262 | Build an army of agents using Claude's Skills with no technical background with Isar Meitis

0:00Hello and welcome to the Leveraging AI podcast, the podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. This is Isar Meitis, your host, and I have a life-changing episode for you today. Really nothing short of that. We are going to talk about Claude skills, and we're going to talk about why they might be the biggest unlock for productivity and dramatic changes in your business and in other operations that you're doing in your personal life as well, and how easy it is to actually create new skills and use them in order to build swarms of agents to do more or less anything you can imagine.

0:40The reason I decided to record this episode is because I started using more and more of Claude Cowork. And I was using skills actually way before Claude Cowork showed up last week, but I did not understand how critical and important and powerful they are until I started combining it with Claude Cowork. So first of all, if you don't know what Claude Cowork is, go and check the previous episode where I talked about this. And I'm going to record another episode on some of the things that I'm doing right now to give you the biggest unlocks possible with the combination of these tools. But today, again, our focus is Claude Scales.

1:18And let's first of all, define what the hell is a scale? How does it work? and then we'll go and build more and more complex concepts from there. So a single skill is basically teaching Claude to do something very specific, such as your brand guidelines. So this was the very first skill that I've built for my business was the brand guidelines scales of Multiply, which is the name of my company. And what it enables me to do is now every time I'm anywhere in the Claude environment, I can ask you to create something based on my brand guidelines. and then the fonts are correct, the colors are correct, the style is correct, the tone is correct, the background, everything within the brand guideline just works magically.

1:57And all I have to do is ask for it in Claude and it knows how to do everything else. Other examples for single scales is how do you want all your SOPs structured in your company? So you can define that. And then whenever you create SOPs, they're all going to be structured in the same way and so on. So that does it actually work? Well, it works very similar to a custom GPT or a cloud project or a gem, depending on the universe you're in. So it has access to multiple pieces of data. These could be different kinds of files, references, links, et cetera. And it has instructions. The combination of these two things allow it to basically do whatever it needs to do.

2:36So it has the examples or reference materials combined with the instructions of what it needs to do. And now it knows how to do these things, such as to apply your brand guidelines or to create specific structure for any kind of document, whether SOPs, proposals, and so on. But the biggest unlock for me was the understanding that you can take any process in your organization, as long as it is a digital knowledge work kind of something versus a physical world kind of something. So as long as it is a knowledge work thing you're trying to unlock, you can create anything with combination of Claude skills.

3:14So let's break this down. A Claude skill can activate other Claude skills, meaning you can build an entire hierarchy of workers in the AI language, agents that can talk to one another and collaborate and coordinate and share information and make progress in parallel, and then aggregate all the information, review the information, and so on. Which means you can take really complex processes, break them down into steps, and sometimes even sub-steps, and each and every one of them can be its own skill that can then work together in parallel or in sequence with other skills in order to complete the overall task.

3:53Basically, the biggest promise of multi-agent orchestration, which is the holy grail of AI automation and a complete rethinking of how a business can operate, all can happen with Claude skills with zero third-party tools and with zero technical knowledge. Now, I'm sure you're asking yourself like, okay, how do I even get started? How do I identify what I need to do? How do I create these skills? What are they? How do they talk to one another? How do I figure this out? This is exactly what we're going to cover in detail in this episode. So let's start with where do these skills live and how do you update and create new ones.

4:28If you go to your cloud, either application in this particular case, or if you are on the web environment, it is the same exact thing. On the bottom left, you have your account. If you click on your account, one of the options is settings. And then inside of settings, you have multiple things. One of those things is capabilities. If you click on capabilities, then you have artifacts and then code execution and file creation and so on, memory. And then you have skills. If you look at mine, I have multiple scales, probably a couple of dozens, but you have an add button. If you click on add, it says create with Claude, write scale instructions, or upload scales.

