a discount? And if we run it, then in a couple of seconds, you have a map of exactly what your whole market is running. Just like that. You can see the selling points that are working for them, save time, make money, easy, no skill. You can break down the types of hooks, the funnel stage, anything and everything you can think of in mere seconds. Hiring is the same thing here, but with higher stakes. You can hand Jeff every applicant and your criteria and it can score each one and cite the exact points that earn that score. One company, for example, put Jev in front of their recruiting search and went from minutes to practical seconds. And at the same time, the data was accurate. So, fast and accurate. You’re not letting it hire essentially, you’re letting it hand you the short list much faster. And finally, this is the one that makes everything we’ve covered so far click. You put Jev in front of your whole stack. Everything that comes in, it looks at and decides where it goes. Cheap stuff, send it to our automation bots. Questions to the thinking bots. Important stuff, send it directly to me. That’s the essential idea here. You stop paying a frontier model to read your spam and you only spend the big money on the things that actually need a brain. Now, here’s a fair warning. This is where all the hype on social media falls apart. Jev is dumb and that’s on purpose. It can’t reason. That checkers game you saw me play against it from the video’s intro, for example, it lost because it makes a snap call every move and forgets the one before. It has a tiny context window, a fraction of what Claude or GPT give you and the answers aren’t guaranteed right. The format is locked, but the concept is a fast guess with a confidence score. So, test it before you trust it. That rhymes. And here’s a kicker, the single most popular Jev demo online is from someone who used Jev to compact Claude’s memory. The post got thousands of likes, but it’s one of the worst things you can do with Jev because Jev doesn’t have the reasoning or the context to compact well. Same story with using it to judge another AI’s answers or to write text one letter at a time. You can have it do those things. It’s just that that would be a bad idea. It’s a snap decision model. So, give it snap decisions. So, here’s where I land. Jev is not a better model than Claude or GPT. And it’s not trying to be. It’s the cheap fast layer that goes in front of them. If I’m building anything that makes the small decision thousands of times, I would put Jev there and save the expensive models for the actual thinking. That is the unlock Jev gives us. It’s not the model that replaces your AI. It’s the model that stops your AI from burning money where it’s not needed.

If you’ve been looking for the one guide that takes you from zero to a full working AI agent, then welcome. You’ve come to the right place because in the next 30 minutes, we’re going to be setting up a Hermes agent together from scratch. You’ll have an AI agent that runs 24/7 on its own, texts your phone the second something matters, and teaches itself new skills while you sleep. And the best part is there is no code required. you can just follow along and copy paste every single prompt right away. And I know what you’re thinking. Isn’t this just chatt with extra steps? So instead of telling you it’s not, I’m going to show you instead because near the end of this video, you will watch it write its own automation and text me first with nobody at the keyboard. We’re going to be hiring your first AI employee. I’ll teach you how to give it a place to live, a brain, a phone number, a real job, and a schedule. And I’ll show you how to run the whole thing for free. Let’s get started real quick so you know what we’re building. Hermes is a free open-source AI agent. The software itself costs nothing. You only pay for the brain it thinks with and where it lives. And I’ll show you how to make both of those basically free as well. as the different paid options if you want to upgrade it as well. There are three things that makes Hermes agent different from a chatbot. Number one is it runs on its own. It can act on a schedule and message you first instead of sitting there waiting for you. Number two is it remembers. It remembers your preferences, your project, the way you like things done across every device. And lastly, it improves. When it figures out a task, it saves it as a skill and reuses it forever. That’s the difference between a chatbot and an employee. So, let’s give it a place to work. The first decision you’ll have to make, and it’s the one a lot of people overthink, is where does it actually live? And there are two roads you can take here, and one of them is free. The free road is if you’ve got a spare machine lying around like an old laptop, a mini PC, a Mac Mini, anything that can stay powered on, run it there. That’s what I personally do. Mine lives on a Mac Mini on top of my desk. It’s turned on 24/7, costing me basically nothing, but it’s electricity. But what can you do if you don’t have a spare machine? Well, you can rent one. Those are called cloud servers and they can be cheaper than your monthly cost of coffee. And you’ve got options here. So here’s the rundown. The easiest one to go with is anything that is a one-click host on the Hermes agent website. You can click on deploy to the cloud and then deploy to your cloud. Sign up and from there you can sign up to the Nouse portal which is exactly that. Another option here that I really like is Hostinger, which will install and deploy Hermes for you in one click of a button without you ever touching a terminal. They’re also the sponsor of today’s video, which is great news because that means you can use the promo code Charbell to get 10% off of your subscription. To install Hermes, click the link in the description of this video. Then, we’re going to pick a plan. Typically, anything with about 8 gigs of RAM is more than enough. So, we’ll go with the KVM2 plan. We’ll click on choose. To use the promo code, we’re going to click have a promo code. Yes, we’re going to put Charel. And watch this. Boom. Just like that, we have 10% off of our subscription. Then we’re going to uncheck anything that we might not need, like a readytouse AI. We’re going to be picking an AI model. Next, we are going to check instant web scraping because Hostinger offers a,000 credits for web scraping through Oxyabs. This is completely free if you select the 12 month or more option. Then you click continue. Then once we pay, that’s going to be it. We go to our H panel. You should be here at this point. I recommend you keep your username Hermes. Then for the admin password, you reveal it and you copy it and you put it somewhere in your notes on standby. Then you click deploy. It sets the whole thing up for you. And a couple of minutes later, you’re sitting on your Hermes dashboard. So let me show you that. Here we get this survey. You can decide to skip or complete it. And as you can see here, our Hermes instance is currently deploying. Once it is done, it’s going to give you a green check mark along with running as a status. We will simply click open. And this will lead us straight into our Hermes dashboard. We’re going to put our username and the password that we copied from earlier and click sign in. Just like that, you have made it in. Congratulations. Now, one fair warning before we go further. This agent can write files. It can run code and browse the web on its own and a web page it reads could try to trick it. That’s a real thing called prompt injection. So, two rules for you to carry forward here. Number one, give it its own machine when you can, not your main one with all your personal files on it. And number two, leave approval gates on, which I’ll show you how to set up later in this video, so that it asks before it does anything risky. But, okay, let’s install it. This part is thankfully very simple. On the free road with your own machine, you go to the Hermes site and you download it for your operating system and then you run the installer. That’s pretty much it. It will take you through the installation step by step. And once you’re done, you will land on this window, which is the Hermes desktop app. Now, I will be showcasing the rest of the setup through the lens of someone on a cloud server. But you can easily follow along by either following the flow of what I’m doing. Meaning, if your UI looks a little different, don’t be afraid. Just follow the flow of what I’m saying and what I’m doing on screen. Or you can run Hermes dashboard on your terminal which will open the same exact screen I will be on. So literally you can open your terminal and write Hermes dashboard. Click enter and boom you’re on the same exact screen on a one-click host. If you’re doing it through Hermes agents website then you sign up and pick a plan. I recommend starting with the free plan and then click on create agent. Then click create instance. Once it’s done, you can open the dashboard and there you go. You’ve made it. Whatever way you decide to install Hermes, you will land on the dashboard. This is your agent’s control room. Sessions is every conversation. Models is the brain. Craw is the schedule. Skills is what it taught itself. Right now, it can’t actually think yet because there’s no brain connected. So that’s next. This is the step most people get wrong. So, let me give you the way to think about it, not just what to click. Your agent’s model is its brain. It’s what it uses to think. Think of it like the engine of your car. Without an engine, your car can’t do anything. And you do not need the most expensive model for everything. You should think about AI models in three tiers. The first is top tier. And top tier is the boss. Those are the top smartest models on the market. Think GPT6 Astra or 6.1 Astra or Claude Opus 5.5. They’re extremely smart and pricey. You save these for the real thinking. If your agent has hard decisions to make, coding, anything where we want to avoid being wrong or where being wrong costs us actual money, then the second is the middle tier. And the middle tier is your workhorse. It has solid reasoning for a fraction of the price. It’s great for planning and for everyday jobs. It’s smart and capable, just not as smart and capable as the top tier models. For the middle tier, think of models like GPT 5.6 Soul or even Luna or Clawson Sonnet 5.5 or even Haiku. Bottom tier is the Grunt Workers. Those can be cheap or even free models. They’re fast and totally fine for doing tasks like sorting, tagging, summarizing, formatting, the stuff that doesn’t need a genius. So, if there’s a manual task for an AI to perform, these would be perfect for it. Here’s a helpful model tier list for you with all three tiers. Top tier, middle tier, bottom tier. Now, here’s the trap. A model being free doesn’t make it smart. Most of the free ones you can actually run are cheap for a reason. So, don’t hand them the thinking or at least test. see how they do before, you know, just throwing hard tasks at them. I recommend giving those the grunt work and keeping a strong model for the brain. In other words, match the model to the job. So, how do you connect one? There are a few ways. So, let me show you the cheapest first. For free models, a provider like Open Router has models with a free tag. They’re rate limited, so you get a capped number of runs a day, but they’re genuinely fine for testing and light background work. Then there are local and free models. If your machine can