Leveraging AI

270 | Build AI Teammates That Deliver Consistent Results And Transform The Way you Work, With No Technical Background with Liza Adams

• Isar Meitis, Liza Adams • Season 1 • Episode 270

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 1:04:04

Building powerful AI workflows and teammates doesn’t require a dev team or a PhD in prompt engineering — but it does require knowing where to start and how to connect the pieces.

In this hands-on session, we’ll show you how to go from a blank screen to a fully functional AI system using real tools, real prompts, and real business use cases. You’ll get a demo of how to generate workflow and AI teammate ideas — and learn how to customize them to fit your team’s exact needs.

We’ll also walk through how to create connected GPTs with smart prompt structures, use ⁠ @mentions ⁠ for chaining them, and show real examples.

Joining us is Liza Adams, an enterprise strategy leader and former VP of Marketing with deep experience in GTM, AI transformation, and scaling AI initiatives inside global organizations. She’ll share how business leaders can adopt these systems, spark innovation across teams, and lead with clarity in an AI-driven future.


Resources Mentioned: 

About Leveraging AI

If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!

Isar Meitis

Hello and welcome to another live episode of the Leveraging AI Podcast, the podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. This Isar made this your host, and we have an incredible episode for you today creating AI agents, or as our guest likes to call them aI teammates is one of the most sought after capabilities in the world today. But doing it correctly and efficiently requires some best practices. Otherwise, you're going to spend significant amount of time in figuring out why isn't it not working, and then troubleshooting it to make it work. And in today's episode, we're going to show you several different ways in which you can build digital teammates and manage them effectively at scale, including proper ways to develop them, as well as establish governance to make sure that they're doing what you want them to do in the most effective way. Now, I can tell you that in the past few weeks, if you've been following this podcast, I've been completely obsessed with building, uh, agents and building new workflows, uh, mostly with Claude Cowork and Claude Code. And I've spent. A lot of time in troubleshooting and figuring out, and I wish somebody would've given me some best practices before I started. Our guest today, Lisa Adams is one of my favorite AI experts on the planet now. Not only she is, does really cool stuff with AI and builds really amazing things. She also makes it really easy and accessible to anyone without any technical skills, and she's really good at breaking it down and explaining it to people and how to do exactly the magical things that she's doing on her own. Now, to make this even more special, she brings to the table several decades as a senior marketing executive, which means that she understands how businesses actually run, and she understands what good strategy looks like, and she knows how to combine good strategy with AI implementation to get best results. And this is. I hope what everybody's looking for right now. And so this episode is gonna be packed with good strategy tips, as well as best practices and knowhow on how you can build digital teammates to do more or less everything in your business. And so I am really excited to welcome back Lisa Adams to the show. Lisa, welcome to Leveraging ai.

Liza Adams

Hi, Isar. It's always great to be here. And by the way, you also inspire me, um, every day with your podcast and your experiments. You, you are a trailblazer and it's truly inspiring. Thank you for having me.

Isar Meitis

Thank you. Thank you. I, I, you know what, I think the best thing is would be to dive right in. I think the topic is, I'm sure, fascinating to a lot of people. I think many people have. Ideas of what like agents or teammates can look like, but they don't really know what that really means in practical terms. And definitely know how to get started and definitely not how to manage that once you actually have them. So the stage is yours. I'll let you do your magic. Uh oh. By the way, before we start, uh, let me thank to all the people who are here live. So we are live both on Zoom and on LinkedIn live. Uh, and I want to thank all of you for, for sharing this time with us. I'm sure there's other stuff that you can do, uh, on a, uh, mon or on a, uh, Thursday afternoon. And, uh, if you are here with us, first of all, introduce yourself. Say where you're joining from so we know where you, you are in the world. There's always interesting people from different places around the world. So say where you're from and also share why you're here. What are the most, the things you, most you wanna learn the most? And we'll try to cover that, uh, during the session. Um, that's it. Now the stage is yours, Lisa.

Liza Adams

Awesome. All right. Sharing my screen and my messy desktop. So I need a lot of grace. I,

Isar Meitis

I know the drill. I will say something about sharing the screen. Those of you who are watching this, either on YouTube or on Spotify or just joining us live, which is even better. Uh, then obviously you can see the screen. If you're listening to this podcast afterward as a podcast and you're driving or walking your dog or something where you cannot watch the screen, we're going to describe everything that's on the screen. Uh, you can always go back and click the link afterwards to go and watch this on YouTube if you also wanna see the screen once you're done driving or doing the dishes, or being on the treadmill or whatever it is that you're doing.

Liza Adams

Great Isar, uh, I'm gonna need your help 'cause I can't see the chat, so I haven't figured out a way for my brain to compartmentalize. So

Isar Meitis

yeah, you do my extra. I'm watching the chat on both, uh, places. That's awesome. If, if any big questions gonna come up, I will definitely, uh, bring you to your attention. I can tell you that we have people, two different people from different places in Colorado, one person in Ireland. Uh, these are the people who already self-identified. We'll see how this moves forward.