5:04So let's go through all of them. If you click on create with Claude, it is basically going to start a Claude conversation that will allow me to create whatever scale I want just by describing what I want it to do. Write screen instructions basically allows you to write your own instructions just by typing them on your own, basically in the same way you're building custom GPTs. I don't recommend doing that. I will get to that in a minute. Or upload a scale. How do you create a scale file? Well, Claude will create it for you. So another way to get to this is one of the scales that exist inside of Claude that comes with the system.

5:36If you are watching the screen, you can see under scales, I've got your scales, which is all the ones that I created. And there are example scales. Example scales might be canvas design, might be internal comms, MCP builder, Slack, GIF creator, theme factory, etc. But one of those is a skill creator scale. And you can see that each and every one of those scales has three dots, basically ellipses menu on the right side. And if you click on that, you can select try in chat. If you select that, it will take you to inside of the Claude chat and we'll say, hey Claude, can you make something amazing with skill creator scales?

6:13This is just a placeholder, but then I can tell it to do whatever I want. But now let me explain something important about skills. How do you use skills is the simplest way you can think about it. You don't need to know anything. You literally just have to ask. So even if I didn't use any of this, and all I requested is I went to the regular Claude and say, I want to create a new skill that will do one, two, and three, it knows to automatically, dynamically, on its own, to activate the skill creation skill because you're asking to create a new skill. If you ask it to analyze an Excel file, it will use the Excel file analysis skill.

6:49If you ask it to create a Word document, it will use the Word document skill that it already has built in. So many of these come behind the scenes inside of Cloud, but then you can create new ones just by literally asking in a regular Cloud chat to create a specific skill for you. But the right way to do this is actually to do this in a much easier and yet more sophisticated, complex and organized way, which is to explain to Claude or to Claude Cowork or in Claude Code, it doesn't really matter, what is it that you're trying to achieve. So describe the broader task and ask it to come up with the different skills that will be required to achieve your goal.

7:28This episode is a really advanced episode. It is talking about building swarms of agents and multi-agent orchestration solutions that can automate your entire business. However, it will be very hard for you and your team to do this effectively without knowing the baseline, the basic building blocks, the do and don'ts, the core capabilities that AI provide across image generation, video generation, how to prompt properly, how to keep your data safe, and so on. To know that basic, we have created the AI Business Transformation course. I've been teaching this course since April of 2023. So we're now almost three years of teaching this course to multiple organizations and thousands of business people that have taken advantage and are transforming their careers and or their businesses with the knowledge they acquire in the course.

8:17So if you are looking for a way to really start 2026 strong when it comes to solid AI foundations, don't miss this opportunity. The course starts on February 2nd. It's a Monday and it's going to go for four weeks on Monday, noon Eastern time, two hours every single week. It will give you incredibly strong foundations for any AI transformation you want to perform, either for your career or for your entire business or department or team or whatever it is that you are managing. Since I am extremely busy with company specific courses and work, I usually teach these public courses about once a quarter, which means if you're going to miss this course that starts on February, you probably have to wait until around April or May in order to take this course again, which means you missed an entire quarter of getting AI benefits in whatever it is that you're trying to do.

9:09So do not miss this opportunity. Make this your New Year's resolution to start 2026 strong with solid AI foundations and come and join us next week and join us on February 2nd for the AI Business Transformation course. And because you are a listener of this podcast, you get$100 off with the promo code leveragingAI100. Don't miss this opportunity. And now back to the exciting and amazing world of Claude Skills. So let's look at an example. In this particular case, I am showing a financial analysis of a really complex set of files. These are a planning file, an actual file, and a P &L file for a really large international project.

9:52It has multiple regions and multiple roles and multiple types of things that are involved. And it has probably over 20 columns and a few thousands of rows. So going through that file to a human would have taken a few good days of understanding how it works, creating pivot tables and analyzing the information. And I want to be able to create a process that does it automatically. So what I wrote is I wrote, you are the executive VP of finance at a large corporation. I'm going to provide you with three files and then I'm explaining what the files are and I'm asking you to understand the structure of the files and the data that's in them.

