299 | The magical world of AI skills - stop chatting with AI and let it work for you with Isar Meitis

Hello, 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 today we are going to dive into the magical world of skills. If you haven't built skills yet, this episode is going to change your world. It is going to take you from knowing chat as something you chat with to knowing something that can actually do work in your business more or less across any kind of knowledge work.

I'm going to give you examples of skills that I'm using. I have dozens, maybe hundreds at this point. I don't even know. I don't count them.

And I don't actually create them because Claude creates them for me or ChatGPT creates them for me. We're gonna learn how you can use these skills across different tools, including ChatGPT and Claude and PowerPoint, et cetera. And so you're going to learn how to completely up your game just by using skills which are now available more or less in every AI tool. Now, I'm going to be sharing my screen, but it's mostly to help me go through the steps in the correct order to make sense to you.

But if you do wanna follow on the screen and see what I'm doing, You can either watch this on Spotify or on our YouTube channel, and there are links to the YouTube channel in the show notes, so you can easily do that. But don't worry about it if you're driving or walking your dog or doing whatever it is that you're doing, while you're listening to podcasts. I am going to share with you exactly what's on the screen. So the first question is why.

Why do we even need skills? And the idea is very simple, that you can have them do things for you in your business versus just chat with you. But I wanna show you an actual example. So the Excel I'm going to show you has fifty-four columns and over thirteen hundred rows, and the team that needs to use this needs to generate six different reports every month based on this file.

They need to do a trend analysis, a variance analysis, a regional and labor analysis, a WBS analysis, and movers analysis. All of these are based on the raw data from their actual ERP system. That comes out to almost two full weeks of work, of analysis every single month to put this set of reports together that they need to present to their leadership and the board. That is a lot of work.

However, now you can build skills to do the same thing, which is exactly what I did. Now, they only generated a Excel report or reports every single month based on this information. What the skills are able to do, they're able to create a Excel file with all the required reports as they did before, plus a Word document that includes all the explanations on what's happening, including graphs and charts and tables and so on, and explain what's in each and every one of them. An executive summary that is in the format of a PowerPoint and a dashboard that is active that can go and use filters and so on.

All of that while only dragging the file into the process and asking Claude or ChatGPT or whichever to do the financial analysis. So let's look at an example. You can see in this particular case, I'm looking at Claude Cowork, and this is just a function, or if you want one of the options inside of Claude Desktop, which you can install on both Mac or PC. And what you can see is that I drag the file in here, and I wrote one simple prompt.

"Use your financial analysis skills and create the the relevant reports." That's it. There's no long details, there's no prompt engineering, there's no nothing, because the tool, in this case, Claude, knows how to find the relevant skills. We're gonna talk about this in a minute, and then it creates all the different outcomes.

So if we scroll down, you can see I didn't write any additional prompts, and you can see the four outputs. The one we're seeing right now is a Word document. It has an executive summary, it has tables, it has graphs and charts. It has detailed explanations of what's happening in each and every one of them.

I didn't write a single word in it, and it is branded to my brand. If we look at the Excel version, then you can see the Excel has multiple tabs. Each and every one of them has specific tables with specific kinds of analysis, which is the six types of reports that we talked about before. Each and every one of them is built properly, has graphs and charts, and has everything it needs to cover that particular type of analysis.

The third one, as I mentioned, is an executive summary in the shape of a PowerPoint. You can see it's a beautiful PowerPoint, or you need to trust me if you're just listening. It's all branded properly. It has big, large fonts, not too much text.

It is clearly following what is in the Word document, just as an executive summary with the charts and bullet-pointed explanations in each and every one of them. really well done, better than I could have done myself. And then the last one is that is available, and you can actually, click on it, and I will make it a little bigger. You can click on the different tabs, and it will move, and you can use filters and see all the charts and graphs and everything that needs to be there, which can be rendered every time you drag the file in.

So now because it takes you're not actually doing anything, to get this outcome, you don't have to wait till the end of the month to see the status. You can just drag the raw file into Claude or ChatGPT, ask it to create the dashboard, and it will do it for you. But how does this thing work? What do the skills actually do, and how do they actually work, and how do they become interoperable between all these different systems?

So the first thing is, in this particular event, there are multiple skills. there is a skill for each and every one of the reports, meaning for the trend analysis, for the WBS analysis, for all of those. each and every one of them knows how to look for specific information in the big file and how to aggregate it in a meaningful way. There are also four skills to generate the actual reports.

