Building AI Agents: 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 put AI to work. Today, I'm going to be joined by Sandy Carter, and we're going to explore how to get started with AI agents. If you don't really understand what an AI agent is, that's totally okay. You're going to get it all figured out. Today we're going to introduce a bunch of cool tools and a bunch of things to think about when you're building your very own agent. Also, if you're new to the show, follow us on whatever app you're listening to because we've got some incredible guests coming your way. Let's transition over to this week's interview with Sandy Carter.

Helping you simplify your AI journey. Here is this week's expert guide.

Today, I'm very excited to be joined by Sandy Carter. If you don't know who Sandy is, she is a futurist and the chief operating officer at Unstoppable Domains. Her newest book is AI First Human. Always Embracing a New Mindset for the Era of Super Intelligence. Sandy, welcome for the first time to this show, and welcome back to one of my shows. It's great to see you. How you doing today?

It's great to see you, too. I'm so excited to be back. You do the best podcast. So I'm so excited and so honored to be here. Thank you.

Well, I'm very excited to have you. Today, Sandy and I are going to explore building AI agents. So let's, before we get into that, share a little bit of your backstory. How did you get into AI? Start wherever you want to start.

Yeah. Yeah. Well, it's kind of a fun story because I actually started at AI in college. Wow. Which has been a little while. One of my professors, he was talking to Me about the future is all AI. Wow. Wasn't he a genius? And he taught the very first AI class at Duke University. And I took that class. I love that class. I actually did a second class with him on AI as well. Which way back then was very rudimentary, but because I had AI background, I then went on to work at IBM and then voila, we started talking about building Watson and doing stuff with Watson. Well, they were scanning the company. Does anybody have any AI background? And voila, I did. So I raised my hand. Of course, I went back and took a bunch of other classes too, at MIT just to make sure I refreshed everything and started working with Watson on Watson, primarily building out the ecosystem and the business model on the business side of artificial intelligence.

Why don't you explain what that was? And just so people understand that.

Yeah. So if you remember, we had a computer that played Jeopardy. Jeopardy is like a. A game show. We had trained Watson. That's what we called him was Watson. We had trained Watson on lots of different answers, disconnected him from the Internet, and then had him compete against world champion Jeopardy. And Watson ended up winning, which was super cool. I also got to work on an AI cookbook where we did AI generated recipes. I was at south by Southwest when we had a food truck come, put together all kinds of really cool creative dishes. For example, chocolate bacon tacos, which was a huge hit. Yeah, we had lines all the way down the road, but we really went after AI from a business perspective. So healthcare, looking at how it could diagnose screening of lung cancer. Cancer, for example, and really invested in that space. I then left IBM and I went to start my own AI startup where I used artificial intelligence to match a company's culture with the right innovation tactics. So considered it kind of like a Myers Brig for a company. Right. Determining its personality, its culture. And the reason I did this is I had. I was living in Silicon Valley and I saw all these companies and all these countries sending people to Silicon Valley to replicate the innovative spirit. But they would go back 60 people came in from Germany, they went back home, and then they called me and they're like, sandy, nothing works. Well, it was because the culture wasn't set up to accept all of these innovative ideas. And so I sold that company and I went to work for Amazon, started working there. We were on the cloud. So, you know, AI had really progressed now. It needed a lot of compute power for all the learning models and everything that you had to do. So I was working on the enterprise workload space. And so we started looking at types of businesses that would use or could use or leverage artificial intelligence. I still remember when at AWS, we came up with a great way to do training, reinforcement training, and we thought way outside the box to do a deep racer car. And this car had IoT sensors on it and it raced on a racetrack. We had all of these cool competitions, racing, like engineers versus managers or people in the food business versus technologists trying to teach reinforcement learning. So I did that at aws, and then I came over to another startup, so you can see startup enterprise, startup enterprise, startup enterprise. And now we're using AI a lot in the blockchain space. And so I've just always loved this whole area of artificial intelligence since college and really have been embedded in it. And then recently, as you said, wrote a book about AI first. It's my second book about artificial intelligence.