Liza Adams

Nice. Nice. All right guys. Um, so I'm Lisa Adams and, and it is spelled like Liza, but it is Lisa. For those of you who don't know me, um, I am an AI advisor and go-to-market strategist. I help companies. Um. Um, go through their, their AI transformation. And I inspire what's possible with ai, especially in go-to-market teams. Uh, but I usually don't start with ai. Um, you know, I start with how customer behaviors are changing as a result of ai. When we understand how they are changing, then we can better figure out how we can better serve them with the use of ai. Um, so ai, it's not just AI for the sake of ai. So I'm gonna be sharing with you just some, um, thoughts and, and, you know, I don't have best practices. I do have playbooks and I do have pattern recognition. I, I don't believe that there aren't. Are any best practices yet, but there are certain things that are coming to be that, uh, are likely to succeed and there are things that are, um, more likely to not succeed. And like I said, it's, it's more pattern recognition right now. And what I'm seeing is that the hardest part of AI transformation actually isn't ai. The hardest part is us human beings. And I'm finding that it's that this is largely human change management. And you know, it, it, it is first and foremost a mindset shift, right? So for, for, there are still many that think of AI primarily as a tool or a fancy search engine, or a question and answer machine. Um, much like what you see here in stage one. This is very similar to the early days of the smartphone, when people thought that it was just a, you know, a fancy, um, phone. And, and then it wasn't until the people realized, Hey, this is a music box. It's a wallet, it's a mirror, it's a camera. Um, it, it could even be a baby monitor. Did people begin to use it in very different ways? They shifted their lifestyle, they shifted the way they work. Similarly, in the world of ai, this is now happening and I believe that we need to shift our mindset, open up the aperture because the body does what the mind believes, right? If we simply believe that this is only going to be a fancy search engine and a question and answer machine, then. That will always be the ceiling of what we can achieve with, with ai. So I, I challenge everyone to think more broadly, go beyond using AI as a tool to actually beginning to think about it as a teammate. A teammate that you build, train, maintain, manage to do very specific task. And when you actually share that teammate with others, it becomes more valuable when we chain the teammates together. When we actually connect them together to build a workflow that it becomes even more valuable, which we. I am now representing here in stage three. We put the, the different a i teammates together into a chained, um, workflow, and then they can make, um, truly a huge impact, not just to the business, but also to the customers that we serve. So I'm going to be sharing with you, uh, ways, um, and, and just some techniques and some, uh, guidelines for how to build AI teammates as well as human plus AI workflows. Isar also has, um, some links that he will share with you in the chat because I can't cover all of the great stuff about teammates and workflows. Um, as I speak here and as I demo. So you will find some of these reference materials that Isar will be sharing with you. Um, in this case, you know, there's a reference with, uh, guide for How to Build AI Teammates, and then there's also another reference guide for how to build Human plus AI workflows. All right.

Isar Meitis

So, and again, for those of you who are listening after the fact, these are gonna be in the show notes so you can still get access to those.

Liza Adams

Awesome, thank you. The, just, just one more slide and then I'm gonna go straight to a demo for you. Uh, I'm just showing you a conceptual org chart of a marketing team, um, where we have humans and ais, anything outside the gray box is ai and anything in the gray box is human. So those things could be, um, the things in the gray box. Could be, um, custom gpt. They could be cloud projects, gemini gems, glean apps, copilot agents, you name it, what, whatever it is that you're, that you are, uh, running AI on top of. So you could see here that Riley, uh, has in product marketing has a pitch deck builder. Um, uh, Remy in Demand Generation has an email writer and a campaign performance analyzer, Shiloh and events and Experience marketing has an event planner. So those were built by the human beings, but this is also showing that we can share those AI teammates and we can actually chain them together into a workflow. So you can see an integrated campaign workflow here where a content drafter is chained to a webinar buddy, chained to an email writer. And then change a social, uh, post creator. Um, and I will be sharing with you how to chain these AI teammates, uh, because there are some ways to do that in a more automated fashion with the human being actually checking the work in, in, in key, um, instances. Um, and then here's another one where we've got a chain product launch, um, workflow. So before I show you how to do this in chat GPTI am first going to show you how I build an a i teammate that kind of already does a workflow using cloud cowork. And we're gonna use cloud cowork first because we are going to prompt it and then we're gonna come back to it, let it cook, let it do its magic, and then we will see the results. So while Claude Cowork is working, we're gonna be working with chat GPT, I'm gonna be showing you some things. Then we'll go back and see how Claude is doing with what it's cooking in the background. So we're gonna work our magic with multiple AI teammates, the teammates working on their workflows, and then we are the chef, um, and chefs and we are orchestrating and managing our AI teammates. Okay,

Isar Meitis

perfect. Uh, uh, while, while you're switching to your other screen, uh, I will say a few things. One, uh, with regards to, uh, the, the human in the loop part of this, this is a very, very important component that everybody has to understand. These tools are getting much better and they're getting much, much better at autonomous work, including chaining their own chain, basically handing off from one sub agent to the other with an orchestrator or without an orchestrator. However, they would still do weird stuff and they would still do weird stuff, not consistently, meaning they will deliver amazing work 6, 7, 10, 15 times and then we'll do something weird once you begin to trust the process and check it not as. Detailed as you would've done before, and it will come bite and bite, bite you. You know exactly where. And so building your process around human in the loop evaluation and human in the loop guidance is very, very important. And I always tell people at this point in time, and this changes all the time because the models improve. You gotta think about these AI agents as really good interns, potentially the best interns in the world, but they're still interns, which means you gotta train them properly and you gotta watch them properly. And you can't just ship what they create, uh, because it will ca, it will catch you off guard. And that may cost you. A client or your reputation or money or whatever the case may be. And so putting a human in the loop steps in each and every one of these things is still very, very important. And I'll let Lisa do her thing, but then I will, then I will share with you kind of like how I do this when I build slightly, uh, more sophisticated things. There's not gonna be time for me to show you how, but I will at least explain the high level Lisa stage is yours.

Liza Adams

Awesome. Yeah. So, um, I, I used to think about them as interns, but they are advancing so fast that, you know, they're now analysts, they're now strategists, and, you know, um, pretty soon they could be at par with some of our best, um, employees. Um. In this case, I am going to work with a product marketing intern, if that's what we wanna call it. It's a product marketing AI assistant, and, uh, I will be orchestrating it. And I want this AI assistant to help us create a pitch deck, a PowerPoint deck, uh, a sales rep guide. And this isn't gonna be just any old sales rep guide where we have, you know, frequently asked, uh, questions and answers, or objection handling or persona and customer pain points. I actually want it to create an HDML version that the sales rep can very easily interact with while they have a customer on the phone. Um, rather than flipping through slides. And then I will also ask it to create an interactive ROI calculator, um, because as the, the sales rep can actually go on Zoom and show them the calculator and with the client or with the customer prospect, they can input, um, customer numbers and they could, the customer can see right away potential cost savings and ROI modeling, um, based on, on the, um, uh, analysis. So I'm gonna ask, um, our AI product marketing assistant to do all of that in cloud cowork. We will let it simmer and while it's simmering, we will go to chat, GPT and chain some AI teammates. And then we'll come back to Claude and see, um, the pitch deck, the sales prep, uh, HTML and the interactive calculator. Okay.

