Agentic AI: Build Your "Micro-Agency" of One

[00:00:00] Did you know that out of all generative AI users, only 1% can be considered experts? Most of us, we're talking over 80%, are still just dabblers. And here's the thing, it's not because AI is too complicated, it's because we've gotten comfortable using it for basic content creation when there's so much more we could be doing with it. In this episode, I sit down with Matt Collette, former Global Managing Director at Edelman, who now helps companies unlock

[00:00:30] the real power of AI. Matt is going to show us how to move from beginner to intermediate AI user by leveraging custom GPTs, projects, and even building our own AI micro-agency, one agent at a time. If you've been wondering how to leapfrog your competition when it comes to AI, this episode is going to be an absolute game changer. So stay tuned for the next episode of the Your Digital Marketing Coach podcast.

[00:00:57] Digital, Social Media, Content Influencer, Marketing, Blogging, Podcasting, Vlogging, TikToking, LinkedIn, Twitter, Facebook, Instagram, YouTube, SEO, SEM, PPC, Email Marketing. There's a lot to cover. Whether you're a marketing professional, entrepreneur, or business owner, you need someone you can rely on for expert advice.

[00:01:22] Good thing you've got Neil on your side. Because Neil Schaefer is your digital marketing coach. Helping you grow your business with digital first marketing, one episode at a time. This is your digital marketing coach, and this is Neil Schaefer.

[00:01:47] Hey, everybody. Welcome to episode number 441 of the Your Digital Marketing Coach podcast. This is your digital marketing coach, Neil Schaefer, and I want to be one of the first to wish you a very happy new year. It has been a while, but all is good, and I want to get back on track here. So, as you know, AI continues to evolve, and how we use AI continues to evolve, which means for us, we need to continue to up our game.

[00:02:16] And that's really the theme of today's episode. I'm really excited to bring in someone who's going to help us all raise our AI game to that next level. My guest today is Matt Collette, founder of Sequencer.ai. Now, Matt comes with some serious credentials. He spent 20 years in marketing communications, including time at Ogilvy in China, where he actually worked with Microsoft Research Asia, back when they were doing what they called Blue Sky Research on machine learning and computer vision.

[00:02:46] More recently, he was the global managing director for digital growth at Edelman, where he was looking at new technologies and introducing them to clients. Now, when ChatGPT launched in November 2022, Matt saw the opportunity and took the leap to start his own company. Now, he helps marketing and communication teams adopt and scale the use of generative AI, and he builds agentic workflows for companies. Today, we are going to cover a lot of ground, from the difference between custom GPTs and projects,

[00:03:15] to creating instruction sets that actually work, to building AI workflows using automation tools, and even how to create your own AI microagency with agents. Matt's also going to share a killer use case using Code Interpreter that I think is literally going to blow your mind. So, without further ado, let's get into it with Matt Collette. You're listening to Your Digital Marketing Coach. This is Neil Schaefer.

[00:03:45] Hey, everybody. Welcome to another edition of the Your Digital Marketing Coach podcast. AI, AI, AI. You know, this episode is going to be released near the end of this year, 2025. And as we look into 2026, you know, we know that AI is not a fad. We're probably already using AI every day as part of our marketing, part of our business. But how do we get to that next level? We've heard of this promise of agentic AI. We hear about companies that have employees that are all AI bots.

[00:04:13] How do we go from just creating simple content using ChatGPT to being more sophisticated and therefore leapfrogging the competition when it comes to AI? I am really excited today to introduce a special guest who is an expert, has his own AI company, and is going to help guide us through getting from beginner AI user to intermediate AI user, if not expert. So without further ado, I want to introduce my special guest for today, Matt Colette. Matt, welcome to the Your Digital Marketing Coach podcast.

[00:04:43] Hey, Neil. Thanks for having me. Really appreciate it. Excited to be on. Yeah, no, I'm excited to have you here. And I definitely want you to introduce what you currently do now, your company and all that. But before that, I'd like to take a step back because AI, I mean, AI has been around since we were born, but not in the form we know it as today. Okay. So where did this all get started? What is your angle? How did you get into AI? What were you doing before that? And how did you see this opportunity with AI that many others did not? It's kind of a full circle story, actually.

[00:05:11] When I started my career in the early 2000s, I started in communications with an agency in China, actually, of all places, called Ogilvy, which is actually a global marketing communications firm. And one of my clients was Microsoft Research Asia. And at the time, they were doing what was called blue sky research, essentially working on the innovations that would come out 10 or 20 years in the future.

[00:05:34] And they had a bunch of researchers there that were working on things like machine learning, computer vision, a lot of AI technology that were actually has come to life in a very, very real way today with ChatGPT. So that's kind of where it started. I ended up spending 20 years working in marketing and communications agencies.