10:27But then down here I wrote your first goal is to think about what skills you would need in order to do similar analysis in the future. Such skills may include analyzing specific aspects of data, doing online research on conditions and changes in specific regions and countries in the world, comparing different things such as seasonality, etc. The goal is for you to then develop multiple skills that, when combined together, will provide a thorough analysis of similar files in the future. Try to think of each skill as a highly capable financial analyst who specializes in finding very specific kind of information in similar data.

11:05Some skills should allow to coordinate between all the other skills in order to perform the handshake in the most effective way. As an example, if skill number one finds that there is a quarter over quarter increasing cost, it will hand over the research for that specific quarter to additional skills. One would research the type of labor that was performed and another would research the project or the WBS components that were involved. And so I'm basically explaining what are the tasks that I'm trying to perform, giving it an idea of what kind of analysis I want to do. And then it goes to work.

11:40and then it reads the three different files that I gave it. It does the analysis. It tells me that it has X number of rows and four regions and 11 resource countries and 58 roles, et cetera, et cetera, like all the stuff that's in this gigantic files. But then it also starts and comes with ideas for the skills it is going to create. So it is going to start with a financial analysis coordinator. The purpose of that is the master orchestrator that coordinates all other skills. Then it has variance analysis, trends analysis, regional cost analysis, labor cost analysis, WBS analysis, profitability analysis, and external research.

12:15This is just step one. Later on, I ask it to create skills that will allow it to create the outputs in specific formats. So what I want to show you next is I want to show you what happens when I actually run the analysis. So now that I have the skills already in place, all I have to do is write a very short prompt because the scales will do all the rest. So I'm writing the following prompt. You are an expert financial data analyst. I would like you to review the following files. And then I give it access to the three files and use your financial analysis scales to analyze information in the files provided and create the three reports.

12:52That's it. I don't explain what kind of analysis. I don't provide it any additional details. Very short. And now I click on let's go. And what you're going to see that's going to happen is it is going to start working on the process, but the first thing it is going to do, it will start using the scale. So on the right side where it says context, you can see the first thing it launched is financial analysis coordinator, which is the one that will invoke all the other scales. The second scale that it's using is XLSX scale, which will allow it to read the files that I gave it. It is asking me a bunch of questions, which I'm going to answer very quickly.

13:25It's less relevant to our current situation. And now once it's doing the analysis, you will see that on the right side, it will start invoking new skills and it will create a plan. So now it's already uploading PPTX and DocX because it will need to write and create those. It uploaded the Multiply brand and then it's now looking at the financial report generator as well. So it's adding more and more skills that it has created on its own previously in order to do the work. So I'm going to stop it right now because I already have created those and I want to show you how these look like. What is the final report looks like once it knows how to use all these skills?

14:03So it's going to run for about, I don't know, 15 minutes or so to do all the different steps of the analysis that it needs to do. And it is going to generate three different files because these are what the skills of the generation tells it to do. One of them is an Excel file that looks like this. So the Excel file has multiple tabs. The first one is an executive summary with high level information. The second one is variance analysis, then trend analysis, regional analysis, labor analysis, profitability analysis, each and every one of those tabs was created by a separate skill, or if you want a separate agent that knows how to research specific kind of data and generate a specific kind of report.

14:40They all report together to the orchestrator that invokes them, and then the orchestrator gives the work to the skills that generate these reports. So one of the skills generates this Excel report, the other one generates a Word document, which I will show you next. So you can see this is a highly detailed report. It has multiple pages and it has color coding that is consistent across the entire report. So you can see the color legend with green, red, cyan, orange, red, and yellow. Each and every one of them means something else. And you can see here an executive summary with the written analysis and a table and the color coding.