There is a skill that knows how to package the Excel report, the Word document, the PowerPoint, and the dashboard. Each and every one of them is a separate skill. So there are multiple skills here, but because each and every one of them knows how to do something very specific, they always get it right. If you try to build one skill to do all of this, you're going to get a result that is, eh, not exciting and in reality, for real life usage, not usable.

But because I've built, and when I say I, I mean Claude, built skills to do every one of those things specifically, then it knows how to follow the instructions in a very detailed way. So let's go a layer deeper and first of all talk about what the hell is the difference between a skill and an agent, because I know many of you are like, "Oh, that sounds a lot like an agent." and it does, and it's true. So let's explain for a minute what is a skill and how it's different from an agent, and then we're gonna dive a layer deeper and see how you're actually building skills and how these tools know how to pull the skills when they need them.

Because you saw in the example I gave you, I only told it to do the process. I didn't tell it which skills to actually use. So a skill is basically an SOP, a standard operating procedure, a set of instructions to give the AI that it will know how to do a very specific task. Now, it is written as a package of several different things, but the two main things are a .

md file. MD stands for a markdown, which basically means it's a Word document without the fancy design. So it doesn't have fancy headers or tables or all these kind of things. It's just plain text.

And more or less everything right now in the agentic world is running using these kind of .md files. So the .md files tells it exactly what are the instructions that it needs to follow to do the thing it needs to do.

as an example, I have a proposal writer. I actually have multiple skills that write the proposal, but one of them actually writes the proposal, and it is called the proposal writer, which is not too sophisticated. but it knows how to write the proposal based on my guidelines and so on. The other thing that is packaged in every skill or can be packaged is reference materials, which just provides the skill additional context.

In, let's say we use this example of the proposal writer, it will have a template proposal. It will have a PDF document with all the different services that I provide and description of why they're beneficial and exactly what's included in them, and so on. So it knows how to use it when it is writing the actual proposal. So you can attach information to the skills.

To make this simple for all of you that has built either projects or custom GPTs, Think about a custom GPT or a project that doesn't live just in that project or the custom GPT, but is available basically anywhere in your AI universe. It is basically the same thing Now, that sounds a lot like an agent, right? It's something that knows how to follow instructions. But an agent is a more complex version of this.

An agent, as a first step, might be able to do multiple skills instead of just one. more like an employee. An employee doesn't do just one thing, it does multiple things. does one thing, he or she does it in more steps than just one thing.

It's a more complex process. So a agent can have multiple skills, can have an orchestrator between the skills, can have the brain, if you want, between them to move the data around and coordinate exactly what needs to happen. It receives a goal and not just a specific process, and it will know how to reason through the different steps that are required to get to that goal. It decides which skills to use and when, and it keeps going basically until the job is done or until it needs help, or you pulled it to ask for your permission to do different things, and it will do that as well.

So that being said, many cases people would call a single skill an agent because it would still perform work for them Now the next question is how the hell does Claude knows which skills to pull? Because as I told you, right now in my Claude environment, I have either dozens or maybe hundreds of skills, and I never actually call them by name or very rarely call them by name. So how does Claude know? every single skill has a top section that if you want is an executive summary, a very short description about a paragraph long that explains what this skill does.

When Claude starts any conversation, and the same thing with ChatGPT, and the same thing with Hermes, and the same thing with whatever tool you're using that has skills, the tool loads the first paragraph, that descriptive section of all the skills it has available, which means it uses it as a restaurant menu. It now knows all the things it can do, and based on what you're asking, it is going to always look, does it have specific skills that support this thing that you asked?

So when I told it to do a financial analysis, it said, "Oh, do I have financial analysis skills? Oh, yes, I do. Let's see what they are and are they applicable for this particular case." And in that step, it will actually load the entire instruction section.

So the instructions could be multiple pages long, the entire SOP that it needs to follow. So the process is it reads the description, it loads the full skill, it follows the instructions, and it delivers the result. And then if there is an orchestrator, which we're gonna talk about it in a minute, it will then report back to the orchestrator with the output or in any steps in between. So this is how it works behind the scenes.

Now, the biggest difference, again, if you haven't used skills before, is that instead of just chatting with you or answering questions for you, it is actually doing work. I just showed you one that generates multiple reports when I write a single sentence, and that is the biggest unlock. Most people today are still using AI as a chat when in reality it could be a coworker working with you and doing different things, and the entry point for that is learning how to build skills.