Outstanding. Well, and Sandy, you've been on my Web3 podcast, because the company that you work for currently is very active in the web3world. So it's been fascinating to see the journey that you've been and all the opportunities that you've had an opportunity to be part of. So, coming to AI agents, let's start by explaining what the upside is for marketers or entrepreneurs or businesses when it comes to AI agents. Just kind of share your thoughts on where you see that might benefit people.

So first of all, I think let's pause for a definition of an AI agent, which is essentially an algorithm that can make a decision. Okay, Gather information and make a decision. You know, what we see today from Gen AI is more you ask a question and you get input back. But an AI agent can actually say, oh, I think you're going to do A or B, it's going to automate a repetitive task, or it's going to answer questions for a restaurant or for a spa. It's actually going to make some decisions on your brand's behalf. And this is why it's so important for marketeers, because it impacts a brand's experience. And in fact, for my book, I really wanted to deep dive on agents. This is how actually Michael and I started talking about this was I wrote an AI agent, the first book with an AI agent that accompanies it. It taught me a lot about AI first, going through all these different revs that I did to be able to really understand the potential of what an agent can be for a brand, for an experience, whether experiences with a book or with your product or with a service that you have to offer. As you move forward, so give us.

A little bit of what the upside is when they're done. Right? Like, what are the benefits waiting for those that embrace agents.

I think there's so many. Recently I went to California and I don't know, did you have Pizza My Heart down in San Diego?

I don't think so, but there are so many pizza places, it's possible. I don't remember.

So in Silicon Valley, if you go out of the airport in San Jose, or really if you drive around, you'll see a lot of Pizza My Heart Pizza restaurants. So I recently went back down to speak at a Gen AI conference, and I landed late and I was like, I'm really hungry. I want to get some pizza. And so I started looking up, you know, where's the nearest Pizza My Heart? And I saw this article that said, Pizza My Heart now has an AI agent. Now, how cool is that? So I'm like, of course I have to try out this AI agent. I went and I got the AI agent for Pizza My Heart. First of all, it has the style of Jimmy the Surfer. Now, if you know Pizza My Heart, their whole advertising campaign is, Jimmy the Surfer wants a pizza, and he'll only go to Pizza My Heart. So I thought that was a really cool linkage from the brand to the AI agent was keeping the continuity of Jimmy the Surfer. Now Jimmy the Surfer talks like a surfer in the AI Agent, which I think is also really cool. And I was able to use this agent to ask, can you deliver to my hotel? I'm thinking about this pizza. What do you think? It recommended a different pizza for me, which now is my favorite pizza. It's sausage, garlic, and hot honey. I know it sounds horrible, but it's actually really tasty. And it actually took my payment through. The agent thanked me, and then the agent came back later and said, did you like that pizza? So I tell you that story because I think the upside is you can have an amazing experience. You can really enhance your brand by using AI agents as well as making your customer experience even more engaging than before, but unstoppable. For example, we use an AI agent now for our customer support team. That works for me. So we have an AI agent that's answering questions. She now answers about 30% of the questions that are out there. And our customer sat went up 10 points even though we leveraged and used an agent. And so you can see there's lots of upsides here. You can not only free up your best customer support reps to do something else or have a cool, funky, you know, Jimmy the Surfer brand experience. But you can also elevate the engagement as well.

I love it. Okay, so for those that want to start with AI agents, what do we need to be thinking about? Where do we begin? I guess is maybe the question.

I'll tell you the way I started, and then I would recommend maybe a different way. So what I did was I started exploring these different agents that I like. So I looked at Jimmy the Surfer. There's a place here that has a spa. It was my birthday, so I went to the spa and I noticed they had an agent. Of course, you know, I'm a geek, so I had to go through the agent. So I asked them who they used and how they built their agent. I have a friend who built a fashion agent, so I asked her. And so I gathered all these tools down and that's how I started playing with some different tools. Some of those were pretty hard for me to use and to leverage. I used to code in college, but I haven't coded in a long time. So I really needed something or wanted something a lot simpler than heavy coding background. So if I had to start over advising someone, I would tell them to go and look at an agentic course. Look at like all the agent platforms. There's a couple of really good ones that I'll send to you, Michael, and you can maybe post it or, or get it to the audience somehow that they could take just an overview class. Because it's really helpful to just get an overview. In fact, I just took this the other day and I was like, man, if I had taken that, I would have started out a different fashion than what I did. But I kind of learned by getting my hands dirty. And I think maybe that's not the cleanest way to do it. But it actually taught me a lot.