15:14It has the variance analysis with different things to pay attention to, whether minor, significant, or critical. It has the quarter over quarter analysis. It has graph charts, bars. Everything is extremely well organized and even recommendations in the end with color coding and using my brand guidelines as far as the colors and so on. And all of this because it has skills to create all of that. And the third file that it generates is a PowerPoint presentation. Again, fully aligned with my brand guidelines. Each and every one of the slides includes a combination of text and charts and graphs. Again, color coded with clear highlights for everything the analyst wants you to pay attention to in their specific areas of interest.

15:52such as rate versus volume decomposition and top variance contribution, whether up or down, by project, by region, and so on. It's an extremely well-organized PowerPoint presentation. It is way better than what I could have created on my own. And all of this was done by me asking for that simple prompt of just doing the analysis. And the reason it knows how to do all these things is because it has all these agents, each and every one of them focusing on something very specific and an orchestrator that knows how to make them all work together. So now let me explain to you how does it work from the moment I ask for these skills to the moment they actually exist in my Cloud universe.

16:31So it is actually very, very simple. The way these skills look like, they are files that are.skill in the type of the file. And the way Cloud works is when it creates these files, it either allows you to download them if you're in the regular Cloud chat environment or if you're using Cloud Cowork and you're giving it access to a folder, it usually creates a new subfolder in that folder. It calls it Scales or Scales Update if you've done multiple versions of them. And I will show you now exactly how that works. So I'm going back to my Cloud environment. And again, you can do the same exact thing, not on the Cloud desktop.

17:06You can do the same exact thing in the Cloud over the web. And you can see this is the folder that it's working in. And you can see that it has an updated Scales section. and you can see that it has files that are.md and file that are.scale. The ones that are.md you can completely ignore. The ones that are.scale are the ones you need. And so I just made an update to one of these scales, asking it to make some changes to it and it updated the file. And all I need to do in order to update the scale in my environment is again go back on the bottom left corner to my profile, click on settings, go to capabilities, and now I don't even have to scroll anywhere.

17:40All I have to do is drag the file that is the dot skill file into this environment. And you can see that Claude understands. It says drop to upload skill. I just dropped the file. It already identifies that it has this skill in its memory. If it will be a new one, it will just upload that. And it says replace financial analysis coordinator skill. There's an existing skill, blah, blah, blah. And I said, yes, upload and replace. And it's going to upload it and create it for me. And then I get a message on top that it's ready to go. That's it. Now, if you go back, you will see that it has all these other skills that I created that do multiple things in my business right now.

18:16Some of them work as a standalone skill. Some of them work in tandem with others to do really sophisticated and complex things. So why do I think skills are life-changing and so critical and so important? The first thing is they are a very similar concept to custom GPTs or projects, but with two huge differences. One, Claude is really good at building them for you. So you don't have to be an expert. You just got to give it the right information and it will help you build a skill. You don't need to know what to do. You just need to know what you're trying to achieve. The second thing is they do not take space in the context window unless they are called and invoked.

18:58So different than a custom GPT, all that data is in your conversation as soon as you upload the custom GPT versus here, it can have 300 skills in your company, but it will only call the one that it needs as it needs them. And hence, it's going to save space in the context window. And hence, it will allow you to be more flexible with what you're doing without running out of space. The second thing is they are extremely powerful because they know how to work in a multi-agent orchestration environment, and you don't need to know anything about what the hell that means and how it gets set up. All you have to explain is what you are trying to achieve and let Claude do the magic for you, and then just fix from there.

19:37As I just showed you, it doesn't always work perfectly the first time, but then you go back and say, hey, this is what happened. I would like to update that capability or that component. Which skills do we need to update? And sometimes it will get it perfectly and sometimes you need to help it, like it will forget the orchestrator as an example. But with very simple attention to details, you can have it fix whatever it needs to fix. You drag and drop it back into the cloud environment, and then it works. So that's the second reason. It is extremely powerful multi-agent orchestration building capabilities.