handle it, you can run a small model right on your own hardware with something like O Lama. Nothing leaves your computer and it costs nothing. Just know a small local model is a grunt worker, not a genius. Then we have the pay as you go models. I recommend you use open router if you want these models. You can drop in $5 and on a cheap model that lasts weeks. This is what I’d actually start on. And finally, here is a hack for you. If you’ve already or you’re already paying for a Chat GPT subscription, you can actually run OpenAI’s codeex subscription on that same plan. That way you don’t have any new subscriptions to pay for. To connect that you go to keys inside your Hermes dashboard. You choose to connect to your codeex subscription. Log into your codeex account. Copy the pairing code. Paste it in and ta you’re done. I am going to connect my GPT subscription for this tutorial so that I don’t pay for anything new. But if you don’t have a GPT subscription, then I recommend you start with any of the free models or the pay as you go models. If you’re picking a pay as you go model, I personally recommend you start with DeepSeek V 4.1 Flash. It’s a great all-around fast and cheap model. Kimmy is another great option. GLM’s another great option, but I don’t want to confuse you with too many model choices. Deepseek V4.1 flash is honestly a great one to start with. And finally, send it a hello. If it answers, the brain is alive and we can move on to step three, giving it a phone. Now, let’s make it reach you when you’re nowhere near the screen. In channels, pick Telegram, then scan the QR code with your phone, and it spins up your own private bot. Set that as your home channel. That’s where it sends you your updates. And that’s it. Just like that, you’re on the allow list so nobody else can talk to your bot but you. Now, here’s the hack I actually use, and it’s a big one for staying organized. Don’t just DM your bot inside of the one-on-one chat you have with it. Put Hermes in a Telegram group and turn on topics. Now you’ve got rooms for your agent. Split by job. You can have a general topic where you just throw whatever at it. Quick questions, quick tasks. You can have an email topic where it drops new leads and drafts. That one’s not for chatting. It’s a feed. You can have an alerts topic where it only pings you when something breaks or when there’s something that you want watched, where it pings you, hey, this happened. You can have a content topic, whatever you want really. That way you have one single agent with one unified memory but organized into rooms so nothing gets buried. This one thing makes it feel 10 times more usable. All right, it’s got a home, a brain, and a phone. Let’s give it a real job. This is one of my favorite jobs that I have set up on my very own Hermes. Every morning before I’m even up, Hermes sends me one message on my phone and it’s, “What’s your number one priority today?” And while I’m still in bed with my coffee, I just send it a voice note back. Something like, “Today, I want to build an email watcher that qualifies leads and sorts them into groups wired to Jev. Group A qualified gets template A. Group B, qualified gets template B. Disqualified gets the disqualified template. And if it’s just spam, then ignore it. And by the time I go to sit at my desk, Hermes has already started it. Sometimes it’s even already finished it. Imagine that. Starting your workday with the most important thing in your day already finished. I mean, to set this up yourself, you can send this prompt to your Hermes agent. Every morning at 7 a.m., message me and ask my number one priority for the day. When I reply, start working on it right away and get as far as you safely can before I sit down. Don’t spend money or send anything externally without asking me first. And that’s it. It’ll ask you the same question every single morning from here on out. You can also point it at your inbox to draft replies in your voice ready for you to approve and send. Or you can set it to watch something that matters like your web signups, a competitor, a price, and message you only when it actually changes. And by the way, if you want more high lever jobs, you can install into your Hermes. I made a free resource you can grab from the link in this video’s description. you’ll find a bunch of prompts you can copy paste and steal for yourself. And if you want more resources like this, then subscribe to my channel and to the newsletter. And this brings us to the coolest part of setting up a Hermes agent. Let me show you. All right, watch what happens when I hand it something to watch and let it figure out how. Keep an eye on my niche for anything new worth knowing. Build yourself a skill for the check. Keep track of what you’ve already shown me. Run it every hour and prepare a reel for me to film about it in my voice and only message me when there’s something genuinely new. I gave it that one file and it went and it wrote its own skill, set its own schedule, and a few hours later, my phone buzzed on its own. That’s exactly what makes Hermes agent so special. A chatbot waits for you. Hermes starts the conversation and sometimes even finishes it without your input. There are two quick things you can also do that make it feel like your agent, not just any generic one. First, it learns how you like things. Just correct it once. Tell it something like, “That’s too long. Keep it short. No bullet points.” And tell it to remember that. Now, it’ll become like that in every future chat you have with it, even brand new ones. You correct it once, not forever. And the more you use it, the more you will feel how personalized a Hermes agent can get. It’s one of my favorite things about it and why or what makes it so hard to switch to another system. And second are skills. When it does something well that you will want again, tell it to save that as a skill. if it didn’t already do that on its own and next time it just runs the whole process start to finish. That’s how it gets better every week instead of starting from scratch. Now, let’s talk money and safety because this is where people either relax or panic. Starting with money first, the server is a few bucks a month or free on your own machine. The model is the only real variable and it’s tiny because Hermes cashes most of what it rereads. $5 on a cheap model genuinely lasts weeks. A guard rail I would set though, always put a spending limit on your model provider if you’re going for a pay as you go model or a pay as you go service. So it can never surprise you. And for safety, put up approval gates on for anything that spends money. Send something to a client or delete stuff. You can do that by telling it in any of the chats. Let it do all the prep and you approve the final click. That’s how you hand it real work and still get to sleep at night. Once this clicks, there’s a whole other level. You can give Hermes a whole team, different bots, each with its own job talking to each other and handing off work between each other. I covered that in my bot mode video, so watch that one next if you want. You can also run separate profiles, a work agent and a personal one, so they don’t share the same brain. But honestly, don’t rush here. Get this one agent doing one real job first because that’s where it clicks. You can also watch my learn 95% of Hermes agent video if you want to master more of Hermes agent next. That video is more for intermediates if you want to level up later on. So tell me in the comments what’s the first job you’d hand to your agent. I want to see what everyone automates first. And I put my exact setup, all the prompts, the crrons, and my model picks into a free doc. It’s the second link in the description, so you can copy the whole setup. And if you’d like to use Hostinger’s one-click install, then don’t forget to click the first link in the description so that you get 10% off of your subscription. If you enjoyed this video, then leave a like. Or if you’re new to my channel, then subscribe because I’ve got a ton more content like this coming your

Anthropic’s own engineers just said they stopped building a new agent for every single job. And that sounds completely backwards until you see what they switched to building instead. Think about your phone. The AI model is the processor. The AI agent running it is the operating system. And skills skills are the apps on your phone. You don’t buy a whole new phone every time you want a new feature. You just install a new app and all of a sudden your phone can now do things it couldn’t do a minute ago. And skills are exactly that. By the end of this video, you will stop babysitting a pile of halftrained agents and start building one that gets better every week. So, in this video, I’m going to show you exactly what an AI skill is, how to build one for any job you want done in a couple of minutes, and where to steal the best ones other people already made. Let’s get started. Here’s the trap I fell into for a month. Every time I wanted my agent to do a job, I’d spin up a whole new agent or sub agent for it. I made one for content. I made one for emails. Uh made one for outreach. And every single one needed time, monitoring, training, enhancement. But the moment I started using and familiarizing myself with skills, everything changed. All of a sudden, all I had to do was teach the agent the job once, save it, and then have that skill at my disposal forever. When you learn skill building, you’re moving away from the need to build more AI employees and instead you’re giving one employee a bigger toolbox. And the best part is a skill is genuinely just a folder with a text file in it. So, let me show you what’s actually inside. A skill is essentially nothing more than one folder with one file, which is typically called skill.md. The top of that file has two lines that matter, the name and the description. The description isn’t for you. It’s for your agent. It’s how the agent decides when to reach for this skill. Then below those two lines, you just write the instructions in plain English, aka how you want the job done every time it does it. And from that moment onward, your agent does something clever. By the way, it doesn’t read every single skill you own on every message. That would burn through your context very quickly. It only reads the name and the description until your request actually matches one. Then it opens the full skill. This is called progressive disclosure, by the way, which is a fancy way of saying the toolbox stays closed until you need to use the tool. So, a skill has two jobs. To tell the agent when to use it and to tell it how to use it. Now, let me show you how you can build one in minutes. Okay, let’s build a skill together. For this video, I’m thinking we build a YouTube research skill, which is one of the skills I personally use the most and teach our clients to use as well. What I want the skill to do is find us viral pockets inside of YouTube and outline what the top five are in order of virality. If you create content on YouTube, this allows you to find great video content ideas. The way it works is we go to YouTube or a YouTube search engine like Vid IQ or one of 10, doesn’t really matter, and we search for the niche that we want to look for. So in my case, for example, it’s AI. Then what I typically do is filter by date. So publishing date, let’s put this week or this month. Then we filter by at least an outlier score of two. That’s my range. also view range above 50,000 and apply. Then we fetch all of the