Creating an AI Assistant: How to Get Started

Happy 2026. Hey, how's this for a New Year's resolution? How about you Master AI marketing? The AI Business Society gives you live training with experts who are getting real results. No hype, no fluff, just proven workflows that work. And you've got access to an incredible library of training from many of the incredible guests that you've heard on this show. So secure your AI transformation by visiting socialmediaexaminer.com AI welcome to the AI Explored podcast, helping you put AI to work. And now, here's your host, Michael Stelzner. Hello, hello, hello. Thank you so much for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzer, and this is the podcast for marketers, creators and business owners who want to know how to use AI. Today we've got a really, really good show. I'm going to be joined by Igor Pagani, and we're going to talk about how to create an AI assistant. We're going to talk about all the foundation steps you need in order to use artificial intelligence to assist you to do the tasks that you don't like doing or the tasks that you wish you could do, but you don't know how to do. And I think you're going to find today's episode absolutely fascinating. Also, I am at Stelzner on Facebook and at Mike Stelzner on X Twitter. If you want to let your friends know about this show and you want to tag me, that's the place to do it. Also, if you're new to this podcast, click that follow button on your favorite podcast app so you don't miss any of our future content because we've got some really good stuff coming your way. Let's transition over to this week's interview with Igor Pagani, helping you simplify your AI journey. Here is this week's expert guide. Today, I am very excited to be joined by Igor Pagani. If you don't know who Igor is, you got to know Igor. He is a tech enthusiast and the founder of the AI Advantage, an online school designed to help people excel with AI. His YouTube channel covers generative AI use cases and and his membership is called the AI Advantage Community. Igor, welcome to the show. How you doing today?

I'm doing so well. So glad to be here. Michael, thanks for the invitation. Looking forward to talking about what has been happening, but also what is coming and how people can adapt and make the best out of this new tech.

Yeah. And I'm really excited to dive in Deep with you today, Igor and I are going to explore how to create AI assistance. But before we actually get there, I'd love to hear your story. How'd you get into AI? Go ahead and start wherever you want to start.

So it depends. There's really many timescales I could take here. But I'll just start out by saying that I've always been a early adopter by nature. If I look back into my early childhood, I was the kid that got a new phone. I think I got my first one at age 11. And first thing I did was go into the settings and explore every single submenu. The camera had 12 sub settings. I had to try every single megapixel count, take a picture, compare the results. I was that kid. The other kid were like, oh, there's Snake, we can play nice. No, I was still in the menu, sitting at home by myself. So. And that ethos just carried forward throughout my life. So when I grew up, I at a certain point started my freelance video career. And there I just got into all the technical sides, all the, all the microphones, the audio, the editing, the different editing software. Then I discovered Adobe After Effects. It just. I'm a very curious person and I love to explore technology. So now fast forward 15 years from me being 11 or 12, and I got to a point in my career where I was creating videos for clients and with clients basically all budget sizes and all crew sizes, all the way from a one man show to like medium sized advertisement productions. And I just felt like I want to do more. There's nothing left to explore. There's no new frontiers in video production. Most of the things that I learned, I applied. So I started exploring more. And then it occurred to me that I've always didn't just love YouTube, but I've always been creating for YouTube on the side. I never considered it a job or anything like that, but ever Since I was 17, I was just spinning up different YouTube channels for different passion projects of mine. First, Magic the Gathering tutorial channel on Xbox, teaching people how to build decks. Then later on just filming my friends how they're having a good time and filming vacations. So at this point in my career, and this is somewhere, this is three years ago now, I decided to really commit to YouTube and to start, no matter what happens, to start creating an average of two videos a week where I'm in front of the camera for the first time and where I'm presenting. So I did this. And obviously, as many of you might know, YouTube is a long journey and nothing happens overnight. So for the first one and a half years of this journey, basically nothing happened. I just kept creating, kept pouring money into it. But now I get back to the first point, which is this, like, curious tech enthusiasts mindset. I just kept learning the new stuff and I kept trying new stuff on YouTube. And it turns out that with the release of ChatGPT, I already had my eyes on generative AI for images. But when ChatGPT came out in the circles that I navigated in one of my communities, it came up on the day of release. So, like the little nerd I am, I just like locked myself up in my apartment and just tried various things. It was so interesting, it was fascinating. Nobody knew about this, nobody used this. Some people know about GPT3, but nobody was really using this. And all of a sudden with ChatGPT, like, it just felt like a lot changed. So now four days passed. I released my first video on YouTube because that's what I was doing. I committed myself to two videos a week, no matter what. Right. So I uploaded the first video on ChatGPT and just went to bed. And then I woke up and I was blown away. I was like, oh, this is 70 views. 70. People watch this overnight like they care. People want to see more about it. So I created another video. Right. Well, turns out that second video is, I think it's approaching 4 million views now on YouTube. Wow, that was the second upload on the channel. And mind you, this is back in December 22, early December 22. So when all the hype in January, February came around and all the people figured out that this is actually going to be, you know, game changing and world changing, the next, like, paradigm in software, the next platform shift. Yeah, I already had a library of videos and I already had a month of experience of like obsessing over ChatGPT. And it turns out if you spend 50 hours trying one thing and somebody else is just getting started, you could be really helpful to that person. And that's what I built entire AI advantage on. It's just I do my best to be extremely early on the new releases. I spend unhealthy amount of time exploring them and then I turn around and use my communication video production skills to teach people. And that's basically the story behind how it all got started. And now we're doing so much, but that is at the core of it, just staying curious and teaching people what me and my team find.