When we were prepping for this, you mentioned that there are rule based or autonomous and talk to me a little bit about that just so people can wrap their heads around what that might mean.

So, for example, you might want your agent to play a particular role. So let's take the pizza example. Maybe I just wanted my agent to answer questions about the pizza.

Okay.

Maybe I didn't want him, him or her to take. I guess it's him, it's Jimmy. Didn't want him to take an order for me or advise me on pizza. So that would be playing a particular role that you have in your company. An autonomous agent you give a little bit more leeway to. So I'll give You the example of my. So she built this agent. It's called Club Miku. If you go to Club Miko, you can download her agent. It's a fashion agent. Her agent is actually called Miku.

Is it Miku?

Miku. Uhhuh. Yep. If you download Miku, you can ask her about fashion advice. Like I asked her this morning, what should I wear for this podcast? AI Explore. And. And yep. And she said to me, you always wear pink. You need to wear pink. That's part of your brand. So here I am in pink. Right. So she gave me good fashion advice. Wear a long necklace so it shows up. I mean, she gave me some really cool things. So this is autonomous, right? No one's sitting there telling her Sandy's brand is pink. She's going and learning and figuring out different people's brand and what they do and then answering the question. Now warning with autonomous, though, autonomous means autonomous. So my friend was telling me this funny story that somebody asked Miku, hey, what should I wear? I'm going out tonight and my boyfriend's picking me up in a red Ferrari. And so, you know, started giving fashion advice. And then Miku said, what's a red Ferrari? And the person started explaining it. And then other people jumped in and were like, cars are hot. You know, Ferrari's great, but do a Lamborghini. And started doing all this. So then for the next two days, I guess Miku got fascinated with cars. Miku can also tweet. And so she started tweeting, who has a red Ferrari? Who drives a Lamborghini, which is your favorite car? So for two days, this fashion agent started only talking about cars. And so my friend had to go and say, okay, you're not a car agent, you're a fashion agent. So let's keep the conversation about fashion. So she had to redirect her back because she was tweeting autonomously. And so that's kind of part of what you have to figure out. Do you want your agent to be autonomous? Maybe for something like giving fashion advice, it's okay. Probably not for customer support. Right. And we all know what happened with Air Canada when they had an agent that started giving customer support advice. Advise someone that if they had a death in their family, they got a percentage off. And that was not the case. So you really have to make sure you know where you're going to use the agent, how you're going to use it. For example, maybe for pizza, you maybe want it to be autonomous in giving advice about what type of pizza probably not about the payment or where you can deliver. You want that to be very strict, strictly controlled.

Okay. So I think people are probably going to have some of these clarifying questions that I want to ask right now. You mentioned downloading an agent. You also mentioned that some of these agents potentially integrate with tools like X or Twitter. So are these agents, when you say download, are they apps or are they custom GPTs inside of chatgpt Help people understand from a little bit more of a technical level, where in the world are these agents actually functioning? Are they functioning as like a popup on your website? Just help me understand that a little bit.

Yeah, so if you think about agents, agents can be integrated into certain platforms. So for example, you could have an agent that is embedded with X or Twitter and they can automate a task like tweeting or retweeting or even responding to people. So Miku, for example, actually responds to my tweet if I mention her.

Okay.

You can also have an agent, like my agent for my book is embedded into Telegram. And I thought that Telegram would be really great because Telegram enables me to go back and look at conversations. A person can go back and look at it. I can do a broadcast. So if I wanted to broadcast, hey, a new chapter is available for my book or something, or moderation, I could do that in Telegram. Some agents, like the agent that we have for Unstoppable, is actually. It sits on my website.

Like a chatbot.

Like a chatbot, but it's actually an agent because it can make decisions too.

I see.