videos here that make sense for the niche we’re working with. And then we try to find the common patterns in formats and titles and topics. We’re essentially trying to see are there viral topics right now or within the last month. Are there viral formats? For example, how to fill in the blank in 30 days. That’s a format. We’re trying to fetch all of those and turn them into a skill. So, let’s do that right now. And there are three ways to build a skill. I’ll take you from the most manual way all the way to the one where you barely lift a finger. The first way to build a skill and also the way I would not recommend my worst enemy to ever do is to write it manually. I included it in this video because I think it’s important that you at least see what a skill looks like on the inside at least once because it’ll show you and help you understand the mechanics of a skill better than the other two ways we’ll ever do. In order to do this, you can click on the little plus sign uh then on skills, then browse skills, and we’re going to click on add and then create a skill. At the top of that file, we will include the name of the skill. Let’s call it YouTube pockets. Then the description, which remember this is how the agent will know when to fire that skill. So it’s important to there you go in the example use when asked for updates, right? So let’s write finds viral YouTube pockets. Use when asked to find viral YouTube videos. for example. Or you can even be more specific. Use when use when I say use YouTube pockets. I already have a skill that is exactly like it. So I don’t want to confuse my skills too much. I want to do the skill as a demo instead. Uh so I’m going to go with this. But if you’re creating a skill for the first time, then describe when or what action you would be performing for it to pull up that skill. And then here we describe how we want the skill to work or the instructions. Those would be the exact steps you wanted to follow when applying that skill. Think of it like handing a rule book to a new hire. I wrote down every time I say the keyword YouTube pockets, I want you to use Vid IQ and search through the niche I want. Filter out uploads older than 4 weeks. Also, filter out uploads with an outlier score below 2x. Then read all of the top titles that come back and find pockets patterns in the titles. For example, how to fill in the blank in 7 days or X ways to get rich using fill-in- thelank. Those are title formats. Before you write anything, ask me questions until you have everything you need. And I’m going to press create. And boom. We just created our first skill with its own skill.md file. The great thing about this is you can click on download if you want to share it with a friend or a colleague or teammate and eventually you can update it and have multiple versions of this. But I don’t know a single person who builds skills like this dayto-day. So, let me show you how I actually do it. Okay, let’s move on to way or method number two. This is the one I use for almost everything, which is we don’t write the skill. You describe the job and your agent writes the whole skill for you. And the whole thing lives or dies on the prompt you give to your agent. So, here’s a good prompt. You can send your agent when creating a new skill. Great thing is Claude has a starter prompt. If we go to skills, browse skills, add, and then create with Claude. Then it gives us this skill. So, here’s what we’re going to do. We’re going to input it, and then it’s going to start asking us about the skill. Now, I already have the previous instructions that we wrote. I’m going to send it to it, and we’ll go from here. There you go. It asked me a bunch of questions so that it really understands how that skill is done. And I’m going to go ahead and answer each and every single one of them. 1 to 12. All right, we just spoke a lot. I didn’t write this. I just spoke it out into existence and we’re going to see what it comes back with. All right, it has three more follow-up questions and then it says we should be done. All right, we answered its questions briefly. Um, I mean, it already had good guesses as to what would make for a good skill here, which is great because that’s what it’s all about with you using claw is, you know, it can think and reason what is the best way to get from point A to point B. And now we’re going to let it run and see what it comes back with. Here we go. It created the skill. I’m going to click on save with that. Let’s test it. Use YouTube pockets for the AI niche and let’s see if it works. Look at this. It’s taking all the right steps. Have no idea what these searches are, but that’s part of its process. It’s fine. Searching for the keyword. It’s ranking by top outliers. Then it’s validating them. So obviously, none of these videos would fit in our pocket. AI opportunities, AI robots, faceless AI YouTube channel. Yay. There you go. So it found the most viral pockets inside of AI right now, which is expected. AI danger explained calmly. This is very nice. Highest potential uh with title suggestions. A reasonable person’s guide to how AI takes your job. That is very grim. The dark future of AI agents. I can easily see this being in the millions if that is your niche. Want to be an AI commentator. Number two, contrarian AI opportunity takes. The biggest AI opportunities aren’t where you think. Why the AI agent boom is just getting started. These are great new titles for your channel. Five insane and this is the third pocket by the way. Five insane AI robots in 2026 you’ll want at your desk. It actually gave us four pockets even though its requirement is a minimum of two but it didn’t just stop at the minimum. This AI history niche is exploding. This faceless AI niche is exploding right now. So those are four different pockets meaning four different subniches within AI as a niche itself as the mother niche as I like to call it. and it gave us great video suggestions for just that. This is exactly how I work on both my videos but also our clients videos to come up and ideulate what videos would work best for each and every one of their niches. Just like that, we just created a skill by telling Claude itself to do it for us. Now, what if I told you there was an even lazier way to do this? And honestly, it’s a pretty cool function. And finally on claude there is a if we go to browse skill click on add there’s a record your screen option which if we click on it we can start recording ourselves as we’re doing the thing so that cloth can literally learn step by step how we want something done. Now, this skill isn’t maybe the perfect skill to do it because I wanted to use an MCP like Vid IQ to access the data. But say you want data extraction, you want very specific things that you typically do manually by hand, then you can record yourself doing it and it will learn every single step of the way and it will see what you’re doing. Okay, he opened this folder on his desktop. Then he created a new file and he named that file this. And he got this name from that place. And inside that file, he put in all of these different other files that he downloaded off of his email. So it will learn every single step of the way that you previously did manually. And then it will be able to recreate that for you. That way you never have to do it again and you can just ask it to do that process and it will do it for you. Just know that this record feature is Claude’s own. If you’re on another agent like Hermes agent, then way two or method number two is your main move because a good prompt works literally anywhere. And once you can build a skill, the real shortcut is not having to build them at all. Now, here’s the part that saves you the most time. You do not have to build every skill you need yourself. always always look up if the skill already exists or not. And there are two places where you can find skills. The first is the official one. You click on the little plus symbol, you go to skills, you go to browse skills, and from there you can find a whole array of skills that are free for you to download and install onto cloth. There’s a ton here that are absolutely gold. But as you can see, you are limited to, I don’t know, over 200, let’s say, skills. The second and much richer place to look for skills is the community. There’s now an open standard for skills, which means one skill folder works across Claude, Codeex, Cursor, Hermes, you name it. And there are full marketplaces built around it. GitHub is by far the most popular place you can find a ton of free and open- source skills. You can use the search bar to find anything you’re specifically looking for. Say for example, marketing skills and there you go. But GitHub can be very keyword search specific. This is also where you can find a lot of my free skills that I’ve open source like a Hermes agent mission control dashboard. uh X bookmarks where you bookmark things on X and then your Hermes or your OpenClaw agent goes and reads them for you at the end of the day and actions the things that you’ve bookmarked. There’s a ton that I’ve open sourced. You can also go to the trending section of GitHub which will show you all the not just skills but all the repositories that are trending on GitHub. Now for skillsp specific libraries or repertoires there are skill marketplaces like claude marketplaces.com it has a ton of great skills. One of the skills is called find skills which is literally a skill that you install and then it helps you find skills. So when you say things like how do I do X or find the skill for X or is there a skill that can it will trigger and it will find or search for the skills for you. There’s skills MP skills marketplace as well. There’s LoHub. There’s a ton of skill marketplaces. And to install them, it’s simple. Say, for example, I want to install one of my skills. Like my most popular one, I think, is the Hermes agent mission control. You just copy the link to that skill. You go to your claw chat, to your Hermes chat, whatever it is. You paste it in and you ask it to install it for me. A quick disclaimer though, always always before you ask it to install, but before you do, read the entire skill and scan for anything malicious or any malware. Always do that because these are third-party skills. You never know what a stranger is uh trying to do because skills run code or scripts and you never know what they’ve injected into them. So, always have your agents scan and make sure that that skill is safe for you to install. Most skills are fine to install, but it’s always better to be safe than sorry. And just like that, you don’t need more agents anymore. You need one good agent and a shelf of skills. You build the three skills you use every week. Then you steal the rest and supercharge your claw or any other agent you use with all the skills it needs because you don’t need to buy a new phone for every new task. Comment which skill you would build first. I’m curious to see what people use claw for the most. And I dropped my starter skill.md template plus the full list of places to find skills plus all the prompts that I used inside of this video in a free doc. The link is at the top of this video’s description, so you can copy the whole setup right away. If you enjoyed this video, make sure to leave a like, and if you’re new to my channel, then subscribe because I have a ton more content like this coming your way. Oh, and the algorithm gods just told me you’re very likely to enjoy this next video as well. So, click it and I’ll see you there.