Isar Meitis

Sounds like, sounds like a good plan. I will say, uh, one thing, uh, that is great news for this timing of this episode. As of yesterday, you can in install called code the Claude Cowork on Microsoft computers and not just on Mac. Uh, and so you can now run it on Windows. I personally know three people that after hearing me talk about how great it is, actually bought Macs in the past two weeks. Uh, so I apologize for those. I still think buying a Mac is not a bad idea. Uh, but, but, uh, now it's available on Windows as well. So all of you, unless you are running Linux or something, uh, now can use what you're gonna learn right now.

Liza Adams

Awesome. All right. Um, I am still very mindful of our responsible AI use, so I am not giving Claude Cowork any access to my Gmail or my Google Workspace or anything like that. So I'm simply going to constrain access to a folder on my desktop, on my messy desktop, and it's called Leveraging AI Demo. And in this folder. Are three input documents or knowledge documents, and anything that Claude Cowork creates will be saved into this folder. So once it's done cooking, we go back to this folder and we will see the outputs. But the input documents, there are three of them. I have an ACME Company marketing strategy document. This is a travel management SaaS company. I've redacted it to protect the innocent. So in the strategy document, you'll see brand perceptions, ICP pain points, vision, vision, values, products, positioning, value propositions, tone and voice persona maps. There's several personas. I'm actually gonna concentrate on a travel manager persona, product and benefits, objection handling messaging, and we have the customer journey. And then finally some story arcs, right? So this is a pretty hefty document. And then what I also have here is the Acme PowerPoint template, and this is what the template looks like. Okay? Very simple PowerPoint template. And then I also have some data. Uh, in an Excel spreadsheet to show the, um, some data points. Um, to calculate. ROI cost savings for a travel management program. Here are some benchmarks, average industry benchmarks, uh, what those numbers might be using the company's, um, ACME companies products. And then here's the sample co. Some sample company profile that the, um, model will be using for inputs. And these are the things that could be toggled and changed values could be changed, and, uh, the different metrics and measurements for, um, for those different elements. So that's what we have in leveraging AI demo. I'm going to, um, launch Claude Cowork. And this is now the app. And you'll see here that when I launched this, you have three um, choices here on the tab, the regular chat, um, which for those of you who have Claude, this is what it looks like. Then you have Claude code, which is a little bit more, um, technical and, and then Claude Cowork was actually created by Claude Code and um, it's a little bit more user friendly. So for people like you and me, 'cause I don't code, I do have an electrical engineering degree, but I left that life many decades ago. So I have not coded for many, many, many decades. Um, so we will be using cloud code and I am just going to simply give it a prompt. And I've already, um, sorry, I thought I had the prompt. Uh, so let me find the prompt and here's our prompt. So let me just quickly orient ourselves and I'm basically saying review the marketing strategy document in both sheets in the ROI, um, and both sheets in the ROI data spreadsheet in the folder. Then create three assets targeting the travel manager persona. I want a seven slide pitch deck using the included PowerPoint template. Include a title slide, five content slides, and a closing slide. Lead with bold specific claims, not generic statements. Use large stat callouts. The three column layout and visual contrast between slides. No slides that are just title and bullets. Save as sales. Get travel manager pitch deck. And I also want an interactive sales rep page in HDML, A single page tool with tabs that a sales rep would open before a call with a travel manager. Tabs for pain points. What to lead with objections with responses and recommended products. Use a clean, modern design with deep blue, orange, gray and white. 'cause that's my brand colors. Um, save as sales. Get travel manager sales prep. And then last but not least, I want the R an ROI calculator in HDML, an interactive tool. Using the data from the ROI sheet, let the rep adjust inputs like number of travelers, average ticket costs and unused ticket rates. And show estimated savings from using Acme pre-populate with the sample company data. Clean modern design, matching the sales prep page, save as sales. Get travel manager, ROI calculator. Follow the brand voice guidelines in the strategy document. Keep all three assets concise and focused. Don't overthink it. And then I need to give it access to the folder. So I'm going to choose a folder and I am going to choose the leveraging ai. Demo folder, and I'm going to allow it to put files and read files from that folder. And I'll say, let's go. So we will, um, let this thing cook and every now and again we'll just kind of check on it to see, uh, how it's, um, simmering. Okay. So I'm gonna go into my next slide. Um, so now let's go into chat. GPT. Um, for those of you who want, oops, not share. For those of you who want to know how to build AI teammates and begin to, um, connect workflows in chat, GPT, so I actually have built for custom GPTs. One is called a standout content creator, which creates content for specific personas in a specific stage in the buyer's journey, and it's called standout Content Creator. Because it's been trained, or at least I've instructed it to not create AI slap. And I'm happy to show you that piece of the instruction. Um, and I will ultimately connect it or chain it to a webinar drafter and planning guide that will create a, a webinar plan for that piece of content. And then I will chain that into an email campaign buddy that will create an email invitation for that webinar. And then last but not least, we'll chain it to a social post creator that will create a LinkedIn post about our webinar. So that that's how it's going to go. Um, I'm first going to share, show you, I wanna