When did you know that you ought to go all in on creating content Specifically around AI. Was it after that second video came out or was there just give us a little bit of that journey?

I would argue that I kind of went all in on just creating content at that three year mark, you know, three years ago from now, I just made that decision that, hey, I'm doing everything I'm doing. I'm, I'm working with all the clients. And it was interesting. Like that career was like on a good trajectory. Like I was getting more and more interesting jobs and clients. But I just knew it wasn't enough for me. I wanted more responsibility, I suppose more like I, I always had a very kind of productive and like ambitious mindset in terms of what I wanted to do. So I already set my mind to like doing a lot back then. So you asked about when I decided to go, yeah.

When did you decide that AI was going to be the main thing that you're going to focus on?

So like, as I had to set up in my mind of like, yeah, I'm just doing this YouTube thing no matter what. It was really easy to lean into the AI side of things. As soon, as soon as the first few videos showed the numbers, at that point I was uploading. Well, at that point, kind of like it was a mix of two channels. There was another channel, I was doing my jobs. So it was really easy to cut some of the other chords when this started, like really progressing. So I would say it took about two weeks until I started refocusing. But again it was, it was not like a hard decision. It was, I knew I was doing YouTube anyway. Now it was just a question of like, which one of these topics do I really lean into?

So as of now, where is your YouTube channel as far as like size? And then tell us a little bit about your community.

So the YouTube has been, I actually have to check. I'm at 233,000 subscribers. But as you might know, it's, it's a bit of a vanity metric. What really matters is the views and mainly the watch time and how many people return. And I've been fortunate enough to attract a certain crowd of people that, that are very proactive and they're very, they're practitioners, they're people who want to use this in their everyday lives. Whether, whether it's marketers or entrepreneurs or others that just have this mindset of like, I want to do stuff rather than just like being informed. I attract a lot of doers. And that has been kind of the ethos of the channel. And that's also my focus topic. So I don't just do AI, I do practical AI tutorials or updates that you can implement today. That's like my whole thing. And as of the community, it's just a very natural progression of that idea because there's only so much you can teach on YouTube. You can't really upload a five hour lecture or five one hour lectures that link to each other on how to like build personalized GPTs and assistants. Doesn't really work. The algorithm is going to punish you. The viewers are not going to watch the whole thing most of the time. Then your channel is going to get suppressed. It's just, it's just terrible. It's just many things do work on YouTube, but not everything obviously. So the demand for that was there and that's also what I wanted to lean into. I wanted to teach people more. I wanted to give them all my experience with automation that I actually built over years. I had experience with no code automations and basic development and coding before the whole, you know, YouTube channel here. And if you want to share that, it's just not a good format to have a YouTube channel where YouTube is expecting 10 to 20 minute videos of you. So that's where the community kind of like spawned. It's more one on one time, it's longer format, it's a smaller circle and it's also, also letting the people vote. You can't really like leave it up to the comment section what YouTube video you're gonna make next in the community. You can do that, you can give him a survey and look at the results and just form the curriculum around that. So, so that's kind of the setup right now.

Very cool. And for folks that have been listening, Igor is in a community of YouTube creators that's a very small community of really great people like Matt Wolf who's been on the show, who are really just crushing it and keeping up with the ever changing, fast paced world. I used to say that I thought that social media was the fastest changing industry in the world. But now I know that AI has eclipsed it by a long shot. So my next question, Igor, is a lot of people have, that are listening to this show, may have been dabbling with AI, maybe are even skeptical they might be here to learn. Some of them maybe don't understand why AI is going to help us improve our work. I would love to your take on why AI is actually going to help make us better.

I'm confronted with that question regularly and I think, you know, there's, there's Many answers, but one that I keep returning to is just, it's just super, super zoomed out and super macro and it's, it's basically just asking people if you're trying to be productive, if you're trying to create something, if you're trying to move the needle on whatever metric you pick. Do you think that being more intelligent about your decisions in the process is benefit or is that going to hold you back?

Obviously it's a benefit.

Obviously it's kind of a no brainer, right? Like having a little bit of, you know, more reasoning and planning and just making wiser decisions that have been informed by past experiences. It's probably a good idea. Pretty much every human agrees on that. So then if you introduce digital intelligence with all its flaws and all the, I don't want to say negative, but dystopian views that surround it too, sure, there's that side too. But I think we can all agree that there is some level of intelligence already accessible today through an API, through a chat box. If you introduce that into your work, you're going to be more productive or less productive. Also an easy answer. So I think that's where it really begins. Just accepting that sure, there's many narratives around it and there's many narratives people are spinning up for their own benefit and many of them have weight and truth to it. And there's the whole ethics discussion. But at the end of the day, if you can work more intelligently, you probably should.

I fully agree and I think the more that I am hearing guests like you on the show, the more I'm wrapping my mind around the possibilities. We're going to talk about AI assistance and having an AI be like your assistant. And I would love you to. First, before we get into like the how part of it, I'd love you to define in your view, what is an AI assistant.