Clot code just dropped mods and it is the biggest update since skills. Here’s what a mod actually is. If there’s something you wish Clot Code did or had, you just tell it and it builds it into itself. Whether it’s new panels, new buttons, new commands, new screens, mods essentially make your cloth code as customizable as you want it to be. And in the last few days, people have built hundreds of mods. Most of them are junk. A few of them I’m never turning off again. So, in this video, I’ll show you the five mods everyone should be installing. Some of them will save you money, and I’ll show you how to build your own in about 2 minutes. Let’s get started. Okay, so real quick, Cloud Code announced mods last week. Here’s the announcement itself. A mod is a tiny plugin that runs inside of Cloud Code, not around it like a skill or a connector, but inside of it. Here’s an example of one of the mods I’ve installed running literally on my Cloud Code screen so it can watch what Claude’s about to do, change it or stop it cold, and it can draw its own stuff right onto your screen. You install them the same way you install any plugin. Uh you can type in /plugin or open the plug-in menu and you can turn them on or off accordingly. And they work inside of two places. The Claude Code app, the desktop app, or inside of a Claude Code terminal window. So, let me show you the first one I’d install before anything else. This one’s called Next Steps, and it’s from one of Anthropic’s own engineers, and I absolutely love it. And again, to install it, literally all you need to do is copy this and paste it into your cloud code window. And this is what it does. Every time clock finishes a turn, I’m going to run a prompt just to show you. A little panel pops up at the bottom with three suggestions or what to do next. You don’t even have to type anything. It understands what you’re doing and based on that, it suggests what you can or should do next. You just click whichever one you want. So here it’s suggesting me one of these actions that I could do next based on the prompt I’ve given it. I can select let’s say this one and boom it’s past it as prompt and all I can do is just press enter and it goes. This one’s called cache control. Let me show you what it does. So here’s the thing most people don’t realize about cloud code. Your whole conversation is cached. Every time you send a message, you’re not paying full price to resend the entire chat as long as that cash is warm. But it only stays warm for about 60 minutes. So if you step away from your clawed session and come back 60 minutes later, then you pay full price for that entire conversation. Again, this mod right here, ash control, puts a timer right on your screen. Right now, I’ve got about 55 minutes before it goes cold. So I know I can fire off a message now or just let it go. And I want to show you how customizable these are. So if I want to, for example, edit a mod like this one, I can just ask it in plain English. So let’s do I want to add a clear cache button to my cache control mod so that when I click it, it clears cache and starts fresh. And it sends the chat a handoff brief. And it also shows us how much of our cash is still free as well as how it’s being used. I’m gonna let this run and come back in just a minute. All right, so it’s done just that. So I’m gonna put in reload plugins. Reloaded plugins. And now if I type something in, there you go. We now have a reset button built right into our clot code too. So, if I’m about to run out and my context is getting heavy or if I’m about to pass my 1h hour context timer, I can press just one button and it clears the conversation for me and starts one with all the relevant information from a clean state. So, here you go. I can press it. Let’s test it out. Handing off and boom, handoff arrived. It cleared the session. It pasted everything it needed to know from the context before. and we’re back on a clean slate on a big session. That cash going cold can cost you a few dollars every single time. This just saved you a few dollars. But money is one thing. The next one is about your sanity while Claude’s actually working. Number three is called plan progress and you install this one straight from a marketplace. You know how when Claude’s working, your screen just fills with code and tools calling flying by and you genuinely can’t tell how far along it is. This kills all of it because instead you get a clean checklist and a clean progress bar with a percentage and it shows you where in the process it is. Is it auditing, researching, planning? There you go. Look at that. It’s progressing and we can see the steps ahead as well. rank the opportunities. We can see it complete them one by one and how much of that task is left. I mean, this is a small thing, but when you’re running something that takes a few minutes, being able to glance and know it’s at like 70% done instead of frozen, that is the difference between trusting it and having to babysit it essentially. And this mod also does something that I absolutely love. When Claude is done with a task, it makes a little popup noise like And if you use cloth code a lot, you know how often you’d be typing something into it, going to something else, going to a chat window. There you go. I don’t know if you heard it or if it was recording, but imagine that. Imagine you’re working. you’re going to a meeting or you’re typing on Slack to your team and you have to keep scrolling back and forth and back and forth between your cloud code window and the other thing that you’re doing and you never really know when a task is done. This just fixes that just by making that small helpful sound. Now, every mod I’ve shown you so far, somebody else built, but the real power here is building your own. And it’s easier than you’d think. So, let me show you. Here’s the part that makes mods different from everything before them. You don’t need to find the perfect one for you. You just describe what you want and Claude builds it into itself. Claude Code already ships with a built-in skill that teaches it how to write mods. So, you literally just talk to it. And here’s the best way to get started. You don’t have to guess what to build. You can let Claude look at how you work and tell you. So you can paste in this exact prompt right here. Audit how I actually use clot code. Read my last 20 sessions and find three things, the questions I keep asking you for the commands and how I check your work. Then suggest five mods that would fix those or enhance my workflow. Show me the ideas first. Do not build anything until I pick one. Here we go. And a minute later, we got this along with the questions we keep asking, the commands we keep giving, how we check our work, and the five ideas. Now, look at all of these mods coming in together. The progress bar lets us know, you know, as soon as the prompt is done, and then the next steps mod gives us the suggestions. We can literally just press let me go through them. This one is very helpful, the prompt deck, the prompts that I keep using the most. I’m actually going to go for that gopher number four. I love it. And with that, it’s going to build me that mod just for me, a custom mod for myself. And once it’s done, make sure to tell it to save it and install it. That way, it’s live and available on every single session and every single time that you open Claude. Just made the sound so I knew to come back and check in. And there we go. Just like that. Absolutely. the install prompt tech permanently. There we go. Just like that, it gave me V1 of this. What I would do to even enhance it more is make it pop up somewhere like here uh in a better UI cuz right now what it did and I like V1 because it gives me the templates that I would otherwise have to go into my notes, copy, paste, so on and so forth. Okay, back to the list because this next one is one of the most important mods in this whole video. Number four is called Blast Radius, and this one’s from Anthropic as well. Here’s the problem it fixes. Cloth code can run real commands on your machine, which is amazing, right up until it decides to run something on the wrong folder and deletes something you actually needed. Blast Radius steps in before any destructive command runs. So, the second Claude’s about to delete something or force push, it stops and it shows you exactly what’s about to disappear, like we can see in this photo. And then it waits and asks you to proceed or cancel. But this is one that nobody thinks they need until the day they really, really need it. It’s essentially a seat belt. You put it on before the crash, not after. And the last one on this list is pure quality of life. So, let me show you. Number five is called the image viewer. So, normally if Claude generates or pulls in an image or a screenshot, you can’t actually see it in the terminal. You’ve got to go dig to the file and open it for yourself. This mod just lets you actually see the images just like that. I’ve went ahead and installed it. And if I take a screenshot, for example, here, paste it in. There we go. We can see it the moment that I put it in. Look, I know this is a small mod and it’s mostly for the terminal window, not the desktop app. But if you touch images at all in your work, not having to leave the window adds up faster than you’d think. And before I wrap, there’s one more I have to show you just because it’s ridiculously funny. Okay, honorable mention. This one is completely useless and I love it for that reason. The same creator who made the previous mod actually built a mod so that every time you’re sitting there waiting for Claude to finish a task, it drops you into a multiplayer game of Doom. Let me go ahead and mute it with other people who are also just waiting for Claude to finish their task. I mean, that’s the whole point, right? It’s your clot code now. You can make it do whatever you want. So, just have fun. For the first time, clot code isn’t a tool you take as is. It’s a tool you shape. I’ve been waiting for this for month now and it’s finally here. So, tell me in the comments which one are you installing first. Oh, and all of these five mods that I’ve shown you, their install commands and that exact audit prompt. I’ve included all of these for free in the first link inside the description. And if you’ve enjoyed this video, then leave a like. Or if you’re new to my channel, then subscribe because I’ve got a ton more content like this coming your way. Oh, and would you look at that? The algorithm gods have decided that you’re very likely to enjoy this video next. So, click it and I’ll see you there.

Every AI money video makes it sound like there is a secret button to print money with AI. I have used AI inside my businesses for years. I’ve used it to increase our company’s revenue by 50,000 a month. I’ve used it in ways that have lost us thousands a month. So, in this video, I’m ranking the only ways I would actually consider making money with AI in 2026. Not the easiest sounding ones, not the ones gurus are selling us, the ones that I’ve actually done myself and that connect to real obvious demand. I’m going to rank each one on a tier list as well. S tier means I would actually build this. A tier means it’s a strong opportunity. B tier means it can work, but there are caveats. C tier means possible, but not my favorite. and F means I think you should stay away from this. By the end, you should know which AI money methods are real and which are not. Let’s get started. The fastest way to make money with AI is usually helping a business find customers. This is why AI lead generation is first on my list. Every business wants more pipeline. They want better prospects, cleaner lists, and more money. That’s what we all want. And AI can help with all of that. What this essentially means is you either selling services to businesses where you help them find and acquire clients on X, LinkedIn, email, so on and so forth, or you doing this with your own business in order to increase your own sales. The important part is that this should not just be scraping 10,000 emails and blasting garbage outreach. That is what I like to call spam. The real value is using AI to understand who the company is, why they might need your help, and what message would make sense for you to send them specifically. For example, my business partner developed a lead generation scoring automation and uploaded an entire video about it that I find very interesting because the core idea is not just that it finds more leads. It helps our sales team better prioritize who looks like a good fit, who is showing intent, who should get a different message. That is where AI starts becoming super useful. If I were starting this myself, I would pick one single niche or one single offer and one single source of leads. Then I would build a simple system that finds prospects, scores them, gives me the reason they might care, and drafts a first message I can improve. The trap in this method is thinking volume fixes bad targeting because it doesn’t. AI makes it easier to send more bad messages, which means it can only destroy your reputation faster. My verdict on this is a tier. If it is niche specific and tied to a real offer, it can be S tier, but generic AI lead genen is already crowded. Finding the lead is only step one. The next question is, what happens when that lead replies, fills out a form, or lands in your inbox? Onto method two. A chatbot only makes money if it saves a human from wasting time on the wrong lead. I would call this category entirely chat bots, but I think that word makes it sound a lot smaller than what it actually is. The real business here is lead qualification and routing. Here’s what I mean. Someone emails you, someone fills out a form, someone asks a pricing question or replies to a cold email. You typically avoid all those emails because you think they’re just a waste of time. But all of these are actually sales opportunity waiting to be seized. The AI’s job here would be to understand what this person wants, qualify them, route them, and help the right human respond faster. We already have versions of this internally. And I have a fun story about this actually. About 2 months ago, my