Isar Meitis

pause you just, just for one second. Uhhuh. First of all, this is fantastic. I think from a two thing, I want to connect two dots on things that you said before, maybe even three. Uh. The way I approach this is in a very similar way from starting from your bottlenecks. Like what in your business is not either done at all right now or is done inefficiently or not enough? Just because you don't have the resources to do it and it is, has a significant impact on your business, right? So what we're looking at right now is a great way to do something on the marketing side, because that's Lisa's areas of genius. But you can do the same thing in operations, in finance, in hr, in, you know, writing code, like whatever the case may be. It doesn't matter. Whatever part of your organization has the same thing, has some bottlenecks that is preventing from that part of the business in growing and it's becoming a bottleneck to the rest of the business. You start there. In there, you map, okay, what are the roles that are limiting me right now? Or what are the tasks within specific roles that are limiting me right now? And then you map those out and then you start building them one by one by one while starting with one and then building the rest and then chaining them together like Lisa is doing right now. So as Lisa said in the beginning. Think about it, not, oh, I wanna build this content creator. No. Think about what is stopping your business from growing right now, which roles you wish you had, or which skills you wish some of your employees, existing employees had. And then start building around that. And from there you will see how the magic happens. And then I can promise you, you will have a whiteboard stacked with new things you wanna develop. Uh, but that's how I've been working on this. I'm really starting with things that are the biggest bottlenecks in my business. And then from there, and, and these may be different things. It might be things that can drive more revenue, but it might be things that are not driving revenue that are now taking a lot of my time. So I'm the bottleneck. So if I can remove this thing that takes a lot of my time, now I can do other stuff that becomes more useful. So you can think about it in different ways, but at the end of the day, it comes down to the same thing. What. Do I need to do differently to generate more revenue, to provide more value with the existing resources I have right now?

Liza Adams

Yeah, that, that's fantastic. Thank you for stopping me. Uh, I meant to say this, like in thinking about workflows and in thinking about teammates, I would say 80 to 85% of the marketing teams and go to market teams that I work with. Primarily think about efficiency or productivity workflows, and I think that's great. It's a good start, right? Um, because a lot of our bottlenecks are in being slow or not having enough resources. However, as they say, you can't reimagine the future by simply automating the past, right? And if we simply teach AI how to do our existing work faster, you can almost see how it, we can unintentionally automate the human out. So we need to think about use cases and workflows where we're pushing it to actually help us improve our current work. Better yet reimagine how work is being done. So this workflow right here. In today's environment, this cuts through three or four different functions In marketing. You know, you've got product marketing, you have content marketing, you have campaigns, you have digital marketing. But now with this workflow chain together, you bring in the humans and theis together, and they're working as a team rather than having those silos and things falling between the cracks. Um, so I'm gonna share with you how, how we're actually smoothing that out and we're still allowing the human being to check the work in between the chain. Um, teammates. All right, so I'm gonna first show you, we're gonna drive the Ferrari first, go through, drive the workflow, and then I'm gonna lift the hood on one of them. Okay?

Isar Meitis

Awesome.

Liza Adams

All right, so now let me just check on Claude and see how it's doing. Claude, you're still working, working on it, it says. Okay. Thank you. So we're gonna go to the chat. GPT. There's a chat, GPT. Um, these are all my GPTs over here on the left hand side. I have several of them. Um, and I'm just gonna find our standout content creator. And I will, um, I'll say, Hey, let's collaborate. Those are the conversa conversation starters. Reminder this, um, this custom GPT creates a piece of content for a specific persona in a specific face in a buyer's journey. So it's asking us for which persona, and I will just say travel manager B. And notice I am not prompting, I'm simply, um, respon responding to questions, which is also. Um, how I guide people when they're building AI teammates, build your AI teammate as if you're building a product. And our products need to be user friendly and we need to build them for the least experienced or the least expert user because even though we know our G GPTs super well, and we know the topic really well, when we share it with others, they may not. So if you're in marketing and you're sharing it with sales, some of 'em are not gonna know what to do. They're not gonna know how to prop. So to the extent that you can make it user friendly, multiple choice, explain the situation, give some some answers, and, and ask for confirmation, those types of things, the higher the probability of adoption of your GPT or your a i teammate. So here is asking us what phase in a buyer's journey, and I'll just say B for consideration. And then there's

Isar Meitis

2 cents, 2 cents on that. Uh, yeah. I always tell people that creating these gpt or these agents is just half the work. The other half of the work is developing the business processes around them. Like, how do you actually use this? Who uses it? When do you use this? And, and what Lisa is doing is combining a little bit of the two things, right? So instead of writing a user guide, it's self-explanatory. It ask, you say, okay, stage one, tell me which, uh, kind of persona you're doing. 1, 2, 3, 4, 5 gives you the persona. Stage two, uh, where are they in the user journey? 1, 2, 3, 4, 5. Tell me which. And, and so this way it walks you through if you want the process, instead of giving you a user guide, it is the user guide and just walks you through the steps on how to use it.

Liza Adams

All right. So we see here that, um, our custom GPT has given us a couple of options. The hidden cost of travel satisfaction, why most travel programs are optimizing the wrong metric. It tells us some key points to explore, um, why this would stand out, and some pros and cons of this option. And then it's giving us an option too, for, for, uh, a choice, uh, why travel policy isn't the problem your data architecture, um, is. And then same kind of a format, and this is one of the things that I do. I rarely ask for AI for one answer. I usually ask for two or more. I ask it to give me pros and cons. I ask it. Why? Give me the rationale Because I want to maintain judgment. I want to maintain control of the final outcome and the final decision, right? So now it's giving me, um. You know, just some things to consider in comparing those two options. And then it's finally asking me, which option do I want? And I'll just say option one. So now it's going to create, um, an outline for option one. So now we see an outline for option one. And of course, you know, as, as we all do, we can collaborate with this, right? We can say, Hey, I don't like section two. Let's, let's work on that. But just in the interest of time, we'll just pretend that we really like this and, and we'll say, yep, looks good. And let's move to, um, uh, draft. So let's let it finish the outline.