I always like to pull real world examples because that's just how we're built, that's how we're used to operating. So I would just say it's very similar to having a real world assistant, independent of the fact if you ever had one or not, whether a personal assistant, executive assistant, in whatever form that might come part time, full time, if you have somebody helping you in what you do, gonna output more and you're gonna do better work. It's gonna give you the ability to focus on the things that you're good at as opposed to some of the parts of the job that you might want to move to the side or some very repetitive Parts. So that's where I would start. I would say, you know, AI assistant. Right now it's just a simplified version of a human assistant. It can't do as much. It's probably not as smart, but only in certain ways. In other ways, it has basic domain knowledge about every domain on planet Earth. Not every assistant has that. Now are you going to get assistants that might be really well read and well educated in a specific industry and therefore outshine the AI in pretty much like every discipline within that industry? Sure, absolutely, yes. But also consider that we're very early days here. I mean, we're talking, you know, we're in maybe year two of consumers even touching these products. And I mean, just look at the Internet in year two. I would say that. So think of it as a human assistant, but simpler, but also smarter in certain ways. So it's kind of a give and take, but you only learn about the advantages and disadvantages of it by using it. There's, there's only so much we can talk about, but we will definitely talk about both sides of it today.

So again, unprompted, not prepared question, what are some of the things that an AI assistant can do that might supplement the work that we are doing today? And maybe you have your own and you can say some of the things you're using it for.

Yeah. So I'll start by saying, I think today it doesn't really make sense to have some universal assistant, something that would like replace your personal assistant or something like that. The thing that works today is having these super specialized laser focused use cases that you identify and then building something around those. So like if you have a SOP in your company, that could be, you could have an AI assistant for the sop or maybe just a part, maybe just one step of the sop. Practical example would be posting content to social media, right. I think to this audience that might be a very, very typical task. So if you're posting content to social media, there's a pipeline to it, right? It's not just like, oh, I'm gonna post even at the simplest form, even on a tweet, let's say, right? You still have to kind of come up with an idea for the tweet that might be sparked by some other inspiration. But you have to come up with the idea for a tweet, then you need to type it out, then you need to edit it. And then there's obviously also the part where you interact with the responses to that, right? That's the simplest form. Maybe on the Other end of the spectrum of social media, it's like really long form YouTube videos where that whole pipeline doesn't take minutes but dozens of hours to complete. Now if you dissect that pipeline into all the different components, you can think of the assistant as a helper on one of the components. There's not like a YouTube video assistant that gets it all done for you. No, you think about the fact that first you need to do some research probably. Then from the research you gather some inspiration or some ideas or you might get them otherwise. But let's say then you have the idea stage, then you have fleshing out the idea into kind of a bullet point script. At least this is how I work. Then you turn the bullet point script into more detailed research. Then you get into the recording phase where you turn the bullet point script and the research into a video. Then you have to edit the video, then you have to review the edit somewhere along the line. Ideally in the beginning you come up with the title and the thumbnail for it. Then you post it, Then you need to split test it. You need a description, you need subtitles and you need to distribute it of course afterwards and follow up and interact. And that is like one YouTube video.

And analyze it of course. Right.

And look into the analytics. That is again that. And that might be the research for the next video. So every one of those that I just named you could isolate and we could, we could talk about how it can help in that particular step. But you're not going to have a universal assistant now that just creates these YouTube videos for you. It's really helpful in the research, it's really helpful in fleshing out the, the details of the bullet point script and adding like other other things. It's really helpful in ideating titles. It's really helpful in creating mockups of thumbnails. Heck, if you look at my channel lately we've been playing around with like a fine tuned version of stable diffusion. Doesn't matter what that is in detail, it just matters. It can create any image of me in any situation. So right now I think we're split testing thumbnail of me holding up a sign looking like a homeless person saying GPT4O took my job and it's actually doing well. It has 1% higher CTR than the other stuff. But again, this is just like this image generation model is my assistant. And in finding the titles, I love using GPT4O and I have various brainstorming techniques where it just comes up with stuff that I would need a Second person to jam with, to come up with that level of creativity. But there I have it. So that's how it really works. You have to isolate these things and handle them one by one, and then just start with prompts and then flesh them out into something you could call an assistant over time.

I was driving one of my daughters somewhere today, and I was driving back and I called one of my employees because I had an epiphany. I said to myself, okay, every time we do research for guests for the social media marketing podcast, my actual assistant goes through this massive undertaking. And I asked another one of my employees if she could help me figure out a way to get AI to do a lot of the preliminary research.

And.

And we started brainstorming. Here's what the output looks like that normally my assistant does. I gave it to my employee and I said, let's see if we can build a very simple model, for lack of better words, that can do some of these kinds of things so that my assistant doesn't have to spend as much time doing it. The reasons I'm explaining these scenarios to people so people can, in their mind, understand that there could be any one particular task that's time consuming that you or someone else on your team is doing that could be essentially built to. To be supplemented or assisted by AI. So this leads to. My next question is, like, what do we need to think about when we start? Like, where do we start when it comes to actually using AI in an assistant kind of role?