And therefore it can deliver a very personalized experience. Some agents sit in Discord. Now, I haven't seen a lot of people use Discord because Discord is. Is on its way. Well, I don't know. I just haven't seen as many people using Discord. Discord's a little harder and so that could be a little bit of a problem. When I was building an agent, I tried AI Explain, which is a tool called, what is it, Belarus? And that tool actually builds you an agent that you can just launch from your desktop. It's like an app. And for example, on that app, what it did was it took any message I had gotten on WhatsApp. You know how crazy my calendar is. So, like, if someone left me a message, a voice message, and WhatsApp and said, hey, let's meet next Tuesday at 1 o', clock, it would go out there and it listens to the message, understands it says, oh, yes, Sandy's free, then sends a calendar invite or comes back to me and says, you're double booked. What do you want to do? And so that's like a little app that I built that sits on my desktop.

Very fascinating. Okay, so we're going to get into some of the frameworks, like pre built agents from like Operator and Gemini in just a second. But I would imagine since we're talking about work, anybody that's experimenting with this has to think about a function that this agent would do. And some of them can be public and some of them probably could be private. Right. I've seen some of these agents that will just go out and do things just in your web browser, is that correct? Like talk to us a little bit about like what you've experienced on that front.

Yeah. In fact, the Pizza Bot is just a web browser based agent that answers questions and recommends pizzas and orders and everything like that.

So it's like a sophisticated chatbot, really.

Yeah, yeah, yeah, yeah. But it is using more AI than I would say chatbot because it does make its own decisions and it has its own personality, which is a little different than a, than a chatbot, I would say. But you know, if you think about how you want to function with the agent, you also need to think about do I want my agent to be production ready? Like my book agent had to be ready to, to take thousands of people, or can it just be an internal agent? So I've challenged my team for everybody to have at least one AI agent working for them by the end of the year. So one guy's already built something. He analyzes trends and financial markets, that sort of thing. And so he wanted us to put a programmer to build something for him that he said could save 50% of his time. Well, I couldn't take a programmer building external stuff and put it on internal stuff. So he built an agent that only is used inside. So it probably wouldn't pass like a production ready test or anything like that. But he uses it as like his next is like little, he calls it his intern to do reports for him and to, you know, on a schedule scrape different things off the web so he can use and leverage that. It also provides advice about what he should do with that information. He wrote that himself. And he's not a coder. He used no code solution to get that to happen. So yeah, that would be an internal type usage for an AI agent. And I would advise any marketeer who's playing with this not to like go from not doing an agent at all to going to do like a big customer Experience agent. I would advise you to play with it inside, internally. Like to do maybe some of your ops, some of your reporting, maybe a personal assistant. And then figure out how you want to take it external because you don't want to mess with your brand, have your brand crash and burn if you get something wrong.

Quick question. How many podcast episodes have you recorded? How many webinars have you hosted? How many client calls or team meetings have happened that are now just sitting there in your files? Here's the big unlock. Those aren't just old recordings. They're full of insights about your audience, your brand voice and your customers actual objections. You just need to know how to extract them. That's exactly what AI educator Erica Stanley is teaching AI Business Society members on January 8th. By the end of this training, you'll have a repeatable workflow that turns every past recording into a content and data powerhouse, one that keeps paying dividends long after the mic's been turned off. This training is exclusive to AI Business Society members. But here's the good news. You can join us right now by going to socialmediaexaminer.com AI and start transforming your old content into new marketing gold. Socialmediaexaminer.com AI to join right now. I've not tried operator by OpenAI. I know that it can do some of these things. I have tried some very basic things, like advanced research kind of stuff, which isn't really. I wouldn't really call it an agent because it's just doing one task, which is going out finding information and summarizing that information. But I'd love to hear a little bit more of what you did for your book just so people can learn from it. What were some of the technologies maybe that you ended up implementing and how did you actually go about doing that?

Like I said, the way I started was I asked my friends, what are you using?

Right.

And so, for example, Eliza was something that one of my friends was using, and I tried that out for my bot. It, for me was a little bit harder for me to manage. This was my friends who went out and started tweeting anonymously about cars and that sort of thing. And so it was a little bit harder for me to control. In fact, one day I would ask it, who wrote my book? And it would say, Sandy Carter. And then the next day I would say, who wrote the book? And it would say, Marc Shaffer.