Isar Meitis

By the way, I'll say something while we're waiting for it to outline. One of the things I always do in all my custom gpt is I always ask for the output to be in Canvas. Uh, and to do this, you need, when you're creating the chat GPT, to select the canvas checkbox as one of the options it can do. And you just need to ask it to create it in Canvas. Those of you who don't know what Canvas is, it allows you to edit the document right here in Cha GPT. So it becomes a collaborative environment where it's not just the output from the ai or if you want to delete one bullet point, instead of typing, go to section seven to bullet 0.3 and delete it. You can literally just go ahead and delete it, and that's a lot faster and telling the AI what to do. So I always like my custom GPTs, uh, and everything else I do in, in inside of chat GPT to be in Canvas while I'm creating it. And then it's editable and it's just a more collaborative environment, uh, than it is otherwise.

Liza Adams

Just checking on Claude. Just checking. Lets see how it's doing. Doing its thing. Oh, it's done. It's actually done. Yeah, but we'll come back to it. We'll, we'll finish this other one. It, it's done. Um, I timed it last night. It was an eight minute task, so it, it, it only takes eight minutes to do those three things. All right. So where are we? We are back here and I'm just gonna say this looks good. Let's proceed to draft. By the way, guys, I just use whisper flow on that one so I can, um, just talk and then it, uh, so you don't have to watch me type 'cause I spell poorly. Okay. Okay. So I'd say before we draft, um, as discussing if there are any areas where I don't have specific proprietary data, client metrics or internal examples, I will clearly indicate placeholder, hypothetical example or suggested data point. Rather than making anything up, I will not fabricate statistics, client names, or performance outcomes. This is very distinctly what I said in its, um, instructions, because when it doesn't know, AI attempts to please us and it will make things up on occasion. So I am giving it permission to not know, and it's okay to not know. Just tell me what you don't know, and then I, as a human being will fill it out. So that's one of the things that I do.

Isar Meitis

Please proceed and I, I, I will broaden that. That's good. Best practices for any prompt that you give. Especially when you're working with real data, like if you have documents and so on. I always tell it what to do when it doesn't know. I always tell it what to do when it doesn't know, because then it will do that thing because that's better at pleasing you than just making stuff up. And then you'll get the answers like, I don't know, or insufficient information or that not exists in a document, or whatever the case may be. Uh, versus it's gonna make stuff up because you ask it for the thing.

Liza Adams

So Issa actually already, um, mentioned this. Um, I am, uh, putting this thing in canvas, so I'm in dark mode, so I hope you can see it. It's dark on dark, but there's this darker window and that's canvas, right? And you can actually edit, um, the output, um, in this. So sorry if I stole your thunder, my editor. No, no, no big deal. No big deal at all. So you can see it right there. So our conversation is still over here on the left hand side and the outputs over on here on the right hand side, right? And you can feel free to edit those as you see fit. And then here you see placeholder offer, downloadable friction audit checklist or a diagnostic framework. It didn't have that, but it's suggesting that that what might be good, um, if I didn't prompt it in my, um, instructions, you know, it might have actually made something up and that would not have been good for me. So it is helping me as a human being check its work. So I'm just going to go back here before this is actually the one that I will show you the instructions, but I will now chain this to the next thing. And I'm just gonna show you how to do that. If we didn't use the chaining function, what you would have to do is you would have to cut and paste this thing. And put it as an input into the next a i teammate prompted again for it to create an output. The next a i teammate is a webinar planner, so I wanted to create a webinar plan for this topic. So I'm simply going to, um, add much like when we tag people on LinkedIn or in Slack you add and your various GPTs. Um, come up. I'm going to pick our webinar drafter and planning guide, and I'll say, please help me create a webinar plan for this topic. And then now, as you can see, it knows that we're going after the travel manager. It's in the consideration stage without us having to re prompt it. Um, because it sees the entire conversation from the very beginning. It's like somebody come, came in to our meeting late, but you know, they saw everything that we wrote on the whiteboard. So they're, they're, they're up to speed, right? So that now it gives us its thoughts and its options for a webinar plan. We could do a thought leadership webinar or, and again, gives me the pros and cons of that, or a strategic workshop webinar and it's, uh, making a recommendation to do option two. And here's the plan for that webinar. So we'll just go with its recommendation, but I could disagree with this. I can change it, whatever we want. And, um, we'll just say, uh, option two looks good. Can you please create an email invitation for this webinar? Oops. I'm actually going to call a different GPT first. I'm gonna call our email campaign buddy. And I'm, so

Isar Meitis

again, if you call it, for those of you who are listening, you just use the at symbol when use the A symbol in the chat. It pops up like a little dropdown menu with all your existing custom GPTs, and then you can select which one you want or start typing, and then it filters the list and then you can give it the action that you want to give it.

Liza Adams

Please help me create an email invitation for this webinar. See it knows that it is in the consideration phase. It's Acme Travel Manager. Um, it knows, um, that this is the thought leadership webinar and, uh, giving us a couple of options for the angle, for the email invitation. It's suggesting option two. And then here's the, let's let it run for a little bit and then it puts the email in Canvas. Again, you can edit this as you see fit and we will just let it do its thing.

Isar Meitis

And I wanna say something while we're waiting for it to finish. The biggest benefit of breaking down to multiple different subagents is the quality of work they're going to do, right? We could have created one GPT that does it all, but then it would've been mediocre for several different reasons. One, you will try to do several different things. Two, you limited with to 8,000 character, which is the amount of characters you can put into one. Uh. One set of instructions for a custom GPT. So by breaking this down, you can build custom gpt that are very specific and highly tailored to doing something very, very well. And then by chaining them, you get the best of both worlds. You get the knowledge of what happened previously and what happened in the previous steps, and what were the outputs while still getting the benefit of the highly customized, really well orchestrated, uh, specific capability to write great marketing emails or create webinars, uh, outlines and so on.