20:07The third reason I think this is extremely powerful is because you can literally take your existing company's SOP, every SOP you have, and turn it into a cloud scale. Meaning, you can take existing documents. First of all, as I mentioned, you can build a skill that will allow you to build better SOPs, but I'm putting that aside for a second, but you can take your existing SOPs and build skills from them. And then Claude will magically know how to do them. And then the last reason this is so powerful is the fact that Anthropic did something that makes it so simple, which is it works in every aspect of the Claude slash Anthropic universe, meaning skills work in the regular Cloud Chat, skills work in Cloud Cowork, skills work in Cloud Code, and you don't have to know how or why or when.

20:56All you have to explain is what you're trying to do, and the Cloud environment on its own will know how to call and activate the right skill, usually the orchestrator skill, which will then call the relevant other skills that it needs in order to complete the task. This is the ultimate democratization of agentic capabilities that ever happened. This is a huge step forward that enables literally anyone to create extremely powerful multi-agent solutions that can achieve more or anything that is knowledge work in your company. And it takes minutes to set up, a few more minutes to test, and then you're up and running with literally any aspect of operation in your business.

21:38So a quick summary. A, yes, I'm really excited about this and terrified at the same time because this means that a lot of jobs that exist today can disappear in the very, very near future if people learn how to use this properly, or you can keep these jobs and 10x your company, which makes it very exciting. I'm going to talk about a quick summary and then I will go to what this means. So before we end, I want to do a quick recap and then I want to tell you what it means for you and how I think you should use it. So the recap is this. Skills allow you to create either individual, very specific things that Claude can do or really advance complex entire processes with very big goals with multi-step processes that will work as a multi-agent orchestration kind of solution.

22:24You do this by simply asking Claude to help you build those skills and by explaining what you're trying to do, explaining what information and data you can give it access to, and then allowing it to define what skills are required, brainstorm it a little bit, provide it additional feedback, and then ask it to create the actual skills. Once it creates the skills files, all you have to do is drag them into the place that I showed you in Claude, and you have the skills working, test it out, ask it to make minor changes, and you're up and running. It is really that simple. The other thing that I will say, and then what does that mean for the world, and what does that mean for you, and what I suggest you do?

22:58What it means for the world is that multi-agent orchestrations that used to be saved to the biggest companies in the world by hiring multiple highly skilled individuals is now available to anyone. Now, yes, the$20 a month license may not be sufficient to build as much as you want. You may need to upgrade to the max plan that is$100. Or if you're going really crazy and doing multiple things in parallel, like I'm doing$200. But for$200 a month, you can build armies of agents that can replace anything in your business. And by the way, once you finish building them, you can downscale back again because they run.

23:34Because when they run, they take significantly less tokens than actually creating all these things in parallel. But now I'm getting to the point of what I think you should do. If you are in a leadership position, you need to teach your people how to do this. You need to build procedures in your company from a data security perspective, from an infrastructure perspective, from an API and data access perspective in order to make sure nobody does anything stupid. But once you do that, You should give every single employee access to the max plan in Claude and have them start building everything they need, every building block, every bottleneck, every problem they have.

24:13They can more or less solve with this by just explaining what they're trying to do and have it build the skills and the operation for them. Combined with Claude Cowork, you can do magical things in your organization in a very short amount of time. And my suggestion is hire somebody like me, doesn't have to be me, can be anybody else, to teach you and other people in your company how to do this effectively. And then define the rules of engagement, define the right safety guards from a data perspective and start allowing people, you know, forget about start allowing, demand from people to start building these kind of solutions.

24:50because it can 10x the amount of work people can do and do it faster, better, and cheaper than most. When I said in the beginning, this is a life-changing episode, I totally meant that. And my hope is that you will, A, experiment with this, and B, start implementing these kinds of solutions because if you move fast, you can grab significant market share in your industry, in your niche. And if you move slow, somebody else will figure out what I just showed you and will do it to you and you will stay behind doing everything manually the old way. Now on the bigger picture, I think what we will find is that we will stop doing things the old way and just automating them with Claude's skills.