I think that might be the best question you could have even asked. Because honestly, most people where this, like, spark of curiosity hits them and they're like, oh, AI, okay, let me look into this. Let me see, like, what this is about. Very quickly, they just get overwhelmed by this avalanche of information all across the Internet, there's so much, and there's new releases coming in every day. As you referred to earlier, you thought social media moves fast, but then the AI space is just. It's really a wild phase, and it's all driven by innovation. So overwhelm is a real problem. So where do they start? Where did you begin? Well, it's simple. You might have heard this word a billion times at this point, but it's for a good reason. You start with one prompt. You start with one prompt that does something useful to you. And, you know, all across the Internet, people are showing you prompts and, hey, the secret prompt and this super prompt and this giga prompt and this prompt that I crafted is going to change your life and make you $10,000 a month, whatever it might be at the end of the day, there is a point there. The prompt is the primitive. It's the building block. It's the simplest thing. When you break down improving a process with AI, improving your process with AI, it always starts with the prompt. Or at least I think that's a very good way to look at it for consumers trying to implement this into their everyday lives and their career workflows. So how does that look in practice? I mean, let's take the simplest of prompts, a summarizer prompt, right? You can just say inside of ChatGPT, summarize this column and then you paste it something, you give it a blog article or whatever, it's going to summarize it for you. Now, is the summary useful to you in many cases? Probably yes. There's a plethora of other problems, but we won't get into that now, into all the use cases. I really want to answer your question because it's such an important one. So if you start with one prompt and you find that summarizing something might be useful to you, then you have to summarize this prompt. But then when you look at alternatives and maybe you look at other prompts rounded you, you might realize you can extend that. It doesn't need to end that. Summarize this, you can summarize this to me in the style of an 11 year old and all of a sudden you get this more simplistic summary. And then maybe you can add, summarize this to me in the style of an 11 year old in five bullet points, make it extremely concise. You add that too, and all of a sudden you have five short bullet points as if somebody was talking to 11 year old and you get a summary in that style. And now you might move on over to other types of prompts like writing emails or doing other specific things, helping you brainstorm. The list goes on. But all of these things connect because the styling something as if you were talking to 11 year old and making it extremely concise. All of these things mix and match. You could use that for writing emails, email drafts, you could use that for coming up with ideas for any type of brainstorming. These things mix and match. But where it really begins is finding one prompt that works for you, that sparks your kind of imagination and where you're going to be like, wow, yeah, this is good. And then you need some place to save that prompt. So you might start off with your notes. But I definitely recommend having some sort of like Prompt organization thing, whether it's Excel sheet, whether it's a dedicated page in your notes, somewhere where you can store the ones that actually hit a chord with you, and that's really where you begin. Now, we can keep talking about where to take it from that, but it all starts with having one useful prompt to you and saving it somewhere. And then over time, you can, you know, automate things. You can flesh the prompt out, you can make it interact with other, other parts of the, of the chain. But that's where I would always begin. And that's why also the one video that did so well on the YouTube with the 4 million views is it literally starts with a prompt saying generate. I think it says generate me an essay about penguins or something. Or it's the first question. I'm not exactly sure. I would need to double check. It's either that or it's what is the meaning of life? Just that's the first second of the video and that's how you should start too. It's just hitting one prompt, checking out the result, if you like it, saving it, and then developing your skills over time and improving that.

Let's talk about filling the gap between AI potential and AI results. I'm sure you've tried ChatGPT. Are you seeing real ROI? Most marketers are barely scratching the surface of what's possible with AI, and it's because they're missing the training and the frameworks that will allow them to be very successful in their careers and in their future. AI business world allows you to get that training. When you attend, you get two full days and 20 expert practitioners, many of the people that you've heard on this very show, that will teach you exactly what you need to know to build the workflows and to succeed as an AI enhanced marketer. It takes place April 29th and 30th in Anaheim, right alongside Social Media Marketing World. And here's the good news. You can save $300 when you register by January 16th. Head to AI BusinessWorld Live right now. This discount ends soon. Let's talk about some of the more useful prompting stuff beyond that, because I know, for example, we talked about scoring, asking it to score results. Talk to me a little bit about that.

The second question, or like one of the first questions that comes up is usually like, okay, but what makes a good prompt? Like, how do I, how do I know this is, this is good stuff? Like, I'm new to this, like, how am I gonna evaluate it? So what I usually say there is just it's very simple. It's your taste matters here. You gotta be the judge of whether this is useful or not. Okay? And we talked about the scoring, which is like a particular technique where you could ask it for evaluation that is more rooted and reason and more rooted in mathematics rather than just kind of like feelings or emotions. You could ask it to evaluate. And there's certain prompts that help you evaluate the quality of a prompt or help you flesh it out and extend it. And you could use ChatGPT to evaluate all sorts of concepts. Like one example that somebody in our community did recently was he gave it the different songs that Drake made to this Kendrick Lamar and the verses. And then he gave it the songs from Kendrick, and He just asked ChatGPT like, hey, come up with different dimensions for scoring. And then go ahead and score the different verses and give me your verdict on it. And then ChatGPT went ahead like, well, this one really cut deep because of this reference. So this is like a 9 out of 10, as opposed to this one from Drake, which was a little softer and summarized it. And in the end, it gave him a whole, like, evaluation of, like, who's better? So you could do that for prompts too. But I would always recommend your taste is what matters here. And your domain expertise does really matter, because this is just an assistant. At the end of the day, you got to know what you're doing, and therefore, it might do a lot of the work for you, but the burden of deciding if that work is any good is still on you. And that kind of, like, value judgment of, like, ah, this is. This is not good enough, or this is great, or this is, like, decent. It's good enough for a draft that really matters.

Okay, so just to be clear, are we asking the AI to evaluate itself, or are we evaluating the AI saying, you know what? I'd give you a 7 out of 10 here. I think you need to give it another shot. Like, help me understand what your thoughts are on that.

The second one, really, like, you are evaluating the AI at all times. No matter what it creates, no matter.

What you're doing, we're giving it the actual feedback, right? Like, we think you could do. I think you could do better. Right?

And then that is really the next step. Also voicing that with people, sometimes we are careful and we don't voice every single thing. If you're, like, fully blunt and voice every single thought of yours, that's usually a recipe for disaster right away, right? But with the AI, you got to adopt this radical candor where you just tell it. If you don't like something even a bit, you forgot a dot, do it again.

Okay.

All right. Even to the minute details. And don't be shy to do that. It's not going to, you know, the people joke about it, but it's not going to, like come back and like punish you because you didn't say please.