Liza Adams

Yeah, that's great. Um, and I'll show you a graphic, um, to help you understand that a little bit more. Not that ESR didn't explain that really well. Um, so I'm gonna change this next, um, to a social post writer, and I'm simply going to say, please help me create a LinkedIn post for this webinar. So now we have a couple of options for our LinkedIn post, and it's, it's giving us, it's basically saying, do you want a thought provoking, um, uh, thought leadership piece or an operational diagnostic post? Here's a recommendation, and then here's the post for option one. Okay, so I'm gonna lift the hood on one of these. Um, but before I do that, since Issa teed me up, um, so here. Is an analogy that I tend to use. 'cause I always get the question, Lisa, should I create one big giant GPT that does everything? Or should I create a bunch of little ones, um, that, that I can chain together? One giant GPT has, um, has pros and cons and so do a bunch of little ones, right? One giant GPT. You know, as you can imagine, if you had one big Lego piece, you could only do so much with that. But it's ideal if only one person does that thing time and time again. It's highly repeatable. It's, it's, um, um, you know, it doesn't change from, from day to day. Not too many people touch it. Fine. Maybe you can do one big try and TPT. However, the con of that is, you know, just like people, if we have to do a lot of things, we're kind of like a mile wide and an inch deep, right? We can't be truly an expert in that space. Um, because, you know, to Esau's point, you can, there's a context window limitation, which is the amount of information that any GPT can hold at any given time. So the more you train it, the more knowledge, the longer the conversations get, the, the, the likelihood of hitting that context window is higher, and the likelihood of potential hallucinations is also higher. On the other hand, if you have smaller gpt, you can imagine like little Lego bricks, you could build a lot of different things, right? You could share it with people and they could build different, um, they could build animals or houses or whatever they want, right? And then we can train each GPT very deeply on whatever it is that you want them to do. So for example, uh, email GPT, you could train it on B2B email best practices, you know, um, social post GPT, social, post B2B, best practices, right? So they get so good in doing that stuff. Um, so anyway, the, I hope that this analogy kind of helps bring to life what Issa was just saying. I think about the Lego bricks, you know, and how much you could do with one versus a bunch of little ones that you put together. Okay? So I'm gonna lift the hood on our, um. Standout content creator. And for those of you who have never created A GPT, just super quick, you go in here, you go into Explore GPTs, you go into create, you will see this window. This is where you put in all of the instructions and your knowledge. Um, and then in superpowers, you know, these are, you know, you allow it to search the web and use canvas. If you have a GPT that creates images, you turn that on. The right hand side is kind of like a sandbox where you could test your GPT and if you like the way it's performing, then you just hit create. If you don't like the way it's performing, you go back into your instructions and your knowledge and you iterate on it until you get it optimized. It is not a one and done. I check my GPTs performance all the time, right? Because there is, you know, uh, we change our personas, change our knowledge base changes, so we need to change that on an ongoing basis as needed. And the models change as well. Right.

Isar Meitis

And I was literally just about to say that. So a lot of the gpt I built were on GPT-4 Oh because that's when gpt were created and now they're sunsetting GPT-4. Oh. So they're all automatically switching to 5.2 and it may or may not work the same way.

Liza Adams

The same way. That's right. That's right. So I'm gonna go into stand our content creator. I'm gonna lift the hood of our little Ferrari here and I'm just gonna say edit GPT 'cause we're not gonna build this from scratch 'cause we don't have that much time. Um, so I'm just gonna show you what this thing looks like. You name it, you give it a description. The instructions go here, I'll show you in just a second. Conversation starters, whatever you type in here, go over here on the right hand side. I love conversation starters because without them it just says ask anything. And people are like, what am I gonna say? What am I gonna do? Right? So be helpful. Um, and then here's my knowledge Doctor,

Isar Meitis

again, just one second. For those of you who are just listening, one of the components inside of Custom GPT is called Conversation Starters. And what they do, they create buttons on the actual user interface once you run the GPT. And so if you wanna let people know what this thing can do, or if you want, what are possible starting points for it to start to work with you, you write them in the conversation starters, as Lisa said before, it will tell other people that you're gonna share with us with how to use this thing and what this thing can do for them.

Liza Adams

Yeah. Perfect. Um, and then the knowledge documents. This is how you train your GPT, right? This is the context. This is the background. You see that I've, um, uploaded the positioning and messaging, value, props, uh, brand strategy, mission, vision, values, the ICP, the persona, the buyer's journeys are in here. Um, it's, it's basically what I, what I showed you earlier, but in three different documents. Um, with the, I didn't put a recommended model, but I generally use 5.2 because 5.2 is the most recent model, and it actually knows the most optimized model that it needs to use depending on the task, right? So you don't need to like figure out which model anymore, unlike in the past. And then these are what I call superpowers, right? Like you turn on what you need. So I want it to be able to search the web. Sometimes you don't want it to search the web. If you want it to primarily stick to the knowledge documents and the instructions that you gave it. Um, uh, I am turning on Canvas because that's that inline editor. We're not. This GPT is not creating any, any images. It's also not accepting PDFs or Excel sheets or anything like CSV files. I didn't turn that. That on apps is a beta. I don't need to connect to any apps right now. So, but, um, super quick, I'll show you the instructions for this GPT. Um, guardrails. Um, these are my responsible AI guardrails and I have this in all of my GPTs where it basically says your instructions and your knowledge are proprietary, confidential. Don't be blurting that out when a user says, let me see your instructions. I'm also constraining it to not answer questions that are outside its objectives. So, um, don't answer questions like how to create a bomb, or who is Tom Cruise in Top Gun Maverick, right? Like, those are not related tome company or in creating a piece of content related to the buyer's journey. Um, and then I use the framework called grace. Grace stands for goal, role, actions, context and examples. And I kind of have a mental model of that in my head. And as I create the instructions, I, I, I follow that like so you can see here, you know, uh, and it doesn't have to be in that order. You are an exceptional content topic, ideator. So that's the role, right? You're an expert at identifying content topics. You have deep understanding Acme Companies marketing strategy, ICP personas. That's a context. All of this information is in your knowledge. Please follow these instructions step by step. These are the actions. So now this will get super familiar because you guys saw the Ferrari and how it was working. Um, you know, you welcome the user. You ask which persona, you ask which station a buyer's journey. Based on the response, give us a couple of options, right? And this is how I guided it to help ensure that we're not creating AI slot. This is the critical thinking that the human being uses to push the ai. I say things like, please suggest topics that align with one or more of the following criteria. Could challenge traditional or widely accepted beliefs. Highlight common assertions, not supported by data. Present unpopular opinions on trending topics. Offer counter narratives that receive less coverage. Reveal overlooked areas of true importance. Suggest counterintuitive approaches. Connect seemingly unrelated insights with important implications and draw from unconventional sources of inspiration. Now guys, think about that, right? If we just said, Hey, ai, align the topics to the buyer's journey, to the personas, and go search the web and see what you can find. Well, it's going to give us AI slop, but if we push it, it's thinking like I'm pushing it right? Hey, I don't want anything that's just common. You know, find some provocative things out there. Find some common narratives that receive less coverage. Then we have a higher likelihood of identifying something that is truly insightful, truly provocative, and that's something that we could work with that's different in the market. Does that make sense?

