I built an entire finance department inside of Cloud and it is comprised of several AI agents as real employees. There's a revenue manager, a spend manager, a forecast manager, and a reporting manager. Now, each one it's just a markdown file sitting inside of a folder and Claude runs all four. I also pointed at 18 months of raw invoices and expenses for an agency. It closed the books. It built a 12 month forecast. It wrote the owner's report and it turned everything into a board deck. And in the middle of doing all of this, I caught a duplicate invoice that I never had to tell it about. I'm going to be building this in three different parts. Number one, the folder and the one file that keeps every single role telling the same story. Number two, the four roles and how they actually pass work to each other without me touching anything at all. Number three, just one prompt that runs the entire department's end to end. And then we change a single number and we watch the entire plan rebuild itself. Now, even if you've never built anything with AI or automations or done any sort of coding or have any background inside of that, you don't need it. You can follow this because all we're doing is we're just writing in plain English files and talking to Claude. And the thing that makes it all work isn't a model or any specific prompting. It's just one simple rule inside every single file or each role writes its output into the next RO's inbox. Now, the way that most people actually use AI on their numbers, it's typically one prompt for revenue, maybe another one for expenses, separate chat for the forecasting, and now you have yourself maybe three answers built on three different readings of the business. And the revenue figure in one, it doesn't match the revenue figure in another one. And beyond that, each new chat, it's going to be drifting just a little bit further from the last one. And then next month, when you actually need the same analysis on any of your fresh data, you have to start over from a completely blank text box. Now, with this, the stakes are very real because underneath all of this, the numbers still live in spreadsheets. Now, incurring even one significant spreadsheet error, that can cost a company or a firm about thousands of dollars on average. And over half of teams, they hit errors every single week. So, this build, it works differently. There's four different roles, each owning one specific job. Now, all of them, they read the same company context. So, every output, it ties back to one version of the truth. So, you can run them separately or even as a chain. and the numbers they're always agreeing and when new data actually does inevitably come in nothing gets rebuilt. So the same rules they just run over and over again. Now one thing to be clear on this does not replace you. It does not replace your team. It does the execution. You just keep the decision. So, keeping all of your partners, all your highvalue employees from working on the low value work, getting them to work on the higher revenue generating activities inside of your firm, so you can actually grow without needing to hire or, you know, even fire other employees. So, there's going to be four different roles with one shared context, just having one coordinator. That's the map for this entire video. And every box on it is going to be real and running by the end of this video if you follow along until the end. Let's set it all up. All right, so this right here, this is the entire department. And so we just have one folder called the finance team. Now inside of this there's just obviously one folder per ro. And each ro folder it has three separate things. So it comprises of an instruction file. It also has an input folder and also an output folder. That is it. There's no code anywhere inside of this. Now beyond that there's also this shared folder. So this is what makes everything else work. Just three distinct files. So we have one the company context which describes the business. So what copper veil creative is we have the three retainer tiers what leadership actually genuinely cares about. And just a quick note on that copper veil it's just a demo agency that I have built out specifically for this video not to you know showcase any of our clients or any sensitive data. And with this there's about 18 months of realistic invoices and expenses. It's all dummy data but made to be very realistic. It's mock data. So that's just because real client numbers they don't obviously belong on YouTube. So however for your business in this file you just have to describe yours very simple. And then on top of that we have our assumptions. This is holding every forwardlooking number. So anything like the growth targets the renewal rates the hiring plan and then we have a calendar file. So when one roll says maybe Q3 every other roll means the same three months. Now every roll it's going to be reading these before it touches a single number. Now that