What about asking it to ask you what it needs? Like, I know we talked a little about this when we were in prep. Talk to me a little bit about how we do that.

Yeah, so. So this is really where we get into the intermediate stuff. Because in the beginning it's just about finding things that work and taking those first steps and realizing, okay, like, this might not be exactly what I need, but it's definitely better than not having this help of this assistant. And then once you go further, you start realizing that you get to dial in the quality. So to say you get to dial in the quality of the responses you get by. And this is really the biggest word in prompt engineering, which, by the way, is just using ChatGPT. It's just talking to it. It's just communication. It has like barely anything to do with engineering. I think that word has done more damage than good. But the point is this. Once you get into that, once you get into, okay, how do I up the quality of this? How do I get more relevant stuff where you find the base prompts and you're like, okay, how do I go? The next. It becomes all about context. What context are you giving it? And this is where I'll just tie it back to this original thing that I said. And I also teach it like this. It's if you think of it as a real world assistant, and matter of fact, I would go a step further. If you think of it as an assistant that you paid good money to assist you, you have a certain set of expectations there, like, hopefully he or she is going to be helpful and productive to what you do. If not, why are you paying? Right? And if you bring that mindset to the chat interface and you think about the fact that context is really what defines the quality of the outputs, well, then you get to a point where you can start thinking about what context should I give my assistant so he or she can perform a good job? If you hire a new assistant, let's say it's a she, and she just comes into your office and you tell it, okay, I want you to come up with a social media marketing campaign for these two new product launches of ours and Then you let her go, right? And, like, do it well, what is she going to think? She's going to be like, well, I mean, I know the name of the company, but he didn't even mention what product launches those are. And he didn't even show me, like, other materials that they created before. Heck, I didn't even get to say hi and ask him about his name or, like, what kind of, like, what's going on? And then like, as the good assistant she is, she's going to do her best. She's going to create, like, product launch strategies for you. But she didn't have a look at your socials. She doesn't know your history, she doesn't know your writing style. She doesn't know your target audience. She doesn't even know the names of the products. So giving her all that would be good. And this is where I'll loop back to what you said. Because you said, like, how about asking it what it needs? If you go ahead and like, include. But this only goes so far. But if you include something like ask me clarify questions, that will help you get me better results. If you say something along the lines of that, it will go ahead and it will ask questions like, okay, I'm ready to create your product launch, but what product are we launching? What is your target audience? What kind of tone do you want this to happen in? What is your timeline? All of these questions, and then you get to go in and answer those one by one. And all of a sudden you might not even realize, but you gave it the context that it needs to do a good job. So that's why a lot of the prompt engineering at the intermediate level is just figuring out what kind of context can I give it so it can perform better. And a simple question like that can get you started.

And I love the fact that you're talking about, talk to it as if you hired a very expensive person to assist you in your job. Because if you did hire a very expensive person, that person by their very nature would ask you these questions. But AI by its very nature will not ask you questions, at least today. Would you agree with that? So we have to be the ones to give it permission to ask us questions. I found that to be true. Right. So it's not going to prompt you because for whatever reason, it always tries to get it right on the first prompt, doesn't it?

Well, I thought about that. It's a good point. It would be a super annoying user experience for new users if it did that. Right. If you're like, hey, what's the meaning of life? And the answer would be, well, it depends what you mean by meaning. What religion do you subscribe to and when are you born? And like, what ethnicity? Like, what do you mean by meaning? What do you even do in your everyday? You just want like some answer and you're like, whoa, yeah, if it asks clarifying questions, it would make for a terrible experience. It would have never went viral like this.

Yes, except. But in the use of business, it's not going to know to ask those clarifying questions unless you give it permission to ask those clarifying questions. Right.

That's why I'm here.

That's right. So, okay, so so far up to this point, we've been talking about the preliminary stuff. We started with different use cases of how you could have AI assist you in various different very narrow use cases. And then we talked about the importance of experimenting in the beginning with very simple prompts, like a summarizer program prompt, and then expanding that prompt to say, in the style of blank, like an 11 year old. Right. Or in bullet points. And then we kind of expanded into like the fundamentals of a good prompt and maybe some of the more intermediate things, like giving it feedback like, hey, I'm only going to give you a 5 out of 10 on this because you were missing these things. Right. Just like you would with a real person and then eventually asking it to ask you clarifying questions because it won't do that by its very nature unless you ask it to. So now we're at the point where I think this is the fun part of the conversation, like the interconnections between AI tools and all the other cool stuff that you can do. Like, I would love you to expand on this because a lot of these AI tools, but especially ChatGPT, has a lot of integrations and this is where all of a sudden you can have an assistant in a truer sense of the imagination. So share with us a little bit about the tools that integrate with these things and kind of a little bit of the use cases and how you can start getting this thing to not just live in a chat window inside of ChatG, but to do things for you outside of that domain.