By the way, talking about learning how to build skills and agents and orchestrations and so on, this is exactly what we're teaching in the multi-agent orchestration course that allows you to learn how to get started with things like I'm showing you on a very superficial level right now, but over eight to 10 hours teaches you exactly how to go from knowing nothing about this universe to building entire teams of employees in your company that will do more or less everything in your business as far as knowledge work.

I build new employees and new teams of employees for my businesses every single week, and in the course you can learn how to do this as well. But let's continue with talking about skills. So I want to give you a few examples of skills that I have in my system. I have the Multiply brand skill.

It knows what my brand looks like, how I speak, what do I usually say and don't say, what kind of style do I use, what colors, what fonts, how the logo looks like, all these kind of things. So every time it creates documents, descriptions, PowerPoints, et cetera, it knows how to use that and be consistent across the board. I have the same thing for Data Breeze, my other company which delivers automated invoice vouching and reconciliation at scale using AI agents, and so it has a Data Breeze brand guidelines.

There's also a deeper level of design if you want, and I have a Multiply presentation style. This knows exactly how I create presentations because I don't create my presentations. Claude does. How does it know exactly how to create them so they all look the same and follow the brand?

Because there's a Multiply presentation style which creates everything, including the design. It is also using the Multiply brand guidelines, but it has a lot of other additional specific things to PowerPoint There is a PowerPoint section importer. I create a lot of PowerPoints. I have a lot of existing capabilities or existing slides that I can copy and paste, and I don't do that manually.

Claude knows my entire universe. It knows every presentation that I ever created, and if I ask it to go and find specific sections, it will know how to import them in an effective way, which will maintain everything in those sections as far as the transitions and the motions and the style while adapting it to the new presentation that it is creating. So there's a specific skill for that. I have a time tracker skill that tracks everything that I do and gives me a report on what I am investing/spending/wasting my time on, on regular basis, and it tracks that across the board on everything that I do, and it knows how to create a report and where to place it, and so on.

I have a file cleanup skill. It runs every single night. I create a lot of documents in my process of getting to the outcome that either I create or in most cases, the AI creates, and there is version 23 of that file. Do I need versions one through 22?

Sometimes yes, sometimes no. But this file cleanup goes through my entire universe of AI and deletes files that it thinks are unnecessary. Ones that it's not 100% sure about, it puts in a quarantine folder, and then it deletes them after a month if nobody has touched them during that month. So it keeps my folders clean and organized, and I don't have to do it.

So that's a whole other kind of skill. I have a podcast content orchestrator skill, which gives you a concept that there's other skills in there which is true, but the orchestrator manages everything else for my podcast and all the content and helps put it together and organize it, so I can actually deliver it to you. I have a safe import skill. This is more of a data security thing.

If I go to any open source environment and want to import what's in it in order to use it, it will go through the code with a comb and go through everything in it to verify that what I'm about to import and use is actually safe to use, and it will raise any flags, and it knows how to run it every time I want to import anything, so I don't have to tell it to do it. There's a newsletter orchestrator that helps us create the newsletter every single week based on all the information from the Friday AI Hangouts, which is our community meeting, which any one of you is more than welcome to join.

It's on Fridays at 1:00 PM Eastern, and there's now over 30 people every single week that is sharing and discussing AI and how to implement it safely in businesses. Everybody's sharing ideas and solutions that they develop, and it's an awesome place. So that's one source of it. The podcast is another source.

information from other articles are another source, and so on. And the Newsletter Orchestrator knows how to combine all the other newsletter related skills to create it. And then there is my favorite skill that is called Gotcha. I name some of them with actual names.

And what Gotcha does. He or she is my devil's advocate. It is the most critical skill that I have. Every time I design something new, every time I hit a big milestone before I deploy it and so on, I call Gotcha, and Gotcha goes through it and says, "Ooh, you missed this.

You should have thought of that. What about this thing? This seems wrong to me. Let me verify it," and so on.

And then it just helps me not miss gaps, blind spots, mistakes, things that I've done in the process because either I didn't think about it, I didn't have enough time for it, I wasn't aware of different things and so on. And it just helps me deliver better results across the board on everything that I'm doing, and I use it more or less every single day across more or less everything that I do. As I mentioned, I have dozens more. These are just examples to give you ideas of what skills can do for you.

Now, where do skills live? So inside of ChatGPT, you need to have a quote unquote "business account," so either Teams or Enterprise in order to have skills, which is pretty annoying based on the fact you have skills more or less everything else, everywhere else, including in the basic Claude function. So Claude is definitely a better starting point right now if you don't want or if you don't have access to a business account. But if you have a business account, then on the bottom menu where your name is on the bottom left, if you click on that, it pops up the area where it has the settings and personalization and so on, and that will have skills as well.