YouTube channel reached a size where I was receiving a ton of emails from sponsors requesting to work together. Most of these conversations would go nowhere. So eventually I stopped answering. Then on my way to my vacation, I asked my Hermes agent, “Hey man, can you take care and respond to emails for me? Only message me if you need approval for a dollar figure.” I didn’t think much of it at the time or honestly think much would happen. Then a few days into my holiday, I decide to check in on Hermes and I see he’s asked me for approval on rates for a lead that’s interested in our business services. Hermes was asking me if he could send him a link to our calendar for a meeting and I approved it. Then my team come back to me a few days later and tell me, “Hey, by the way, we just closed the client that you sent us.” Me asking Hermes to just do that for me took me 5 minutes to set up before my holiday and made us an extra $12,000 in 1 and a half weeks. That is real value. That is much more valuable than a random website bot saying, “How can I help you today?” Because businesses do not pay for chat bubbles. They pay for speed, clarity, and lots and lots of fewer missed opportunities. You could again build something like this for your own business or you could sell it as a full service. If you can guarantee people results and correlate this to money, you’re good to go. My verdict for this is a tier. It’s not as sexy as AI agents, but it is as close to revenue as you can be. and very easy for a business to understand. Once you can find leads and qualify them, the next level is building the actual AI workers that do jobs inside the business. Method three, look, businesses do not just wake up wanting an agent. They want a job done without hiring another person. That’s what they want. That is why AI agent development is one of the strongest opportunities on this list. The market is moving away from give me a chatbot and it’s moving towards give me something that handles this for me every single day. That could be a sales research agent, a reporting agent, customer success agent, an internal operations agent. The value is not that it is AI. The value is that a job gets done with less human drag. This is also where I think a lot of people will mess up. They will say they can build any agent for any business. That sounds flexible, but it is actually a nightmare. Every client becomes custom. Every workflow becomes messy and every edge case becomes your problem. The better version is to pick one type of agent for one type of business. For example, a client reporting agent for agencies. that’s it. Or a sponsor inbox agent for creators or a lead research agent for B2B teams. The reason this works is simple. If the agent saves a team 10 hours a week, it or prevents missed revenue or helps the business move faster, there is budget for it. The trap is selling the technology instead of the outcome itself because nobody cares that you use the fancy model or framework. If the workflow still breaks every Tuesday, my verdict for this one is S tier, but only if it is specific. Generic AI agent development is going to become a race to the bottom. Specific agents that solve painful business problems, that is the opportunity. But not every AI business has to start with code or agent infrastructure. The next one starts with attention. AI content only makes money when it creates trust, leads or sponsorships. This is where people confuse the tool with the business. AI writes content is not a business. A business is helping a founder, creator or company turn their expertise into demand. That can mean exposts, LinkedIn posts, YouTube scripts, newsletters, lead magnets, sponsorship decks, sales pages. But the content has to do something specific. It has to create trust. It has to create a pipeline. It has to attract sponsors. And it has to move the business you’re selling it to forward. This is the logic behind one of our services, Founderfunnel. The value is not just generating posts. The value is taking messy founder thoughts, sales calls, client wins, product lessons, and turning them into content that sounds like the person and creates opportunities. This can also become a sponsorship business. If you build an AI personal brand and attract the right audience, sponsors will pay to reach that audience. But again, the audience has to trust you. AI can speed up content production, but it cannot fake trust and authenticity. If I were starting this, I would pick one input and one business outcome. For example, take a founders’s weekly voice notes and turn them into posts or their weekly meetings with their teams or their outreach angles or one lead magnet. Then measure whether it creates replies, calls, sponsors, or sales. The trap here is just like you’re going to notice with the other ones is generic AI copyrightiting. The internet does not need more average posts written faster. It needs sharper content, better content, and a real point of view. My verdict for this one is A tier. It becomes S tier if it has taste or proof and a real business outcome attached to it. Content can create demand. But there is another version of AI content that removes the person completely and that is where I get a little bit more skeptical. Faceless AI channels, they can work, but this is only where AI slop goes to die. I want to be fair here. People are making money with faceless YouTube channels, AI voiceovers, AI avatars, short systems, and essentially content forms. AI makes this easier than ever. You can generate scripts, you can generate voices, images, clips, captions, entire videos with tools that barely existed a few years ago. So yeah, this is a real method. I would be lying if I said nobody makes money from it. But I would not personally build my whole future on a anonymous AI slop. The mode for it is very weak and so is the brand you build from it and the platform risk. And if the whole advantage is that you can make cheap content fast, someone else can probably make cheap content even faster or even cheaper. The version that can work is not lazy, faceless automation. It’s a real media operation. Strong topic selection, strong packaging with real research, good editing, and a reason for people to come back. And at that point, AI is just production leverage. If I were doing this, I would not create random motivational videos of an AI generated Trump or generic AI listicles. I would pick a specific niche where the viewer wants utility like finance breakdowns, tool comparisons, or educational explainers. Then build a repeatable format with actual quality control. The trap is thinking faceless means effortless. Faceless removes the face, not the need for actual good content. My verdict here is F tier for most people. uh C tier if you have good editorial taste, a strong niche, and a lot of patience. It is possible, but it’s not my favorite. The next method has even more hype, and honestly, even more danger. Method six, AI trading. It’s useful when it helps you research faster, but it gets quickly dangerous when you treat it like a money printer. I love playing with AI trading bots. I have had for months where the systems helped me make one or $2,000 a month, but I have also had months where the edge were off, the market changed, and I ended up losing money. That is the part a lot of AI trading content does not want to admit. A trading edge is not a magic spell. It decays with time. other people find it and the thing that worked last month can stop working next month. The version I like a lot more is AI assisted research. It track markets, monitor smart money wallets, label opportunities as copy candidate, watch list or skip. This is especially interesting in prediction markets like poly market because there are real signals to analyze. But the system should start paper trading first. It should help you make better decisions before it touches real money. As much as I love trading bots, my verdict for this is B tier for most people. It has high upside and high interest, but it is risky, inconsistent, and not where I would tell beginners to start with their AI journey. Trading is the risky version of AI leverage. But the next method is the boring version. And boring is usually where the reliable money is. We love reliable. The most reliable AI business is often a service people already buy delivered faster with AI. This is one of my favorite categories because it does not require convincing the market to want something new. The market already buys content. It already buys clips. It already buys sales support, operations, editing, lead generation, whatever. AI just lets you deliver some of these faster, cheaper, or better. Founderfunnel is a great example. The offer is not we use AI. The offer is a business outcome, which is better founder content, more consistent output, faster production, stronger distribution. That distinction matters. If you sell AI, people compare you to other tools. If you sell an outcome, people compare you to the pain they already have. This is also easier to start than a full software company. You can begin manually. Use AI behind the scenes. Build your workflow. Learn what clients actually want, then slowly productize the process of delivery. If I were starting from zero, I would pick one painful service category and ask, “What part of delivery can AI make three times faster without lowering the quality?” Then I would package that into a simple monthly offer. My verdict for this one, this is going to have to be the first S tier. This is one of the best places to start because it connects AI to existing demand. Once a productized service is working, the next question is how you operate it without drowning in the work. That is where AI assistants come in. The best AI assistant is not the one that chats with you. It is the one that keeps the business moving. This is where tools like Hermes Agent and OpenClaw become interesting. Not because they answer random questions, but because they can run recurring workflows. remember preferences. They can use tools. They can manage task and they can operate across the systems you already use. A real AI chief of staff can do morning briefs, check your emails, access all relevant files, check on clients, help you with content, and so much more. The money here is often indirect, but it is still real. If it saves you hours every week, prevents missed follow-ups, finds opportunities faster, those are all related to value indirectly. The mistake is treating it like just a random chatbot, just like you would with chatbt or clot. A chatbot waits for you to ask questions. A chief of staff agent has routines, boundaries, skills, and approval gate. If I were setting this up, I would not start with 50 automations. I would start with one daily workflow. What should you know every morning? What needs a draft? What needs approval? And what should never happen without your approval? The trap is giving the assistant too much access before you trust the workflow. This is why I recommend you start with read only, then earn autonomy slowly. My verdict for this is a tier for most people, S tier for operators. If you already have a business, clients, content, inboxes, this can become serious leverage. And that leads to our final and most underrated AI money method. Using AI not to sell AI, but to grow the business you already have. Method nine. The most underrated AI money method is fixing the revenue leaks already existing inside your business. This is the category that does not always make the best thumbnail, but it might be the most valuable one. Every business leaks revenue. Leads do not get followed up with. Clients go quiet. Renewal dates get missed. Churn risk shows up a little late. AI can help with all of that and more. This is where systems like my client poll bot become interesting. It scores clients, analyzes clients for churn, tracks unanswered messages, renewal reminders, upsell signals, the list goes on and on and on. This is different from selling AI services to other people. This is using AI inside your own business to make more money from the machine you already have running. And honestly, this is where a lot of operators should start. You do not need a brand new AI business if your current business is already losing money through slow followup or weak retention or missed opportunities. If I were applying this, I would answer these three questions. Where do my leads get lost? Where do my clients churn? And where does my team forget to follow up? Then I would build one automation around each one of these. The trap is automating random tasks instead of revenue bottlenecks. Just because something can be automated does not mean it should be. My verdict for this one is S tier. It is not the flashiest method, but it is the most operator native. If it saves one client, creates one upsell, or prevents one missed deal, it can pay for itself fast. So, if I had to pick one lesson from all nine methods, it would be this. The best AI money opportunities are the ones closest to real demand.

And here we have it. Here is how I would rank every single one of them. In S tier, we have AI agent development, AI productized services, and AI revenue automation. In A tier, we have AI lead generation, AI lead qualification, AI content engines, and AI chief of staff agents. In B tier, we have AI trading bots. In S tier, we didn’t end up with anything. And in F tier, we did end up having faceless AI content channels. Look, making money with AI in 2026 is not about finding a secret prompt. There’s no secret prompt. Trust me, it is about finding a painful business problem and using AI to solve it faster and packaging the result in a way that people already understand. Don’t lose them with AI gibberish. The closer you are to revenue, the better the opportunity or the easier it will be for you to sell it to clients. This is where AI actually becomes valuable. If I were starting from zero, I would not start with the flashiest method. I would start with one workflow, one niche, one painful problem and one clear outcome. That is how you avoid the common guru trap. You do not need AI to make money magically. You need AI to make something valuable happen more consistently.