Fantastic. Okay, so let's get into it. So let's assume that, you know, you followed all the steps so far and you got to a point where you're happy with the result. You're like, this is a summary I would love for us to have after every single meeting. Now how do you make it happen now one answer to that is you pay a high subscription fee to a tool that has been purposely built to do that. And that makes it super simple. You know, meeting summarizers are actually a perfect example because like dozens of them pop up and then like Zoom integrated it natively and Google Meets is integrating it and now they are like disappearing again. But you can do all of this by yourself too. And I think this example of the meeting summarizer, maybe let's just run with this because you could absolutely create this value yourself. So what do you need to create a meeting summarizer? Well, first of all you need a transcript of the meeting, right? You need something that you can like prompt on top of. Because just feeding it one hour of video is probably not gonna, it's not gonna cut it. There's, there's models that like are getting these capabilities, but that's not what you wanna do. You wanna transcript of the video. Secondly, you want a prompt that works with the transcript. Something as simple as summarize this in five bullet points by the way, in the style of an 11 year old is one option. But you can also say in the style of Vogue, in the style of Tim Ferriss, right? You could replicate brands and people's kind of voices and everything. So anyway, that's just a little side note, but basically you need a transcript and you need a prompt. Now how do I get a transcript in ChatGPT? Like all I can do is run a prompt. And that's fair enough because that's chatgpt. But this is where you need to take a step back and realize that there's actually something behind ChatGPT that makes it work. And it's the model, it's GPT4 and you can access it through either ChatGPT, which is like the easy like Typi interface, or you use the API and the API is simply explained. It's the same thing, but you can connect it to other things. So I like to describe it as like a power plug, like a universal power plug that plugs into all different kinds of walls. And then like what brings it together? The one answer is the simplest answer for most people is a no code automation tool, something like Zapier or Make. There's many others but like I think those are the two popular ones. And there you can take these plugs and kind of like plug them into each other and make them talk to each other. So okay, let's do automation. Crash course in 30 seconds. Are you ready?

Yeah, let's do it. And by the Way, what I love about this is what I'm hearing you say while you're drinking your water, is that, hey, chatgpt by its very nature is extremely powerful and it has an API. And the API allows third parties to connect into ChatGPT and allows ChatGPT. And it allows a communication funnel, if you will, back and forth between tool X and ChatGPT. And we're going to get into APIs, and specifically with Zapier. But before we go there, isn't it also true that ChatGPT also has plugins that do some of this stuff without the API?

It is true. Although they did remove them a few weeks ago. They introduced them in a big fashion and they. They like scrapped it again.

So right now they're not there. They might be back by the time this publishes.

Yeah, absolutely.

This is a way to do it without having any kind of worry, right? Yeah.

But here's a point. All the plugins, they still exist. They just morphed. Now they're actions. And I want to get to that right after this, like automation crash course, because that is really like the next step. And you can do a lot of that. Well, it is kind of tricky. Like right now, if we're talking right now, no code automation tools, that's where it's at. And all, no code automation comes down to two building blocks that you got to understand. First of all, a trigger. Secondly, an action. You can think of it as like a row of dominoes. And it takes for the first domino to fall in order for the consecutive ones to also fall. Right. So the trigger is the first one, like what is going to cause the first domino to fall over and then everything after that are actions. And you can kind of build that infinitely long chain of different actions.

By the way, Google Tag Manager for my fellow marketers is exactly the same thing, right? You got triggers and you got actions, but keep going.

Yeah, it applies to many things. So in our meeting summary example, you need some sort of trigger. So the most basic one that we use and that I like to use a lot, is just if a new file appears in this folder, in this specific folder on my drive somewhere, that's my trigger. So, for example, if we finish a YouTube video, I instructed my editors to always export MP3 file and to drop it into this folder. That's all they have to do. And then what happens if they drop the MP3 file into there? The first domino falls and then the question is like, what did I put in my actions? Well, in this case, very Simply put, there's three more. One action transcribes it. Now, there's also another OpenAI API, and this takes a little bit of education to be aware of this. But there's an API called Whisper, super cheap, and basically it does transcription and translation. So you can just take an MP3 and transcribe it in, like, in a matter of seconds and turn the MP3 file into a text document, all right? Then the second action is just a ChatGPT call. It's like using ChatGPT, you just talk to the ChatGPT API, the GPT4 API, and all you do is you take the transcript that was created in step number two. Remember, step number one, a file appears. That's a trigger. Step number two, the transcript gets created with Whisper. Step number three, you take the transcript and you say, summarize this in five bullet points in the style of vogue, right? And then it does that. And then the last step is just save this to a text file inside of the same folder. That's it. So every time one of my Editors drops an MP3 file into that folder, within a few seconds a new text file appears of a summary of that MP3 file. And that is like a simple example of automation. And now you can, like, interchange the applications. You can have your CRM talk to your Gmail talk with ChatGPT in the middle. You can do all these combinations, and there's a million examples that vary depending on what you use and what you're trying to achieve. But at the very basis, it's this. And then I guess the next step we should talk about is like, how is this going to, like, evolve and is this going to become simpler? Because for this, yeah, you need to be aware of Whisper.

Well, even before we go to the next step, Zapier versus Make, what's your quick skinny on which one you prefer when you're doing AI stuff?

I actually use both because I always used to use Zapier and I'm just like, used to it. But I have become a big fan of make lately, I have to say, because they're super quick with it. I believe Zapier or Zapier or whatever, however you pronounce it. They took two months to integrate kind of the newest API and make. When GPT4 came out, it took them less than 24 hours to have it available. Plus it has a fantastic free plan. So for learning, it's just way better for teaching the interface. Very interchangeable up here is slightly more expensive, I think. I mean, the pricing makes it look like day and night. But the truth is make takes more steps. So even though you get way more steps with at a lower price, it takes more. I don't know, if somebody were to ask, put a gun to my head and it's like zapier or make, I would say make.

Well, what's really fascinating about this too is both of these tools integrate with a lot of other solutions, right? So very quickly, your example of dropping an MP3 file into a folder, could you also create a workflow quote or have your AI assistant or whatever craft an email and say, hey, look, just so you know, it's ready for you to upload to the AI hosting site person X or whatever? I mean, can it, can it trigger that kind of stuff as well?