If you click on that, it will show you all the skills that you have, and you will be able to see that it puts them into different categories. So the ones that are installed, the ones that are created by you, the ones that are available from ChatGPT or OpenAI themselves that you can then install, and the ones that you can get from the organization you belong to. So there could be like a supermarket of skills that other people created that you can use. Inside of Claude, you will find it in the Customize menu.

So on the top left section of your desktop app, there is where it has new chat and projects and schedule and so on. There's a button called Customize. If you click on Customize, it takes you to a page that shows you your plugins and your skills and your connectors. And if you click on Skills, then it will show you a list of all your skills.

Again, you can see here that I have a very, very long list. If you click on any of the skills, it will show you the three components in it. One is called a skill.md, one is your packaged, additional information called references, and one is the different scripts that it knows how to run, and these are all packaged together.

And if you click on that, it will show you the actual content. If you're looking at the one we're looking at right now, you see exactly the structure. This is the PPTX, section importer, and you can see there's a short description on top that tells it what it does. This is what it loads to the memory every single chat, so it knows from every single skill what it can do.

And then underneath that, you can see the actual instructions, and you can see a very short section of it, because you can see on the right side, there's a very small scroll bar, which means there's a lot to scroll. So the instructions of even just this thing that imports PowerPoint sections from one to the other is a pretty detailed set of instructions. And the way they're created is I literally just ask Claude to create them for me. I'm not going to dive into exactly how to do this.

This is way more time than I can invest in just one podcast or as a small section of a podcast. But again, we cover that in the multi-agent orchestration course that you can come and take, and we'll teach you how to do that. But in general, I don't write this. Claude/OpenAI ChatGPT or any other tool that knows how to work with also knows how to create them.

But this is not where it ends. So now that you know that these skills exist and that they can do a lot of cool stuff and that they work in ChatGPT and that they can work in Claude or Hermes or Base44 or anywhere else that uses skills, they actually also work in other tools. So in the Office universe, the Microsoft Office universe, there are also skills running inside of PowerPoint and Excel. So you can actually import skills into Excel and PowerPoint and use them in the tools themselves that you use regularly to do your thing to get to similar outcomes.

So let me show you how this looks like. So if I'm in Microsoft Excel, and you can see here there's a very, very large file, and this is the same file that I showed you in the original example. You can see it has multiple columns and a very, very, very long list of rows. Again, over 1,300 rows.

What I gave it is the raw data that you can see in this tab, but then it created on its own all the other tabs. It created the executive summary, it created the trend analysis, again, with graphs and charts and everything that it needs, and so on and so forth. You can see all of them in here. How does it know how to create it?

Because I gave it the skills that knows how to do the analysis. Where do you add skills into Excel? Well, you need to first of all install either Claude for Excel or ChatGPT for Excel. You can see I have both.

These are extensions that your organization should enable you to install, but once you do, and if you have an account of ChatGPT and/or Claude, then you can install those. If you install them, then you can attach skills to them. So in this case, I have the ChatGPT extension open, and you can see up here on the top right corner, there's the ellipses, if you want these three little dots. And if you click on that, you can see that there's a section called Skills.

And if I go to Skills, it will show me the skills that I have. It also has the ability to create new skills right in here. So they've done something a little confusing with ChatGPT for Excel and the same thing for PowerPoint. You can import skills that you created in external sources.

Either you created them on ChatGPT or you created them, somewhere else, you can bring them in here. But there's no button to import. In order to do this, you need to create a new skill button, which opens the skill creation. And you can see it has a name, it has a description, and it has the instructions, and it has the ability to attach files as the reference materials.

But here on the bottom, there's an option to upload a skill file. So you don't actually need to fill out any of this information. If you have a skill file that you already created, you can click the Upload file, select the file, and then it will install it here on ChatGPT for Excel, and then it will be available to you in the Excel environment. Now, why haven't they done it in a way that every skill you have in the regular ChatGPT is also available in ChatGPT for Excel?

I don't know, but this is how it is right now, and it's not a big deal to import them over again. It's like two clicks, and you have the skill here. And then you can do cool things like this, where you would bring in the skill, you would bring in your raw data in the Excel world you work in, and then you will give it one line and say, "Please create the financial analysis," and it will create all these tabs for you in your Excel while adding all the graphs and charts and everything that you need, which then you can continue working on right in here because this is your regular work environment if you're working in Excel.