You’re building something in clot code, but something on the page looks wrong. So, you crop a screenshot, you save it, you switch windows, you paste that in, and then you write a paragraph explaining which button you actually want to fix. You send that message and it changes the wrong thing and it also breaks something else entirely. You’ve done that today and you’ll do it again tomorrow. Cloth code can read your entire codebase. It can reach your email, your Slack, your notion, but it can’t see the thing you’re looking at right now, and it can’t be reached from anywhere except a chat box. What if I told you that there’s a way to solve both of those problems with one tool? This is exactly what I’m going to show you in this video. We’re going to give Clot Code eyes and ears across our entire workspace. I’ll also cover what the tool is, where it falls apart, and how you can start using it for free. So, let’s get started. Okay, so let’s be precise about what cloth code is actually missing. Cloth code is not trapped in your folder. If you use its connectors, aka its MCPs, it can reach your inbox, your calendar, your notion, and your database. So, reach isn’t the problem here. The gap is much smaller than that. and a little more annoying. It has no eyes. It cannot see what’s on your screen right now. So, anything that’s visual, you have to translate into words. And words are often a terrible format for that box. The third one, no, the text under that. And it’s got no doorbell. You can’t reach for it whilst you’re on another window, for example. You can summon regular Claude from anywhere, but not the world’s most powerful coding agent. Instead, you have to open the claw desktop app, go to clot code, be in the right folder, and ask for the thing you want. That problem right here is exactly what I’ve been looking to solve for months. I get a lot of emails from companies asking me to look at their product. And honestly, I end up saying no to almost all of them. But I recently got an email for the tool I have literally been looking for for a while now. And here’s why that is. The clawed app on your desktop has voice. You can talk to it. But I barely use the clawed app. I mean, I live inside of clawed code and clawed code has no voice or no screen access. Nothing that remotely looks like that. So, I’ve had the thing I wanted sitting one app away from the thing I actually work in for months. So, I agreed to take a look at the product and I have been testing it out for the last 2 weeks now. So yeah, this video is sponsored by VoiceOS. They gave me complete access to the tool for free and told me I could be as honest as I want about it, which is great because you’re going to see me cover this tool’s best features, but also its downsides. The catch is that this is a paid closed cloud tool. You can get started with the 7-day free trial and decide if you like it from there. It is $30 a month if you pay monthly or $12 a month if you go yearly. And if you cancel after the 7-day trial, then you don’t lose access completely to the tool. You drop down to 100 dictation sessions and 25 agent sessions a week. That’s about three agent runs a day, which is enough to keep using it, just not enough to work the way I do in this video. I did include free alternatives at the end of this video in case that’s what you want. However, each have their own limitations which I will also cover there as well. But with that out of the way, let’s actually install it and test it out together. Installing it is simple. You go to their site, you download it for your operating system. You open it and you make an account. That is the whole install. Then there are two things to turn on before any of this works. The first is the apps tab. That’s where every connector lives and you turn them on one at a time. Each one opens an ooth window. You approve that and it’s connected. You turn on clot code. Then you click into it and you connect that. The connector list covers the most common apps like mail, Slack, notion, drive, calendar. But if your workflow lives in something a little niche or if the app you need is not already built into that list, you may need to add or build an MCP integration, which we will get into how to do that later in this video. And with clot code connected, here’s the feature I actually made this video for. Now, this is the part that changed how I work with clot code. You hold the hot key like this. You can see it activated. And while you’re talking, you can circle something on your screen that it will give as context to your prompt. And that circle that we drew, as you can see, just became context and just became the thing that it’s pointing to visually inside of your prompt. So instead of saying something like change the color from red, the one below that text where it says you can, what you can literally do is say, hey Claude, could you change this red text to yellow for me? changing the red text to yellow. And that’s it. Just like that, it sent it its request. It opened a clawed instance, a claude code instance, and it’s currently doing it. This is a big improvement from taking a screenshot, opening the desktop app or the terminal, pasting the screenshot inside, and describing the specific thing you want to change. Here’s what I absolutely love about this feature. The hotkey works everywhere. You’re on your website, and you notice a bug. For example, over here it’s not showing the thumbnail for the next articles coming up. I can literally just hold the hotkeys, circle around the problem, and give it my prompt. Hey Claude, could we add thumbnails next to each title here on my website? We’re on that. There you go. It just sent its request and it doesn’t just fire and forget. There’s a panel on the left where every run lives. There’s another panel at the top where you see your current run. You can see which model it’s on. I can click into it. I can see which model it’s on, which intensity it has. I can modify that. I can ask it to stop. If I don’t want it or I don’t know if it misunderstood me, I can look at the prompt and reread it. And you can even hold the hotkey midun to ask it to change something. Actually, never mind. I wanted to design a specific thumbnail instead for each video. Just like that. So clot code stops being a place you go and sit and it becomes something you throw work at while you carry on doing something else entirely. And once you’ve got that going, you can then go one level further and build integrations for any tool on your desktop. Let me show you how. Another thing I love about this thing is that if you can’t find an integration for a tool that you use, you can just create one. For example, I love Hermes agent, but there isn’t an integration for it there. So, I went to custom integrations and I built one myself. I clicked on create and then I just explained what it is I wanted to do. Hey, I want to create a Hermes integration. And then I had a little back and forth with it until this was done right here. This means you can build integrations for this tool to be able to reach any tool with a usable API or MCP path so that you can use that tool with voice OS from anywhere on your desktop. Now I can just run Hermes and ask it something like Hermes, what tasks did you complete today? One sec. Checking task. I have sent the question to Hermes agent and the summary will appear in your panel as soon as it finishes. There you go. And it will just answer me from anywhere. I don’t need to go to Telegram or the Hermes desktop app anymore when I’m busy doing something else. Hermes completed email audits and partner replies, updated notion metrics, rescored exposts, checked YouTube competitors, and verified all 10 jobs, while demo assets remain outstanding. There you go. It’s that simple. Building custom integrations for more complicated apps isn’t one shot, though. For example, I built a calculator integration and that was literally as simple as asking it to do that. But creating a custom Hermes integration took a bit of back and forth given my Hermes instance runs on a separate device and I want to be able to control it from an entirely different device instead. By the way, I will include the custom Hermes integration I built for free in the description along with everything else I cover in this video. And with that, let’s talk about the places where this tool can fall a little short. Every tool has a breaking point. And after I tested this tool for the past 2 weeks, here are mine. The first and the biggest one in my opinion is that it’s cloud only. There’s no local inference option. If you’ve built your whole setup around running things locally, this might bother you as well. The second is the connector list. It’s good on the obvious apps, but it’s thin outside of these. The good thing is I literally only needed one more integration, which is Hermes agent. And then I was able to just build a custom one for myself in under an hour. The third is that it can sometimes mishar technical words like I would say Hermes and it would hear her mess. But luckily, there’s an easy fix for that. You just go to the app, go to dictation, and add all the keywords you typically use so that it familiarizes itself with them and hears those more accurately. And this brings us to the question I’d be asking if I were watching this video. What are the free alternatives? There are free ways to get some of this, and I thought I’d do some research and dig into each one. I do have to warn you though, every one of these alternatives has their own limitations, which I’ll be covering as well. The first one is chatpt’s desktop app which can also use voice mode for codeex. If you already use chatpt or codeex for your coding, you can enable the voice chat hotkey. It can even see your screen while you’re talking to it. The only downside of that is you can’t point at specifics on your screen using that and it doesn’t really solve the problem if clot code is the coding agent that you use. The second one is already sitting inside cloth code. You just click on the microphone button and say what you want to say, but you have to already be inside of cloth code. You can’t activate it when you’re on another window and you can’t show it your screen at all unless you manually screenshot and paste that into the chat window. The third is voice mode. It’s an MCP server that gives clot code voice conversations. It is MIT licensed and it can run speech to text locally. So your audio never leaves your machine, but it is voice only. There’s no screen watching or screen sharing and getting it set up takes a little bit more work. So here is the honest comparison. The free options are strong on voice. What none of them can do though is hand that specific piece straight from your screen onto a clot code run that’s already working inside your project. So, after using this tool for the last 2 weeks, here’s what I think about it. I have noticed myself reaching out for it a lot more frequently. I mean, it’s just so convenient. And I understand the price is not going to be convenient for everyone watching this video, but at my scale with the business that I’m running, it’s a drop in the ocean compared to what it lets me actually do and how convenient it makes working for me. I’ve even started using it beyond Clot Code and Hermes. I’ve noticed myself reach for it and ask it to send me emails whilst I’m editing a video or set me a task on notion whilst I’m on a meeting. I’m even using the dictation mode whilst I’m on random tabs because it almost feels like a real life Jarvis to me. So, if you’re in clot code every day and a lot of what you do is visual, then take the free trial and try it out