Oh, absolutely. We actually held extra workshop on this in a community, I think, two weeks ago, where you're basically using ChatGPT to create a draft for every single email that shows up in your inbox. So like whatever email shows up, ChatGPT pre writes a simple draft for you based on the prompt that you link it into. And then there's like the advanced version of that, and we're planning on that already is kind of having a router in front of it. Like, hey, when Michael or one of his domains send me an email, then use this prompt. But when I get something from my bank, I want this prompt with a different template. And then you can start kind of like customizing it. But absolutely. Something like this is a simple three, four step, you know, automation.

I mean, this is kind of crazy stuff now, right? Because when you start to think about simply a folder inside of your Google Drive containing a certain kind of file and triggering a transcript, and then having that transcript trigger an article, and then eventually having that article trigger something else, I mean, you could have a literally in a matter of minutes, I would imagine a whole bunch of stuff that might have taken a human forever. Is that right?

You bet. Let me give you one example and full disclaimer. I built this with great passion. I was like, this is going to change my life and then I'm going to change many other people's lives. But it turns out we don't even use it on every single video, but it's still running. So I have this automation that basically takes an MP3 file, creates a transcript, and then I run 16 different prompts on it and save the different files to different folders. And basically it's like, give me 10 titles with corresponding thumbnail ideas and mix and match. And then I included like some creative techniques so it's a little out there and it gives me stuff that I wouldn't expect. That's actually the part that I use a lot. But then we were like, okay, now repurpose it into a blog post. Repurpose it into a tweet, promotional email. I want to see a draft for a short, how my editors could re edit it with this specific formatting. This is what we use. These are the assets we use. Like, I really like fleshed out some of those prompts, gave it a quite extensive Persona with templates attached to it, how I expect, how I expect it to behave. And basically from one MP3 file, I create like 13, I want to say 12, 13 pieces of different content. And then, you know, if we, if we're getting into promoting something or distributing something, we have all these drafts that we can kind of use as a, as a starting point.

How long did it take the assistant or whatever to do all this? I mean, are we talking minutes?

I think this entire chain? Well, I would have to think now it depends on which API you use. So now the new 400 API, it's, it's blazingly fast. So I think in total, I mean, what is that, like 16 an hour maybe?

Max?

No, no, no, no, no, no. I think it's, it's. It might be like 16 zaps out of those. The Dropbox ones happen instantly, where it's just file management, text file creation happens instantly in the ChatGPT calls. They might take like a few seconds. So I would say like this whole thing in a minute, you have the folder populated with all the stuff.

I mean, this is pretty mind blowing. All right, let's look into the future a year from now. I mean, like today we're recording this in May of 2024. Where do you see all this going with this AI assist?

Well, the next frontier really are this agent thing that everybody's talking about. The next frontier for gen AI for consumers. Okay, we need to narrow it down like that because there's like many directions that are happening. There's like hardware, there's foundational models, there's infrastructure. But like what we care about is the applications layer and the applications layer that we have access to and we can reasonably use. It's really this idea of, I think I'm just going to like paraphrase Sam Altman here on a recent podcast where he really outlined this idea of agents. But he tackled it from a bit of a different perspective than usually the conversation would go. Usually the conversation would go like, okay, right now you Go in, you prompt, you get some result and you use it. Maybe you can automate that a little bit, but, like, moving forward, it's going to be more like, you state a goal and then the AI figures out the roadmap and figures out what it needs on the roadmap, what tools, what prompts, what does it need to do to arrive at the goal, and then it follows the roadmap by itself and all it does is like, ping you. Like, okay, I'm done. Here's the result.

That's crazy.

That's the idea of an agent, right? Like, and that idea is like, exactly. Everybody that hears that is like, I want that.

That's scary.

Yeah.

All right. Actions. We were. We said we're going to come back to actions, and I didn't want to forget that. What is it about actions that we need to know?

Okay. Okay. So, like, I need to round out this point and I'll do that with actions, because if that is the roadmap, if that is the idea, then what Sam Altman said, it's going to be less like that. Less like an agent that just does things for you, but more like a senior employee that does things and also has a certain amount of autonomy in doing them. So, like, you won't have to ask every step along the way. It'll just do it. Like, if it wants to send a mail and if it wants to, like, use your credit card, it will just do it because your senior employee might be able to do that. So actions are actually a part of this vision and this transition from, like, me prompting it and doing things to, like, the agent doing it. Actions are the tooling, they're the ability to send an email, they're the ability to use your credit cards to order on Amazon, they're the ability to scrape a website, go out on the web, run some Python code. That's what these, like, actions can be thought of. Now, right now, it is extremely cumbersome to set up for most folks. You have to host a API endpoint. That's where you lose most people because it's like, how do. Well, luckily you can use things like make or like other no code tools to, like, get it done. But fair enough. You host an API endpoint, then you have to turn the API endpoint documentation or like the instructions on how to connect to it into an OpenAI schema that is like a specific format. There's like a separate GPT, like a separate AI that just transforms the API connection docs into this schema. Then you give the schema to your GPT and then it can connect. And when you ask it to do the thing, it does the thing. So it takes a few steps, but this is not where it's heading. Where it's heading has actually been outlined by Microsoft a few days ago here so beautifully. I mean I've been talking about it since mumps, but where it's heading is like log in with Google, connect to Dropbox. It's going to be as simple as using some of these no code tools is where it's like, yep, connect to Dropbox. I log in with my email and password and you're connected. And then you can like do things with Dropbox. Right now it's a bit, it's a bit crazy with building like with building GPTs, but Microsoft introduced this idea of the copilot which is essentially the same thing as a GPT which by the way Google names named gems, it's all the same thing. It's this idea of like a specialized version of ChatGPT with a specific set of tools that does one thing for you. And actions are the levers it can pull. They're the connection to Gmail, they're the connection to Dropbox, they're the ability to scrape the Internet, they're the tools that this agent or senior employee has at its disposal. And with that it can actually get things done. And last point here is that these GPTs that Microsoft introduced, why I'm saying they're a beautiful visualization of the roadmap that we'll be following here with these products. The reason for that is that they went a step further. They're like, hey, give it access to my entire Google Drive folder or like my entire Dropbox folder with my company documents or like drop down consumer friendly stuff, right? And then connect it to my Outlook account, connect it to my this account and then you just tell it to answer all your emails and it'll try to figure it out. Now look, that's going to be a bit of a disaster because these models are not as developed to do this. It sounds really good in the marketing presentation. But even if you build things with more advanced techniques, if you use open source LLMs like open source AIs and give it external files via something called rag, it's a mess if it's unlabeled. You can't just throw 10 Excel sheets at it and expect it to retain all the data. And for it to work with it just doesn't work that well yet. So I think that's a roadmap it's this idea of actions and this idea of specialized prompts doing specialized tasks all kind of packaged into like this idea of a senior employee or agent. But in order for you to do that, you kind of need everything we talked about here and more. You need like basic, no code automation, understanding. So you know like how to and like what you can give it. You need like basic prompt engineering skills so you can direct it when you're not happy with the output. But that's, that's the future going to have these things that are helping us and we're going to be more productive.

