STOP Building AI Agents. Do THIS Instead.

0:00 Everyone is building AI agents right now. Every startup, every tech company, every developer on Twitter. Agents, agents, agents. And it's the hottest word in AI. But what if I told you that Anthropic, the company that actually builds Claude, just said, "Stop building agents." Not me, them. The people building the AI are telling us that the future isn't more agents, it's something they call skills. So, in this video, I'm going to break down exactly what skills are, why they matter, and why I think this changes everything, especially if you're someone who's been trying to figure out how to actually make AI useful for real work and not just demos.

0:34 And then, I'm not just going to explain this in theory, I'm going to show you a real project I built using Claude code skills, which are a set of skills that can take any website, run a full AI search optimization audit on it, and generate a professional report you can charge clients for, all powered by skills. So, let me first set the stage. Why would Anthropic, a company that literally builds AI, tell people to stop building agents? Here's the thing, AI agents right now are incredibly powerful. They can write code, browse the web, call APIs, reason through complex problems. The raw intelligence is there, no one is disputing that at all. But intelligence is not the same as expertise. Think about it like this, imagine you need someone to do your taxes, right? You've got two options.

1:16 Option A, a genius, someone with a sky-high IQ who can figure out anything from first principles. They've never done a tax return before, but they're brilliant, they'll work it out. Option B, an experienced tax professional. They have done thousands of returns, they know every rule, every edge case, and every deduction. So, who are you picking in this scenario? You're picking the tax professional every single time, cuz you don't want a genius figuring out the tax code from scratch, you want someone who already knows it. And that's exactly the problem with AI agents today. They're brilliant generalists, they can figure things out if you give them enough time and the right guidance, but they don't come preloaded with specific expertise. They don't know your workflows, they don't know your industry, they don't remember what worked last time. So, what does everyone do? They build a separate agent for every use case. A tax agent, a legal agent, a marketing agent. Each one with its own custom tools, its own setup, its own architecture. And it's exhausting. It does not scale at all.

2:13 What Anthropic realized is that that agent underneath is actually universal. You don't need a different agent for every job. What you need is a way to give the same agent domain expertise on demand. That's what skills are. Okay, so let me explain this as simply as possible. At its core, a skill is a markdown file. Basically, a document of instructions that teaches Claude how to do something specific. Think of it like a playbook you would hand a new employee. Here's how we do things. Here are the steps.

2:41 Here's what to watch out for. That file lives inside a folder, and that folder can also include other things like scripts that Claude can run, reference documents, templates, examples, whatever Claude needs to do expert-level work in that area. But the heart of every skill is that instruction file. So, at its simplest, you can create a skill in 5 minutes. You write a markdown file that says, "When someone asks you to write a blog post, follow these steps. Use this tone. Structure it in this way. Always include these sections." Save it in the right folder. Done. Claude now knows how to write blog posts the way you want them to written. But skills can also be much more sophisticated.

3:17 You can include Python scripts that Claude can execute. You can have templates it fills in. You can build entire workflows. Now, here's the part that makes this really clever. Skills are what Anthropic calls progressively disclosed. That's a fancy term for something simple. Claude doesn't load every skill into memory all at once. When Claude starts working, it just sees a short description of each skill, like reading the title on a spine of a book. Only when Claude decides it actually needs a particular skill for the task at hand, does it pull it off the shelf and read the full instructions. This means you can give Claude access to hundreds or even thousands of skills and it won't get overwhelmed. It picks the right ones for the job automatically. Just like how an expert knows which reference manual to grab without having to reread all of them.

4:02 And this is what makes skills so different from just stuffing a bunch of instructions into a prompt. With a prompt, everything is clamped into context at once and it gets messy fast. With skills, it's organized, it's modular, and Claude pulls it only when it needs when it needs it. Now, here's the part that really matters for non-technical people. Before skills, if you wanted to customize AI behavior, you basically needed to be a developer. You had to build custom tools, write code, set up infrastructure. With skills, anyone can create one. You're a recruiter, write a skill that teaches Claude your company's hiring process, your evaluation criteria, your interview questions bank. You're in finance, write a skill that teaches Claude how to build reports the way your team does them. You're in legal, teach Claude how to review contracts the way your firm does it. And skills also work alongside something called NCP servers, which are Claude's connections to external tools and data. Think of it this way, NCP lets Claude reach out and interact with the world, pull data from APIs, connect to your tools. Skills tell Claude what to actually do with all of that. NCP is the hands, skills are the experience. Now, when you put these together, you get something really powerful, an agent that can connect to anything and knows exactly how to handle what it finds.

5:15 All right, so now that you have a little bit of idea of what skills are, let me now show you a practical project so you know exactly what I'm talking about. So, I'm inside my Claude code here and I'm going to run a GEO analysis on a real website. By the way, if you're interested in learning more about skills, check out the community link is going to be description. I have an entire skills library that you can download for different business use cases, for research analysis. And then also if you're interested in learning how to monetize this and sell this to real businesses and build real solutions for businesses, that's going to be also inside the community. But, anyways, so I'm inside my Visual Studio Code inside Cloud Code. What I'm going to do is come in the bottom here and I'm going to press {slash} g e o. And as you can see, all of these different {slash} commands, which are these skills, are available for me. I'm going to click on g e o audit. By the way, all of these details, this g e o audit and everything is inside this GitHub repo that I created.