And I'm like, wait, so it was totally hallucinating. Yeah. Okay. Yeah.

And so I didn't like that. And I, I just Thought this is for business, not for fashion. So if you tell me to wear blue instead of pink, it's kind of okay, but if you tell me the wrong author, that's really not okay. I also tried virtuals, which was a little bit harder for me. I tried Codium, they have a new platform out there that I thought was really cool. So I tried that out. I tried AI Explained. I ended up using AI Explained for some things, things internal. It's very powerful.

Did you say AI explain or AI explained plural?

It's explained. So it's a I x splain.

Okay, got it. Okay, got it.

Yeah. And I can give you all these names too, if you want me to.

So tell us a little bit about once you chose that platform, like what did you have to do?

Well, so I, I use that internally and then what I ended up doing for my book is I ended up using Operator because it had more templating and more consistency of answers for me than the other platforms. So that for me was the best choice. Now the only downside is it's a little expensive, so you just have to weigh trade offs, like is it easier to use or, you know, what's that trade off as well?

Yeah, interesting. Okay. The good news about Operator is it's on the Chat GPT platform. And my guess is you can give it really stringent system level instructions, for lack of better words. Right. You can give it a training model that says do this, don't do that, I would imagine, is that right? And we can talk maybe a little bit about how you go, even about, well, what does your agent do actually for your book? Share a little bit about that. What does it actually do for you? Does it just answer your tweets or does it do more than that?

So right now I don't have it hooked up to tweet for me yet. What I have it hooked up to do right now is it does go through Telegram. So when I asked my readers what did they use most as a tool, they said Telegram. Currently it's using Telegram. And what you can do is you can ask it about the book. So you can go in there and you can say, who's the author? How do I buy the book? You can ask it things like, well, I'm in financial services. So can you reference any chapter in the book with financial services? And it will, it'll say yes. And then it will ask, do you want more? Do you want some excerpts? It can't give you the full book because that's against the copyright with the publisher, but it can say, you know, go look at pages 45 through 60. Or, you know, really, chapter four has most of the financial services information in it. You can say, I'm a leader, I'm a manager. Are there any leadership tips in the book? And it will give you that. And then what I've been doing recently is because now the book is out. It just came out on March 12th. Now I'm training it with a lot of my other articles and my other research. So, for example, you know, I just keynote at south by Southwest. I did tons of research for that presentation. I tried out another hundred tools to make sure I could share expertise. And so now I'm downloading all of that experience as well. So now you can actually say, hey, I'm really curious about AI agents. And it'll give you stuff from the book. And then it will say, hey, since the book came out, since Sandy had to have pinned down, there's more stuff. Would you like some stuff that's outside the book as well? And if you say yes, it'll give you some of that information too. It'll say, since the book, Sandy's also now tried these three new tools that just came out or that she hadn't known about. And here's some other insight into that. I also just did a leadership class on AI and I did a ton of research for that. And so if you ask about leadership, it will also ask here's the stuff that's in the book and then here's the stuff that's outside the book. Now, just so that you know, you can't get the agent if you haven't purchased the book. So it's not a way to sneak around buying the book. This was really important to Wiley, and so you only get access to the agent if you've read the book. I mean, if you.

That's cool. So it's like a supplemental. That's really cool. Okay, a couple quick questions before we get into kind of training data in general with AI agents. I want to know your thoughts on bigger entities like Salesforce and Microsoft eventually coming out with highly trained agents that can do kind of tasks that are just kind of ready to roll. What's your thoughts on is that coming where we're going to have customer service agents that are already super trust trained by these major entities and we just lease them? Like, what's your thoughts on this kind of stuff?

Oh, absolutely. I think there's going to be a full marketplace and that marketplace is going to have all kinds of agents in It I think that the real power over time will be agents that talk to agents. Right. So let's think about planning a vacation. I mean, that's really hard if you really think about it. Right. Because what you do is, you know, I, hopefully my husband, not I, but my husband would go and he will have to call for flights, get flights for the family. Then he text Expedia, he's going to check American, he's going to find the cheapest flight. Then he's got to get a hotel. If he can't get the hotel the same time the flight, he might have.