This has been absolutely fascinating. I just want to, on behalf of all of the people that are listening and are watching this thing, say thank you so much, Igor, for sharing your more insights. If people want to connect with you on the socials, where do you want to send them? And if they want to check out you and your business, where would you like to send them?

Well, first of all, thank you for you know, having me and for asking these questions. You might be able to tell, but I love talking about this stuff. This is like, it's my favorite topic and like there's so many directions that you can take it in. And that's what I do on the YouTube. Basically I, I keep things fresh. We upload multiple times a week. I think that is kind of the main hub, the YouTube.

Yeah. Tell them how to get there.

Yeah, on YouTube, simply put, the AI advantage. You can also look for how to use ChatGPT in two minutes and like the guy with the finger up, that's me. But yeah, YouTube AI advantage. And beyond that, I really simplified it last week. I tried a lot with different books and courses and everything trying to like transmit this knowledge to people, but I really dumbed it down. Now it's like YouTube is the hub for up to date information and techniques and tutorials. Then we have the weekly newsletter that is completely free where you get a prompt template with 500 prompts plus a personal prompt organizer, plus a few of these pre built assistants that you can just copy paste for free with the first mail. And then we send out one mail a week that's in the middle. And then at the bottom the community includes all the more higher, touch, deeper learning experience stuff. We do lectures, we have learning paths where we teach you how to build GPTs in a combination of certain lectures that we already held. We have like one piece of content a day, one new workflow guide or lecture every single week. And that is really where people can kind of Hop on the train and like not get left behind because it's, it's really, really easy. But the bulk of, bulk of the stuff and like, if you, if you only want to spend a few minutes a week, then the YouTube and the newsletter are fantastic free ways to stay up to date.

How do people get to the newsletter? Where do they go to get that newsletter?

My aeadvantage.com and then there's a newsletter button on top. You can also go to myaidvantage.com/newsletter.

Perfect. Igor Pagani, thank you so much for coming on the show and sharing your insights with us. You were absolutely spectacular.

Oh, it's been a pleasure, Michael. So good to talk to you. And yeah, I hope everybody can take away something practical, something, something convenient, whether it's an idea or it's something concrete. I shared a few of these techniques. If you check out the YouTube, it's going to be all techniques. I show a lot of like concrete use cases. That's kind of my thing. But all it takes is finding one thing and then just take that and invest those 10 minutes. Spend that hour on a Saturday just playing around. Heck, the best model is free. Now just go to chatgpt.com, kind of type in some of the words I told you about, some of the prompts, something you get in the template and off you go. It's a really exciting time and a lot of people follow this narrative. I just want to end this that a lot of people kind of fall into this trap of this mainstream narrative of it being really dangerous. I think there's a point there if you talk a 10 year timescale. But I'm a big fan of living in the here and now. I'm a big fan of making the most of what we have. Who knows what is going to happen in life a week or a year from now. I tend to not concern myself with that too much. I care about what can we do today for ourselves and our loved ones. And this technology is so helpful today. So I hope some people got value out of this and can improve their own lives or their loved ones lives with some of these techniques. Thank you, Michael.

Hey, if you missed the anything, we took all the notes for you over@social mediaexaminer.com A8, the letter A the number 8. And also be sure to follow this show on your favorite podcast app. And would you do me a favor? If you enjoyed this show, would you share it with a friend? You can tag me. I'm Stelzner on Facebook and ikestelsner on X Twitter. And do check out our other shows, the Social Media Marketing Podcast hosted by me, the Marketing Agency show, and the Social Media Marketing Talk Show. This brings us to the end of yet another episode of the AI Explored Podcast. I'm your host, Michael Stelzner. We'll be back with you next week. I hope you make the best out of your day and may AI help you become more successful. The AI Explored podcast is a production of Social Media Examiner, AI Business World, early bird pricing ends on January 16th. You can save 300 on your tickets. Register at AIBusinessWorld live today.