6:08 Again, this is completely free for you guys, available. Link is going to be in the description again. So, basically, this g e o tool consists of 12 different skills. As you can see, these are the different skills on the folders. Each folder has a skills.md file. So, this g e o audit, as I can click on skill.md, you'll see exactly that all of these details are here and I'm going to talk about that a little bit. So, what I'm going to do is go ahead and run g e o audit on a popular website called calendly.com.

6:36 Again, I'm sure a lot of people are familiar with that. So, I'm just going to press enter. What this is going to do is now Cloud Code is going to look into the g e o skills file inside this particular folder, this skills.md, this g e o audit. That's because that's exactly what we ran. And these are the These are instructions that I was talking about at the beginning of the video, where now we're giving Cloud Code a specific instruction and making it become an expert in a specific area. In this case, it's basically g e o audit.

7:05 Again, g e o stands for generative engine optimization. It's basically optimizing businesses for search in ChatGPT, Gemini, and Cloud software, right? So, what this is going to do is it's going to run all of these analyses, as you can see. So, it's going through and running these these multiple steps. And all of this is again coming from the instructions that's inside this g e o audit skills.md file. So, the first phase is going to be discovery. It's going to discover the homepage of the calendly.com. And then afterwards, it's going to run this phase two, which is parallel sub-agent delegation. So, this skills has instructed Claude code to run five sub-agents simultaneously. Each agent is responsible for a different task, and that's what the beauty of skills are. You can you can have a simple instruction, or you can tell Claude to run multiple different agents doing different things, right? So, imagine if you're building a house. I gave this example before. If you're building a house and you have a main contractor, then that contractor knows make sure that the house is built in an efficient way. They hire five different sub-contractors, right? Each sub-contractor will be in charge of doing something in particular. An electrician, HVAC contractor, one's going to be building the foundation. So, that's the best way to basically build that house in the most efficient way.

8:27 That's exactly what's happening here. Behind the scene, Claude code is running all of these different sub-agents, and they're delegating these different tasks. As you can see right here, everything is going through. I'm just going to fast-forward this so you see the result here. And as you can see, the result is running this GEO report. So, if I open this now, this is the output of this particular tool. And as you can see, it ran this GEO analysis report with the executive summary, score breakdown, right? for different platforms like ChatGPT, Perplexity, Gemini, Bing, all of that stuff. By the way, I did a step-by-step on this. If you want to If you're interested in checking it out, I'm going to put the link in the description of that video.

9:05 But essentially, this runs this incredible service and runs these key findings for this particular company. It says, "Hey, Callanly has no Wikipedia article. This is a critical finding." And at the bottom, it gives it also a prioritized action plan what they can do on a weekly basis to basically improve their AI sellability from ChatGPTs and Geminis of the world. Again, I walked through this step-by-step. If you're interested in finding more details about this. But essentially, this is what this skills these skills files files are, right? This particular one, GEO audit, this is kind of like an entire comprehensive GEO Audit, but there's multiple other skills inside it.

9:43 So, for instance, if you just want to run a GEO site ability analysis for that particular website, you click on this or you run the GEO {slash} {dash} site ability instead of the GEO {dash} audit. And Claude Code is going to look at the skills.md file inside GEO {dash} site ability, and this is going to have a separate set of instructions. As you can see, this one doesn't have any sub agents. So, if you run this, this has a particular set of instruction that tells Claude how to operate for that task that the user is asking for. So, that's basically how you can run multiple different skills or or or give Claude Code multiple skills, and based on what you're asking, it knows exactly what to run, when to run it. And again, Claude Code is smart enough to know that even if you give it plain English, just tell them, "Hey, I want to have a report generated for a GEO audit, right?" You don't have to specifically run the slash commands. It will know exactly what you're asking for. It will look into the skills files and see, "Okay, which one is going to be relevant for this particular ask the user is asking me for." It's going to pull that off the shelf, those particular instruction, and it's going to follow through. And that's what makes this such a powerful tool, and that's why Anthropic is saying that the future of AI agents is actually building separate skills and giving Claude Code access to all the skills.

11:05 And based on what the user is asking, then Claude Code is capable of pulling the exact skill that it needs in order to run the analysis or in order to satisfy the need of what the user is asking. Again, this was just a quick walk-through. I have done a step-by-step on this. You can check it out if you if you're interested, but this hopefully give you an idea of what skills are, why this is the future. I'm going to be doing a lot more videos on this, so make sure you like and subscribe to the channel because I've got a lot more content coming up. I recently also did a marketing one. so check that out if you're interested as well. And again, if you're interested in launching your AI agency and learning how to monetize this, check out the community. Link is going to be in the description. Thanks for watching and I'll see you in the next one.