The same exact concept exists in PowerPoint. So here you see an example of my slides, and you can see that I opened ChatGPT for PowerPoint in this particular case. Again, an extension, and it has the same exact thing. The ellipses, the three dots on the top right, and there's a skill button that you can click and import skills to it Which means whatever you create as a skill can be completely independent of the platform you're going to use it in, which I find really, really cool and exciting, meaning you can create a skill in one environment and then import it to any environment you work in, either a different AI tool like ChatGPT versus Claude or Hermes or Base44 or wherever it is that skills are available, Claude Code, et cetera.

But you can also import them into more and more daily tools that you use, in this particular case, Excel and PowerPoint. And I think over time, they will be available in more or less everywhere. And the interoperability makes it so magical because you can move your skills around with you, which means you are not tied to and dependent on a specific tool. If tomorrow there's a new AI platform that you really, really want to use, you can take all the skills with you, and they're most likely going to work out of the box because they all work the same way.

The tool will know how to read the description in the beginning, will know how to pull the right skills, and if your SOP works in one place, it will most likely work in another. But a single skill is awesome and can do really cool stuff. But as I mentioned, the real magic happens once you start connecting them together. You can build orchestrator skills that work and manage multiple other skills.

So that's exactly what it is, an orchestrator skill manage other skills, and then you practically built an agent. You can call it whatever you wanna call it. It doesn't matter. It allows you a lot more flexibility, and this is exactly what I created in the very first example that I showed you, and this is how the vast majority of my skills work.

It's not just one skill that does something. It's multiple skills that together do something. The orchestrator can decide which skill to call, what information to give it, what to do with the output, and so on, in order to make the process run in the most effective and safe way, and it knows how to pass the data around. It knows how to stop at specific checkpoints that you define to it to get your permission or your approval or your feedback on different things that the skills are developing.

It is basically the project manager of the other skills. So let's do a quick recap of what are skills and why are they maybe the most exciting thing in the AI space right now, especially at an entry level. They are a set of instructions. They are an SOP that the AI knows how to follow.

You can have a very large number of them on your computer, and when you tell the chat something you want to do, it will know how to do it, it will know how to find the skill, and it will know how to follow the instructions. You don't actually need to create them. Claude/ChatGPT or whatever other tool you're using can create it for you, And you can carry them around with you to other tools to benefit from them, not just where you created them and not just in AI tools as I showed you.

Now because there are extensions to Excel and extensions to PowerPoint, and again, I'm sure sh-soon extensions to more or less anything else, you can bring your skills into these other platforms as well and still benefit from the amazing things that they can do. And it is amazing. They can do literally anything you can imagine if you can write the SOP properly. If you can explain to the AI exactly what's the process, it will be able to follow it and do it every single time.

That is it for today. Again, quick reminder, if you wanna learn this in detail and really know how to use this and applies it in your business while connecting them together, obviously, I cannot cover all of that in a 30 to 45-minute podcast, but in eight and a half to ten and a half hours, I definitely can, and I've done this multiple times. The current session, the current cohort that we're selling is for August. We sold out all the ones from May through July, so the next session you can join is in August.

Don't miss out. It will sell out as well. It is selling very quickly already. So if you want to join us, if you want to transform your business or transform your career and understand how to build actual teams of agents that can do more or less any knowledge work in your business, don't miss out on this opportunity.

It is truly, truly life-changing and business-changing, and it will put you on a completely different trajectory than you're on right now. If you are a business leader and you wanna do this for your team, for your business, please reach out to me. There's a link in the show notes where you can reach out to me or just find me on LinkedIn, Isar Matis. There's only one Isar Matis on LinkedIn, which is pretty awesome.

And so if you found it, you found me, and you can send me a message there, and I will gladly chat with you and explain to you how do I do private sessions to teach these kind of things. But that's it for today. I hope you found this, valuable, hopefully extremely valuable. This is really one of the biggest unlocks in the AI space today, and I will see you again this weekend.

But this weekend, instead of a news episode, we are going to drop episode 300 of this podcast, which is absolutely crazy that we are at that number right now. But We are. So episode 300 is going to drop this weekend instead of the regular news episode, and I apologize that in the last three weeks it's gonna be the second time there's no news episode. But I think 300 is a big enough number to celebrate, and it is a unique and very interesting episode that I truly think you would learn a lot from, because we're going to share with you how the most successful AI company I know right now is doing what they are doing, and the journey they went through, and how they set up everything and so on, and the mindset of the CEO.

So don't miss it out, first of all because it's episode 300, but also because it has immense amount of value built into it. That's it for now.