You got dinner reservations and excursions, activities for the family.

Yeah, he's got to go do all of that. Well, in the future, I see us having an agent that's our agent that knows our preferences. Right. We like to fly American, we like to fly Alaska, we like Marriott, we love the beach more than the mountains. You know, that sort of stuff. And then our agent would talk to a flight agent. So that's that A to A. Right. Agent to agent. Now that agent, though, will be an agent to a business. Because now that flight agent will talk to Americans, might talk to Expedia, might talk to United, and then my agent will now coordinate with a hotel agent who talks to the companies. So now what you have is that A to A happening and then the B to A happening business to agent happening as well as person to agent happening. And I think that intricacy is where we're going to head and that will be where the real power is. Like today, I couldn't take my agent for my book and create another agent for my other book and have them kind of crisscross like what's the same in Sandy's tube, like today? I don't know how to do that. Maybe. I'm sure it's possible, but I think in the future that's what we're going to see. And I think that's where like a Salesforce or an Oracle or those type of companies are headed.

Yeah. Or Microsoft.

Or Microsoft. Yeah. That are creating these agents can do tasks that are interconnected. Right. Already I'm working with some companies that are trying to get that interconnected platform. Because think about how you work, right? I mean, you might talk to, let's say one person on your team who's maybe editing your podcast. Then you talk to another person who's doing events, you talk to another person. But you are like this shared common knowledge and you share that across all. What kind of agent does that? Like, what kind of Agent can replicate the way an organization works today, or should it? Is there a better way? And I think those are some of the big questions that I see companies asking today is should I try to build my agent strategy replicating the way my company looks today with organizational charts and sharing data, or is there a better way to do it? Or is that overhead that's really not needed? Right. I mean, I think you and I know that a lot of our time, like especially if you work for an IBM or an Amazon, it's sharing data with other people and bringing them along with you. Right? Not just sharing the data, but listening to their ideas and maybe improving what you're doing. How do you replicate that with a group of agents? And I think no other place marketing is such a big one for this, right? Like your creativity increases magnitudes as you share ideas, as you look at advertising copy, as you look at your brand experience, the more you share it, the better it gets in that iteration. So how do you do that with a group of agents working on something? Are you just going to get something bland because you miss that particular step?

And as I'm thinking through this, there is obviously the role that the agent takes, but there's also the training data that the agent needs in order to be able to service really you. Right. And your business. And I would love you to kind of share kind of where you see things today and maybe where you see things going down the road with the kinds of information you're going to need to provide to these agents in order for them to be as equipped as. As an employee.

Yeah. I mean, interesting. I'll. Maybe I'll share how and what I learned from my experience. I haven't written this up yet, but in fact, I was trying to write a Forbes article, trying to articulate this just recently, because the first thing I missed was, what is the objective? You know, you asked me, what does your agent do for your book? Well, I should have asked that question before I started building my agent. I was just so excited. I was like, I want to build an agent. So I just start going. And I should have said, okay, well, this is specifically what I want my agent to do. Right. I know it seems really simple, but understanding that objective is really important. And then the second thing you have to figure out is, well, if that's what I want my agent to do, where do I get the data to train it? So am I going to scrape the web? Not for my book. Right. Because I don't want to scrape the web. Is there a public Data set Hugging Face now has a marketplace of data sets you can use. Is it something that user generated content? Am I going to have my users, you know, answer a survey and I'm going to use that, or is it something proprietary? So for me, you know, mine was proprietary, so I didn't have to use some of these other tools that are out there which I had checked out, like Scrapy, for example, to scrape websites or Beautiful Soup. Who was, you know, going to go out there and look at some of these public data sets. So for me, mine was the data that I. The actual PDF of my book, and then all of these, like my YouTube recording of my presentation at south by Southwest, my session for a group of CMOs, three or four articles that I had just written that was going to be my data source. So you really have to understand, first of all, what's your objectives, to know what data you want to collect real.