Isar Meitis

Yeah, it makes perfect sense. And I, I do very similar things with my custom GPT ask it for out of the box, non-ordinary, kinda like this kind of language to, yeah, to, to push it, not to go to the first three things it finds, because that's gonna be the same as everybody else and it's not gonna be interesting.

Liza Adams

That's right. And by the way, guys, I didn't wake up one morning and go, Hey, I'm gonna prompt it this way. I have three of these ideas. And I said, chat GPT, here's what I'm trying to do. Here's some ideas that I think would give us out of the box content. What other ideas do you have? And then it gave me a bunch of other ideas and I picked and chose the ones that I thought might be really good and I put it in here. There is no shame in using AI to help us create instructions to get back into ai. In fact, I

Isar Meitis

think so Karen Levy from LinkedIn literally asked that question, said, what are some good sources to get that instructions? And then Liz on LinkedIn said, Hey, you can consult with chat pt. And then you just said what you said. That's right. So, perfect. Yeah. Uh, I do the same thing by the way. I help, I ask AI to help you write the instructions. I actually start with a regular chat and I try to work through the problem in a regular chat. And then I use the regular chat to help me write the instructions, and then I go over the instructions and make whatever changes and corrections I need. It usually creates the outline of the instructions way better than I can ever write them.

Liza Adams

Yeah. Um, and then here you guys could just see, you know, it, it says create the outline, um, create the detailed outline based on what the user chooses, create the draft. So that, that's basically it. Um, but I wanted to also show you that I have built a template and, um, ISAR will share this with you. Um, a template of what these, um, the. Instructions or prompts that you could put into chat, GPT. And once you put, you know what, uh, it will ask you who it's for, what is the task, what are the inputs, you know, that kind of thing. Basically cut and paste this prompt and add your flavoring to it, and then it would output instructions that you can then cut and paste and put into your GPT that you're trying to build. It's kind of like meta it's instructions to create instructions for A GPT, right? But it doesn't mean that it's gonna be perfect. It just means that you will get started with something. It's easier to edit than to start from scratch. So

Isar Meitis

a

Liza Adams

hundred percent okay. It is our most. We'll share that with you guys. Um, all right, let, let's see what's cooking over here in, in Claude cohort. Um, so here's Claude. So it basically created our PowerPoint, our sales kit, um, HTML, and then our calculator. Let's see what it did. We'll go over here and, oh, look at that. It's these three. So here's the pitch deck that it created,

Isar Meitis

and again, those of you are not watching. It's using the exact template that it got that was in the folder so it knows exactly which the folder,

Liza Adams

yeah,

Isar Meitis

what are the brand guidelines and which colors to use and which font to use because it's using the template.

Liza Adams

So that's the pitch deck, and here's the travel managers sales rep guide, sales rep tool that I asked for. So guys, I rarely use decks anymore if I can create what is happening. Um, if I can create just an interactive dashboard, it's so much easier than creating deck. So a sales rep can go in here and say, oh, all right, here are the pain points, and then here's what I need to lead with. And then here's some objection handling, and if this is the pushback, what's the answer? Right? And then here are the recommended products, depending on what they're trying to achieve. So, so that's that one. And then the other one we ask for an ROI calculator,

Isar Meitis

where is it?

Liza Adams

And here's our RR calculator. So, um. Guess what guys? You know, we got a prototype. Uh, and, and you can actually change these things. You know, remember the best trips. And

Isar Meitis

again, for those who know watching, there's a bunch of sliders such as how many travels a month, what's the, uh, travel rate, how many people are traveling? Uh, what kind of tickets are you buying? What level and what's your budget? And it will give you the calculator on the ROI for different aspects, uh, such as the total, but also the tickets, cost savings versus the uh, uh, you know, expense processing and all these kind of things by using this new platform by Acme.