So, Claude can make a bad idea look really good. You ask it to make something, and 20 seconds later, it looks like a final version. There's a clean structure, confident wording, and maybe even a nice layout. So, you start reacting to what Claude made before you really even asked, is this the right thing to do in the first place? And that's the trap I want to talk about here. I'll show you why it happens in three simple ways to catch it before the first draft decides too much for you. So, let's get into it. One of the great things about AI is just how much leverage you can get from it and the speed associated to that leverage. So, you can produce a report, a presentation, even an application in just a few minutes or an hour that would have taken days or weeks. But this amazing superpower we all have is a double-edged sword. There are pros and cons, and oftentimes people ignore the cons. And my guess as to why they're ignoring them is either they're unaware or the downfall of these cons is delayed and is only noticed after a few months of leveraging AI in this way. So, what am I talking about here? Well, with AI, it can produce things quickly. When it produces something quickly for you, if you haven't put enough thought into it, you're anchored on that first draft. Here's a simple visual that kind of depicts that. So, we have the AI that created something for us. So, it could be a idea, it could be a presentation, a report, or really anything else. In that first version we look at, and we're anchored on that idea. So, anything going forward, any variance of this are going to be based on this output. The reason this is an issue is if you haven't put a lot of thought prior to the AI going in this direction, this is going to be the average of what the AI comes up with. So, whatever output you come up with, you're not going to stand out against your peers or your competitors. Now, what's important the steps I'm going to walk you through here and the things you can leverage to make sure the AI helps you create something that's truly novel or has good taste associated to it isn't for all use cases. You don't have to apply this all the time because it is time intensive, and it is somewhat counterintuitive to the usage of AI cuz I'm going to actually tell you to slow down in your usage, which sounds crazy. But we're doing this intentionally for specific use cases where our judgment and our taste will matter most when it comes to the AI's output. So, hopefully, you're convinced or at least understand why it's important to slow down with AI so we can create something that's truly novel and not just what everybody else is going to get from AI. Now, the first thing we're going to do, of three different steps you can take, either you could take all three or a subset of these, is we're going to ask the AI to give us multiple options before it gives us the final version of whatever path we're going to take. Because by doing this, we then get the opportunity to choose from what the AI provided. Now, when it gives us options, the options aren't going to be the final version. And I'm going to give you two prompts that you can actually copy and paste and use yourself here. So, the first prompt is around making something, the second one is around judgment. Because the AI doesn't just create stuff, but it also gives us ideas. In both of these paths, we want to make sure there's some judgment and taste prior to actually getting those outputs. Quick pause in your regular programming. This video is brought to you by me, as always. Two quick things. First off, below is a 30-day AI insight series, completely free. You'll get 30 insights in your inbox if I can apply AI to your business and your work. The second thing is if you'd like to work with me, below are a series of offerings to see if there's a good fit between the two of us. Now, let's get back to the video. So, the first thing of the AI creating something for us, so a report, presentation, whatever else, you can use a basic prompt like this, where you're simply telling the AI that I want you to give me three ways to tackle this specific direction, making sure that each of these ways you provide are mutually exclusive and collectively exhaustive. This is just a fancy way of saying I want you to make sure that your whole list of what you provide back is extremely diverse in the options there. And we're saying for each option, I want you to give me the best case for it, as well as where that specific case could break. So, give me both the strengths and the weaknesses. Now, it's important in this prompt that I've said three ways to tackle a specific direction. This is important. I've not asked the AI to give me three final versions of a report, a presentation, or anything like that. I'm saying general directions that we could head before we start the final version of this. Because with that tactic, we can choose that direction without getting distracted of the polish and the fine 

Claude Code is now unlimited and completely free. Here's how to set it up. The tool is called Omni route. It's free on GitHub and it sits between Claude Code and the models just like a switchboard and it's completely private. It's wired into 290 providers and over 90 of them hand out free tokens. So the second that you burn through your 5-hour window, your session does not die. It just hands the job to the next free model and it keeps building like nothing happened. It also squeezes your prompts before they leave your machine so the same task it costs up to 95% fewer tokens. Now set up, it's just two lines in your terminal. Type npm install -g Omni route and then you just type Omni route and that's it. It points Claude Code at itself and you just open Claude Code exactly like you always do. Now will the free model write code as well as Opus? No, but it will grind through the boring 80% while you save Opus for the hard parts. And your session will not die mid-build again. Comment need and I will send you all the resources that you need to set this up.