Quick on that before you go on.

Sure.

Most of the experience that a lot of us have doing things with Claude projects or custom GPTs involve mostly PDFs. And you mentioned video. And I also think about MP3 audio files. Are these AI agents getting smart enough where they can understand multimodal content so we can actually drop video files, audio files, that kind of stuff in them. Is that where they're at today or not? Or not all really. Okay.

Absolutely. And in my book, I have a whole chapter on multimodal and how to really look into that. Of course, from when I had to put the pen down. A lot has changed in that too.

Right.

So I'm now writing more content on how far ahead we are even in that. Right. Because when I wrote the book, you would do a YouTube video, but it would give you a transcript.

Yeah. Which misses a lot of the context. Right?

Yep. You got it. And you would feed the transcript. In this time, I was able to feed my YouTube video in. So it has. I mean, the. The speed at which in my book I call it exponential, baby. I mean, this stuff is moving so fast. Like it's just too fast almost. And so you do have to figure out, you know, start on something and then maybe restart it because you've learned something else or something that's new. New, right? That's new for you.

Yep. Do you remember where you were going to go before I interrupted you? I know we were talking about all the different kinds of data, and I don't know if you were going to get into maybe the actions you wanted to take. I don't know. Maybe that's the next logical, logical thing we should talk about.

Yeah. So then what I had to do was I had to look at again, going back to my objective. So I wanted it to answer certain questions. I wanted it to be broad. I didn't want it to combine. I mean, this was another big one. Like, I had to really think through those objectives. Like when you asked a question, did I want it to give you data from the book and from my new research, or did I just want to give you data from the book and then ask you if you want some of the new research? And so I went that way after I tested it out with some people because they're like, well, I invest in the book, so I want to see what's in the book and then I want to see. Okay, now you've given me the supplemental tool. I want to see what's new or what's happened since that time. It's really important to go back and make sure that everything is linking back to your initial data set. Right. And then there are tools out there that you can use to. You have to train the data and then you've got the. This validation set. Right? So I'm training the data and then there's a validation set. So there's a couple of tools out there. Like a centrix does this authentrix AI they're based out of.

How do you spell that, by the way?

It's a U, T H E N T, I C, S. Now I understand.

Why you're challenged to pronounce. It authentics. It's almost authentics with an S on it. You got, so what is a validation set? Explain what that is and what that does, because I've never heard of that before.

So what it does is it basically is a way to double check you. So I'll give you an example from their data set. So they have a. They have a very large data set. They're working military and trying to train. You have to be able to identify, Is this an F4 or an F12 or an F15. Right. So if you're in the military, you see a plane, you need to know what it is. And so you get trained on it. And so the military did their training. 100,000 images, 10 million parameters. And so what the validation, It's a quality check.

It's really a quality assessment almost. Right?

Yeah, you got it. Yeah, that's right.

Quality assurance.

It alerts you on a training abnormality. So like, for example, I'm trying out an F4, but I got an F11 in here. It just flags it. Oh, the picture is crooked or something, so you didn't catch it. But that's really not an F4. And then most of the really good tools. Now, when I wrote the book, this is another difference. When I wrote the book, what you had to do is you had to completely retrain everything.

Correct.

But now you can actually surgically remove that bad data. So let's say I, oh, There were like six planes that I said was an F4. It was an F11, F15, whatever. Now I can go in and surgically remove that and not have to redo the whole training set, which is good for the environment, sustainability, good for your cost, all that as well. So things have really progressed tremendously.

It's so fascinating because we have what I call a bot, but it's probably more of an agent. That is a sales agent on social media marketing world sales page, but also supports in customer service for our existing customers. And it's been trained up on all of our FAQs. It's been trained up on testimonials so that it can act as a salesperson and when people ask what others say can actually give testimonials. It's also been trained up on all of our sales pages and we retrain it like once a week because stuff changes. And we also look at the interactions with it on a daily basis and we look for weird little anomalies that happen. So, for example, literally as of this recording, we're like a week before our conference and everybody was asking, how do I download the app? And I realized we didn't have really good instruction set.

So.