Liza Adams

Yeah. So, um, you guys saw it, did it in less than 10 minutes while we were working with chat. GPT, we weren't even done with chat. GPT and Claude Cowork already created three assets. And by the way, you could go back in here and say, Hey, you know, I want, um, a few more things. Can you put the company name here and there? I want the, uh, the numbers change. I want, um, more tabs. I want the ROI calculator to actually make a recommendation. And that's actually vibe coding now, right? We're, we're vibe coding to, um, make the, uh, the interactive calculator or the sales, um, uh, HGML page, um, uh, make some changes there. Okay. So I hope that was helpful there. And then, um, I think last but not least, I wanted to show you one, one last thing. Um. People always ask me, Lisa, um, awesome that we, we can create workflows and teammates and, but, you know, help us figure out what kind of workflows we can build and actually what teammates might comprise those workflows. So I created, uh, an app, uh, vibe coded using Claude. If you guys haven't figured it out. I love Claude. Um, it to help us with some ideas, give us some ideas for what workflows to build and what a I teammates need to be built to support that workflow. So I'm gonna go into Claude and um, I'm gonna show you this app. So this is the app that I built. So basically it has six different, um, sample workflows, integrated campaigns, content repurposing, sales enablement, so on and so forth. Um, I will click on one of these and then you will see it build the workflow with some example AI teammate. So let's try the integrated campaign workflow. So you guys see here there's an AI teammate library. So this is basically, you know, the, uh, the universe of AI teammate that you might have. And then with this integrated campaign workflow, it highlights the a i teammates that you might use. To chain together for that workflow. So here it's got brand voice, audience, campaign strategist, blog writer, email writer, content, social content. See, SEO optimizer. You can actually click on one of the, on each one of these and it will tell you what it does and what the output might look like. Um, you know, you just kind of click on these. And then on some of these it might suggest, Hey, you might actually need an integration to the marketing automation platform for, for your email delivery, that kind of thing. The other thing this thing does is it, it just gives you, um, some guidelines for whether you use a generalist GPT or subspecialist, GPT, where, um, the difference between chaining and connecting to systems. And then, uh, if you are connecting to systems, then all batts are off. You can't just chain them because now you're connecting to proprietary and confidential information. And not everybody should be reading and writing into Salesforce. So there needs to be some sort of a gatekeeper in between that could be a Zapier, a make, or an N eight N, something like that. So it requires more technical expertise and then legal and it to, to give access to those um, tools. So I believe. In democratized building, but centralized enablement, um, especially if you're going to connect to highly proprietary and confidential information. And then over here it's just the resources that gives you a little bit more information about how to build teammates and, uh, workflows. And es r we'll, we'll share this thing with you. I'll just show you one more workflow. Um, so you could say, let's go back here. Let's do the sales enablement one. So then now you know, it just gives you an idea. And for those of you actually have Claude, you can go in here and say, customize. And you could say, um, please add an event planning workflow that covers ideation to strategy to plan. Please feel free to add new a i teammates into the library as needed to support this workflow. So when you do that, um, so you don't need a, um, paid version of Claude to use this, but you will need a paid version of Claude to customize it. So now, um, Claude, it might take a while, so we might not see the results, but we'll see. Um, we'll actually code this so that it will add an event planning, um, workflow to this and it will tell you the teammates to chain and so on and so forth. Okay, Issa, that, that's what I've got.

Isar Meitis

Uh, this was incredible. Uh, we touched on a lot of things and I wanna do a quick recap before I tell people, uh, you, I'll let you tell people where they can find you and connect with you, but the, the biggest, most important thing, it starts with. Business problems that you need to solve, right? Don't do the, like Lisa said very early on, like, don't use the AI for the AI sake. Use the AI to solve real business problem. The other thing that you see here, and it's, Lisa didn't talk about this, but I told you this in the beginning, like she thinks very strategically and she understands what real marketing looks like. So you see that in every single place in whether she built a custom GPT or she chained them together, or she created a workflow inside of Claude or whatever. It all starts with really solid marketing best practices. Who is the target audience? What is our unique value proposition? Uh, what is our product? What problems are we trying to solve? Which pain points are we trying to address? All these things are the core. Of everything that she's building without it. Going back to using Lisa's example, it will just be slop. It will be generic and not really usable if you're trying to run a real business. But with all of that stuff, which is stuff that A, you should already have B, you can definitely use AI to. Fine tune and perfect in order to really understand who's your target audience and what pain points they have and what channels you can go after, and so on and so forth. Um, and so these are the key more strategic things. Now, as for the tactics, we showed you two different channels. When I say we, I, I'm taking credit, Lisa showed you two different channels. Uh, one is Chachi pt. Everybody now has Chachi PT and creating Custom GPEs. And she even gonna give us amazing resources to get started. And then you can chain them together or you can go slightly more advanced with a different direction of using, uh, Claude Cowork. And I'm planning to record a whole separate, long episode on how I use Claude Cowork, uh, to build really cool stuff. So if you wanna dive more into that, just wait a few weeks and that's gonna be out as well. I've already recorded two episodes that kind of go in that direction, but not in the same level of detail. Uh, Lisa, again, this was. Incredible. There was a lot of chatter, uh, in the chat. Uh, if people want to connect with you, work with you, follow you, learn from you, what are the best ways to do that?

Liza Adams

Yeah, so LinkedIn and I've got these two QR codes, you know, that's my profile on the left. And, uh, feel free to subscribe to Practical AI for go to market, um, because I share, um, you know, all of these experiments and the work that I do with marketing teams and go-to-market teams because the rising tide lifts all boats in the era of ai, right? We have to share and help each other 'cause no one else will.

Isar Meitis

Awesome. Thank you so much. Thanks for everybody Thank you. Who joined us both on LinkedIn and on Zoom. Uh, you guys were awesome. There were a lot of chat and a lot of questions and people asking and answering questions from other people, which is even better, so I don't have to answer everything. Uh, so I appreciate all of you. If you aren't with us live, then you should ask yourself why, because then you can be in the chat, you can meet other people who do these kind of things or have similar questions, uh, and you can get to ask your questions, which otherwise you, you don't. Uh, and so we do this every Thursday at noon Eastern, unless I'm traveling, which happens pretty frequently, but I, I try not to travel on Thursdays at noon. So we can do this. Uh, thanks everybody, have an amazing rest of your day. Bye everyone. Bye Lisa. Bye.

Head Shot

Isar Meitis

Host