A few months ago, I ran one AI agent. Now, I run a team of them, and between them, they handle most of my business. My inbox, my clients, my content, all run by different bots that talk to each other and hand off work on their own. This is Hermes Bot, and it’s the closest thing to a cheat code I found. So, here’s the plan. First, I’ll break down what Hermes Bot actually is. Then I’ll walk you through building your own team of bots step by step. And at the end, I’ll build a new team of bots right in front of you so you can see how to set one up firsthand. Let’s get started. Okay. In plain terms, a bot is a profile. Those two are the same thing. Hermes has always had those profiles. Every bot you create has its own persona, its own model that you can connect it to, its own tools and access, and its own memory. So, you’ve been able to build those for a while, but now with bot mode, we get the ability to interact with different agents or profiles a lot more easily. The bots can also talk to each other. You can hand one a job. It breaks it up and passes the pieces to others on its own and they can sort it out and talk to each other in one shared room or a group chat. So you stop being the guy running work between these different bots or chats and that is the unlock. It was never the bots, it’s the coordination. Before we build anything, you should know that there are three ways Hermes bots works together. The first is one-on-one where you just talk to a single bot. The second is direct where one bot messages another. And lastly is through a group chat where a few of them work inside of one thread or one group chat with a cap on how many times they can go back and forth so they don’t loop forever and drain you or your tokens. All right, but with that, let me show you how to actually build a team of bots. I always recommend you start with building one bot, just one. Then once you understand how to do that and understand the mechanisms behind each bot, then the sky is your limit. All right, step one to create a bot, you hit new bot and you give it two things, a name and a title. The title is what shows up on the side. The name is what you use to tag it with an at symbol. Then you just describe its job in a sentence or two. And here’s the nice part. From that little description that you give it, Hermes writes the bot’s whole personality for you. You can clone from your default so it inherits your setup, you can pick a model and that’s a bot. That description becomes its soul, its personality and its rules. You can open it up and tighten it whenever you want. So on my sales agent for example, I added lines like never discuss prices unless I approve and always hand the result back to my main Hermes instance. Never straight to me. The tighter you make the job, the better the bot will behave and abide by the job description that you end up giving it. Now the step almost everyone skips and it’s the one that matters the most, especially for saving money. I’m of course talking about tools and permissions. There are three types of tools here. The first is skills. Those are saved workflows and the nice part is they only load when the bot actually needs one. So they barely cost you anything just sitting there. So you don’t need to worry about turning these off. Then there are tools. Those are live actions your boss can perform like reading a file, searching the web, sending a message and those load into every message. And then there are MCPs. Those are external tools that plug your Hermes onto external tools like Gmail, Notion or GitHub, but those also cost you a ton of money if all of them are turned on. Every bot starts with every tool in MCP switched on, and you do not want that, or rather every tool in MCP that your main bot is uh connected and turned on to. The more tools your bot has enabled, the more context it will use for every run it has. Some tools have upwards of 50,000 tokens that load into every single message you send. So if a bot does not need access to a certain tool or MCP, then disable it. You give each bot only the tools its one job actually needs, and you turn the rest off. This will make your bots faster, cheaper, and more efficient. This is also the step that takes a bot from a decent bot to something that is genuinely good because the MCPs you set for each bot are extremely important. Think of it as handing one bot a specialist power that the others just don’t have. For my sales agent, for example, I wire it into our sales CRM. This gives it access to edit, to search, uh, and to sort through all of our leads that no other agent has or for that matter needs. It’s pulling from a tool built for that exact job. You add it once in that bot’s capabilities, and from then on, it’s part of its kit. One bot, one upgrade, and your entire bot team’s output just simply gets better. All right, while we’re in settings, there is another setting you can change to save money, and it’s setting each bot’s model. Your orchestrator, aka your your main bot, aka the bot making the calls. You want to put that on a strong model. Think of putting one of your strongest and most capable models here because it will need to be able to make decisions and think through complex reasoning problems. The grunt work bots or the worker bots on the other hand, drop them to a cheaper model or run them on a local one on your own machine for free. You will essentially get the same exact result on the simple stuff for a fraction of the cost. And that’s mostly what keeps a team like this cheap instead of pricey. You’re only paying for the smart model where it actually matters. All right, two more quick final settings to adjust here. For memory, let only your orchestrator write to your shared memory if you have a shared memory provider like Honcho or Memo or Super Memory. And for every other bot, switch that to built-in memory only. We do this because our main orchestrator already saves the important information to the shared memory. So there’s no need to double save it and waste context for no reason. In settings, you can also choose to bump how many bots stay warm from a three to a five. This will make it so switching between bots is instant and they won’t need to reboot every time you open more than three bot chats. And finally, timeout time is by default set to 10 minutes. I personally recommend you set it to 20 or 30 minutes depending on how big your workflow or how much time your workflow typically takes. This will give your bots a longer idle timeout so a long job doesn’t quit or error on you halfway through. Finally, this is the last step to make an ultimate Hermes bot team and it is defining anything you never want an agent doing without your input. For me, those are three things. spending money, sending something to a client, and deleting stuff. Those are the things I always personally want a hand on. Everything else, it can just run itself without my approval. So, before you finish building your team, make your own list. Ask yourself, what could a bot do that I’d actually regret? Those right here, your answers, those are your gates. It will be different for everyone, so be honest about yours. Once you have your gates, you have three places to set them. The first and the strongest one is what tools a bot even has. If a bot doesn’t have the tool to send an email, it can’t send an email, no matter what it decides in the moment. So, the cleanest gate is just never handing a bot the keys to something it should never touch. The second is Hermes own approval setting. You can put it on smart and it’ll auto run the safe stuff and block the genuinely dangerous stuff and stop and ask you on anything in between. When it asks, you get four choices. Allow it once, allow it for the whole session, always allow it or deny. And the third layer is inside the soul.md file itself. You can write the rule straight into the bot’s personality. You can say something like never send anything to a client without my approval. Just know this one is a softer layer, meaning the model will try to honor it, but it’s not a hard wall like the other two that we just covered. So, use the soul to reinforce, but never as your only lock. Setting those guard rails is exactly what lets me hand the team my real business and be able to sleep at night. They run free, but the stuff I’d regret is gated behind a yes from me. Okay, enough setup. Let me put everything we just learned together and apply that. I’m going to build a content team that consists of three bots. My main Hermes agent runs point as the orchestrator or the operator. Then we’ll need a researcher whose only job is going to be to dig into what’s working right now in my space and a content creator that takes what the researcher finds and turns it into actual posts or content. So, let’s start by asking our main Hermes instance to do just that. Hey, Hermes, I’d like for you to help me build a team that will post and manage content creation on an Instagram page for me. Uh, you can ask me as many questions as you would like until you reach clarity. Ultimately, I’m thinking we could have two Hermes bot employees. The first one, a researcher that monitors what is working in that specific niche. And the second one, a content creator that will go ahead take that information and create content posts, graphics in the approved styles that we end up choosing and the direction that we ended up going for. Ultimately, I want you to be the one telling me um or relaying the information back to me. I don’t want to be speaking to the other bots. I want you to be my main point of contact. Can you help me out with that? And we’ll send this out. All right. So, I’ve given it the context. It’s asked me a bunch of questions and it’s currently starting to delegate and plan it out. Got it. I’ve locked it in. It has the brand. Please send me the exact Instagram URL. And it asked me a few questions. I’m going to go through this onboarding process, as I like to call it, and come back once we’re done. Would you look at that? It literally just ran through a set of tasks. Oh, researcher hit an error. That’s a very nice start. It ran through a set of tasks and it created them. I’m It even created the group chat itself. Uh to which I’m I’m just saying hi to my new team. Never mind. They did end up replying. I waited like 5 minutes and none of them replied. So, I went to check that my Hermes is still live. It was. I’m not sure what happened, but there we go. Hey everyone, what are we cooking today? Let’s go. Just as simple as that. Turns out what happened was our provider timed out, which is GPT. Uh we reached our uh session limit. Uh so I had to quickly reset that. Now we’re back to being up and running. All right, so it finished setting them up completely. It created the group chat. Everything is done. The one final check that I want to make before we get started with this workflow is I want to inspect the soul that it gave it mission. Very nice. It gave it a mission, a reporting line responsibilities. This is a very well-defined soul. Much better than what I could have done. Uh, and it’s on our subscription, which is good. It’s plugged into a cheap model, so it doesn’t end up costing us a fortune. Now, for skills, we don’t really care too much about that. Remember, what we want to do is turn off any tools or MCPs that that bot does not need. This is the creator bot. The creator bot will need access to Instagram. So, I’m going to connected to Instagram and disconnect it from anything it doesn’t need. Then, I’m going to go to the researcher and do the same thing. Uh, add and remove whatever tools, MCPS, and plugins it needs or doesn’t need. and then I’ll be back once that is done. All right, done with that. I’m going to ask them team let’s get started. Now what you can do in that group chat say you want to reach out to the researcher specifically then you type at researcher and just like that you can tag it so that it knows this is for them specifically. I can tag the creator. I can tag all everyone or I can tag the orchestrator operator specifically. And there you go. Just like that, it’s following our instructions, which is Hermy the orchestrator/operator is the one I’m talking to. He’s coordinating everything. And he said, I’ll coordinate and quality check the final package at researcher. So, it’s delegating to the researcher. Surface today’s strongest evidencebacked psychology angles. and at creator it’s delegating to the creator turn the best one into a sharp practical concept once the research land I realized I didn’t brief you on this but I used to run a psychology page called daily psych on Instagram a few years back and the job that I gave them was to brainstorm and come up with a plan of content angles that they can create to revive the page and just like that the researcher is doing his job, the creator, strong handoff. They’re talking to each other and you could just sit back, watch, come back in a few minutes and see what the back and forth was like. And this is great because they understood the assignment. They came up. So, this is the the chat. They had a back and forth. The researcher came up with the research, handed it off to the creator. The creator conceptualized how they can make that into content. Then they sent it to Hermy. Hermy, here’s the finished carousel draft. And Hermy, relate this back to us with a caption, with a style. All I asked was great. Now you can create the post and run them by me here. A nice thing you can do over here is if you have multiple workflows running, you can create groups. So I’m going to call this daily psych. And inside that group, we’re going to move, there you go, all of the bots that are meant to be in this group. That