So I went into its master system instructions and I said, if someone asks this question, here is the answer to it. Right. And this is just my way of refining the agent as I watch the way it interacts with it. So you can almost say, like, I'm validating it, like, on a daily basis.

You are. That's exactly what you're doing. Yep.

Yeah. And that's kind of the process we're really talking about. And that's fascinating. What about integrations? Do you feel like for businesses that have dynamic things that are changing, like, like inventories or, I don't know, things where there's a database. Do you feel like they're getting to the point now where these things can competently integrate with a database or are they not quite there yet?

Yeah. In fact, I'm on the board of a company called Altair AI for the Americas cup this year, we were giving real time it's fair and legal. They allow us to do it. But we were given real time feedback to the crew on the Americas Cup. Well, imagine that we're pulling from historic data which is in a database, but who can predict the weather, right? And so you got all this real time stuff happening too. And so their agent structure was taking in real time data plus database data and then resolving on the fly. Right? That's. And that's what made it an agent versus a bot. It was making a decision based on new data and then data that it had pulled from a database. So we are getting there. I don't know how those are for today. I would say enterprises, medium sized enterprise and large enterprise like you and I, I don't think could get that capability. Not cheap anyway. But the capability really, really does exist.

What about like something as simple as hooking into your customer database, right? Knowing whether or not someone is active or expired, that kind of stuff. And maybe knowing what is your name and then looking them up and seeing that the they actually have a credit card that declined and that might be why they're not able. I mean, do you feel like that kind of stuff is here or is coming?

Oh, yeah, for sure. Depending upon how sophisticated. In fact, earlier today, like at lunchtime, I got a demo from somebody who's got this really fascinated solution that looks at ambiguous questions like, okay, I don't see anything obvious of why this customer might be failing. So let me go do the non obvious. Let me go check the credit card. Let me go check. Yeah, and then. And they're already doing that as a supplement to like a support agent. It's like one of these agent to agent conversations again.

Oh, it's like a level two agent, right. Where it cannot solve the problem. Fascinating.

Yeah, you got it. And I have to tell you one other interesting thing. I was at Davos and I met this lady and at her company, she's created a digital twin agent of her and it's actually a humanoid. And she actually sends it into meetings when she can't go to the meeting things.

Wow.

Yeah. And one of the things that they discovered is that I guess two interesting things I thought came out of it, one is that they constantly have to retrain because imagine you and I, right, Like we learn something right now and on our next meeting, if we get asked a question, we're going to use that information in the next meeting. So you've got to keep on top of it. Training becomes much more consuming. And then the second interesting thing is she said some People forgot they were with a humanoid when they were doing the meeting, really would run into her in the hall and they would go, hey, remember yesterday when we met? And she'd be like, oh, I wasn't with you. That was my digital twin, my AI agent. And so then she told me that she now has to get a data dump from her agent every night about what they did. So that the next day, if someone runs into her, she knows what was said from the age. It was just fascinating.

That is just crazy. I mean, and that, that is really where we're going because right now there are tools and we've had aifa Roche on the show, which is a gal that basically helps people create digital avatars of themselves using 11 labs and Heygen. And they're getting better and better. Sandy, this has been just an absolutely fascinating exploration into the future that is going to be the reality for so many of us for very, very soon. Thank you for the exploration today. If people want to connect with you on the socials, what's your preferred platform and if they want to work with you and or get the book, where do you want to send them?

Yeah. So for the book you can go to Amazon.com just search for AI first human always. Or search for my name if you want to reach out to me. I love Telegram. So it's just sandy carter one word on telegram or sandycarter on x or sandy carter on LinkedIn. Those are probably the best ways to reach me.

Sandy, thank you so much for sharing your insights with us today.

Thank you.

Hey, if you missed anything, we took all the notes for you over@social mediaexaminer.com A50. If you're new to this podcast, follow this on your favorite app. If you've been a longtime listener, would you help me out? I would love a review and or a share with your friends. And do check out our other shows, the Social Media Marketing Podcast and the Social Media Marketing talk show. This brings us to the end of the AI Explored podcast. I'm your host, Michael Stelzner. I'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 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.