way we can stay organized around all of our different groups. So say you have multiple content pages, you have multiple, I don’t know, businesses, workflows, uh, levers that you’re pulling, you can organize them that way so that they’re much easier to navigate through. Sorry, but this back and forth that you can watch your bots just go through is insane. It’s crazy how fast this tech and innovation is moving. Essentially with that, this is what they ended up creating, which I absolutely love. This is the first slide, the second one. So, they created the carousel, sign two, sign three, try this tonight, and a call to action. Save this for the next hard day. And it’s insane to watch the back and forth they’ve had. Once the creator finished the content, he sent it to Hermy, said, “Hey, I’m done.” Hermy then tagged the researcher and said, “Hey, like check it or doublech checkck it uh for anything that might be off or not evidence-based.” The researcher checked it and said, “Hey, no research objections in this copy packaged as drafted and Hermy came back and said, “Hey, I’m done.” That is insane. Here’s the part that makes this entire process compound. After a job wraps, we can tell the team to save what it just learned. So the next time I ask for the same kind of thing, it already knows my format, my style, the sources that I trust, and it leans on me way less. At the end of the day, you’re not just running a team. You are training one. This is how you eventually get to it, asking you less and less questions with time and need less and less of your input. So here is my advice for you. If you have a one-off job, don’t even bother with a bot team. One good agent can do that job. I mean, you don’t need a team of agent to do one simple task. For example, like manage an email. One really good bot is more than enough sometimes. Don’t go build a whole team just to answer one email is what I’m trying to say. But the things you do every single week, managing your content or your business or your weekly workflows, that’s where building a team of AI bots really shines. a handful of specialists, each locked to their own job, checking each other’s work and running while you sleep. That’s the whole point. The best part is that Hermes is free and it’s open source. So, the only thing this costs you is setting it up, aka your

GPT6 Astra might be the smartest model ever made, but it’s also eating people alive. Most people are burning through their entire Chhat GPT subscription in minutes on this thing. So naturally, people are scared to even use it, and I get it. But once you understand how the model works, you can cut most of your token consumption out. So, in this video, I’ll show you exactly why Astra burns tokens so fast and the fixes that actually move the needle in order of what saves you the most money. I am saving the nastiest one for last because there is a setting that’s literally doubling your bill without you even knowing. But, let’s get started real quick because you need to know this before we get started. These models have no memory. None. So every time you hit enter, your entire conversation gets packaged up and sent again from the top. So your first message is cheap. By your 20th, you’re repaying for everything above it every single time. You’re paying for every message you’ve sent before and every message it has sent you back as well. So your consumption compounds. Astra makes this worse for one reason. It thinks more. All that extra reasoning is extra tokens you’re paying for, even for the things that you never see on screen. That’s essentially the tax for it being this smart. The good news is you control most of it. So, let’s start cutting away. Before you change anything, you can start by finding out what’s actually draining your tokens because it might be different for different people. In chatbt and codeex, you can either ask it something like what’s my usage like right now and what’s eating it the fastest. It will tell you how much of your limit you’ve burned through when it resets and which model and settings is draining you. Alternatively, and I really recommend you do this, you can take a screenshot of the prompt on the screen right now and send it to your GPT session. or for the sake of convenience, that exact prompt and the other two prompts I use in this video are in the first link in the description. So, you can just go there and copy them straight away. Run that first prompt first and you’ll know exactly which of these next five fixes matter the most for you. And once you know your leak, here’s the biggest lever and almost everyone gets it wrong. Astra has a dial for how hard it thinks. There’s light, medium, high, extra high, and ultra. And everyone slams it to max thinking that they’ll get a better answer, but you usually don’t. You just end up paying more. And these numbers aren’t mine. Artificial analysis benchmarks these models independently. They give the models the same set of tasks and measure their results versus their token consumption. On low effort, Astra costs about 80 cents a run. But if you crank it to the max, that same exact run or work costs $326. That is four times the money for the same exact job. And on their coding benchmark, max effort runs about $7 a task just to tie a model that costs a fraction of that. You are paying a premium for a score you can basically match for much cheaper. And this isn’t just me. Open AAI’s own guidance says lower effort doesn’t mean lower capability and that cranking the effort higher won’t always give you a better result. So I highly recommend you start on medium and not the default, which is high. Only if the answer you’re looking for is not landing, then bump it up. Don’t start at the top and then work your way down. Instead, work your way up. And while you’re in there, turn off fast mode unless you actually need the speed for some reason because that one alone doubles the rate you pay. Now, the next one is where you save the most and it’s about not using Astra for everything. Here’s the shift I want you to move towards next. Don’t run every single task on Astra. I know what you’re thinking. I came for this video to know how to use Astra more and this guy is telling me use Astra less. But just hold on and bear with me. Astra is as smart of a model as we get. I’m not saying don’t use it. I’m saying be selective on where you absolutely must use it. Use it like the smart senior developer it is. Let it plan and review for you and then give the grunt work to cheaper models. That way you get the best of both worlds. You get the thing you want Astra for on and then you get the rest the grunt work, the things that any model can do in the world done for a much cheaper cost. This is how I’d set it up. The boring scoping and cleanup, uh, push that down to the cheapest model. You can use a lighter open AI model like GPT 5.5 for this. The planning and the checklist you can hand to a mid-tier model like GPT 5.6 Soul or Luna. Then Astra only touches the hard part at the end, the actual judgment call on medium effort. Now, I wouldn’t recommend you switch these by hand every single time. Instead, you can set it up once and let it route itself automatically. Codeex lets you build sub agents. And there are two ways to set them up. The first is super technical, but you basically create a TOML file inside your project folder that looks like this. The second way is much easier. You just paste these three prompts into your codeex chat. Two of these prompts create sub agents for you. And the third one tells your codeex agent when to call out each of those two sub aents. That way you’re paying for the genius ones on the 10% that needs it instead of every single step. This one change is the difference between a video that cost me 10 bucks and one that cost me one. And speaking of what you feed Astra, there are two things that quietly bloat every single message. Your tools and your output. When you connect a tool, its whole instruction manual loads into your context before you even type anything. A big one can carry tens of thousands of tokens on every single message. So, you can click on settings, then plugins, then look through your list and turn off anything you haven’t used in over a month. People have taken a 55,000 token load to 3,000 with that one small simple change. Then there’s output. You tell it to install something, then you have 800 lines come back, but you don’t see or read any of it. And now you’re repaying for those 800 lines on every message after. There is a free tool called RTK, you point codecs at it, and instead of dumping raw command output into the chat, it hands back a stripped down version of that. I ran it on my own setup. a giant file listing went from 7 million characters down to about a thousand. Two things worth knowing. If you install it on codecs, you have to tell it to actually use RTK. It won’t grab the output on its own. And it only helps on that big messy output, not the fixed cost every message already carries. There’s also a desktop one called Headroom that does the same job across cloth code and codeex and its maker says it cuts costs by about half. Now let’s move on to a few small habits that stack up fast. When you want to start a new conversation with codeex, don’t just type in your new prompt in an existing conversation window. Start from a new context window altogether. Remember, you don’t just pay for the message you send. You pay for every message you’ve sent before along with every response you’ve gotten before, every single time you send a new message. So when you’re done with a workflow, always start fresh. Start from a fresh conversation. Another very important one, pick your model and effort at the start of the session and leave them because the moment you switch models or effort levels mid task or mid-run, none of your cash conversation matches anymore and it reprocesses the whole thing at full price. So, the thing people do to save money, switching to a cheaper model halfway through is actually the most expensive move on the board. And also don’t paste in whole repos or entire old chats. Keep your scope tight because that’s exactly what can bloat a context window. All right, now here’s the one tip I promised you at the start and this one will shock you. This is the one almost nobody talks about. Astra’s got this huge million token context window they love bragging about, but here’s what they don’t put on that same slide. The second your input goes over about 272,000 tokens, the price doubles. Input goes to two times its price and output to 1 and 1/2 times its price. And it’s not just the tokens over the line. It’s the entire request. All of it gets repriced. So, that giant context window is a bit of a trap because if you fill it up past 272,000 tokens, you walk straight into doubling your bill with zero warning. That’s why I’d recommend keeping an eye on your working context and trying to keep it well under that line. If you want to keep an eye on your context window, you can type / status and always be able to look at how much context you have left and how much you’ve spent inside one session. And before you go, let me kill the fixes that don’t actually work because a couple of them are the opposite of what you’ve been told. Number one, writing shorter prompts. That doesn’t change much to be honest. What you type is essentially a rounding error. It’s basically nothing. So don’t bother typing shorter prompts. Number two, compacting to save money. That one’s a little backwards because to summarize your chat, it has to send the whole thing one more time. So the thing you do to save tokens is the most expensive message of that entire session. Only use slash compact for continuity if you absolutely need to continue a chat, not for savings. Number three, screenshotting text instead of pasting it because a screenshot can cost you a couple thousand tokens. The words in it costs way less, so paste the text instead. And lastly, PDFs. You pay for every page of a PDF twice. once for the model to read the words and once for the picture of the page that the model takes. Convert it to a plain text file first and then send the doc instead because that same doc cost you only a quarter of what it would have as a PDF. So look, Astra isn’t the problem. It’s the smartest thing out there and it’s genuinely worth using. It just doesn’t come with a manual. But if you do these six things, at least it’ll stop rinsing you and you actually get to use the good stuff without watching the meter every single second or minute. The setting that will save most people the most money is making Astra the boss, not the worker. So start there if you really want to feel a