[00:05:52] And before starting my company, Sequencer AI, I was working at another communications agency called Edelman, where I was the global managing director for digital growth, which basically means that we were looking out for new technology, startup companies that were pushing the boundaries of what was possible within the marketing and communication space and introducing that to clients and helping them understand how they could take advantage of it.

[00:06:15] And then in November of 2022, basically, ChatGPT came around and kind of blew everybody's socks off. And I decided it was an opportunity I wanted to take advantage of, and I started my company, Sequencer. And basically, what we do is we help marketing and communications users, professionals adopt and scale the use of generative AI. And we also build agentic workflows for companies as well. Awesome.

[00:06:41] You know, I'm assuming none of my listeners are on the fence about this, but this is someone coming from Ogilvy and Edelman, two of the most famous, you know, marketing communications agencies in the world, if not the most famous, as well as advertising, obviously. And using, you know, from that background, leveraging generative AI, agentic AI. So I think this just shows how mainstream this has become. And I need to ask you because you mentioned you were in China. I can speak a little bit of Chinese. Yeah. Yeah.

[00:07:05] I mean, I was, I was, it was just a time where you, you're pretty much an environment where you needed to learn Chinese to get around. And so my, my comprehension is very good. But over the years, since I left, I've kind of lost the speaking ability. No worries. No worries. I'm a fluent Chinese speaker. So I always enjoy the opportunity. And yeah, Edelman, I have friends that have worked at Edelman. I did a lunch and learn in their Tokyo, Japan office many, many moons ago. But yeah, extremely respected company, obviously in the space.

[00:07:32] So amazing pedigree and experience that you bring marketing communications. You saw the opportunity with AI. It's November, 2022nd. ChatGPT is announced. I'm sure you were one of the early users. Tell me, did you at that moment say, oh my gosh, I'm going to start a company or what? How did that, how did that start out? You know, ChatGPT comes about still a full salaried employee. You know, what was the timing and sort of the thought process about making that switch? Yeah. I always wanted to start my own company.

[00:08:01] And I actually was doing a lot of work in digital, social media content, helping companies adapt their marketing approaches and strategies to a new world of how people are consuming information. And ChatGPT, it was just, what is that moment? What's that thing? And, you know, Bitcoin had come around. There was blockchain. I was super interested in that. And I had a whole bunch of ideas, but I never actually took the leap to doing it. And I knew I was kind of getting to the point where either I'm going to do something on my

[00:08:30] own or I'm going to stay as an employee of a larger company. And ChatGPT, generative AI just happened to be that moment where I felt like it was time to step out and try something. Awesome. So, hey, everybody listening. We have a lot of entrepreneurs that listen to the show. So I know a lot of young kids, they want to become YouTubers. And Mr. Beast has been doing this for 15 years. But as you can see, AI companies have not been around that long because the technology has not evolved until recently. So even you, if you wanted to start your own company two years from now, you might be where

[00:09:00] Matt is today. So that's very encouraging. So thank you. I think there's a lot of solid employees who think the exact same as you do, but they don't take that plan. But anyway, this is not a podcast about starting your own business. We want to get into the topic of today. So when you, or I should say when we were prepping for this interview, we were talking about all the ways in which people don't understand how to use generative AI, that the use cases are just way above and beyond. In fact, I think your quote is most people are severely underutilizing gen AI.

[00:09:29] So why don't we start from there? You know, it's 2025. We all have ChatGPT accounts. We might be using Gemini or Claude or, you know, Grok or who knows. And we're already getting comfortable of using AI for basic content creation. What are we missing out on? So for me, generative AI is an extremely powerful technology, as I think most of your listeners will also recognize having used it themselves. There's a super interesting study that came out a couple months ago by Section School.

[00:09:58] They do a bunch of AI training and training on marketing and communications topics as well. And one of the things they found is that there's, out of all of the users of generative AI, only 1% can be considered experts. Most users out there, we're talking more than 80% are considered dabblers or kind of novice users. And part of that is because of how we have adopted the technology. So generative AI really is the first technology that's come around that allows us to engage

[00:10:26] with technology in a way where we're using natural language to do so. Like you just go onto the app, you put in a couple sentences as you would speak to a person or write into a document and you get a response. You don't have to learn any real specific skills or competencies to start using the tool. And you can already get a ton of utility out of it, right? And so that lower barrier to entry for technology has kind of also put us in a bit of a mode of complacency in terms of trying to understand all the different elements of it.

[00:10:55] So the average user is really just using generative AI for content generation. But there's a ton of other ways that you can use it. You can use it to synthesize information. You can use it to mimic different scenarios. You can use it to predict information. You can also use it to game out various different aspects of the work that you're doing, message testing. So there's a ton of applications that go beyond simple content generation that unlock a lot of additional dimensionality in terms of what you're doing.

[00:11:23] So based on your experience at Sequencer working with companies on their AI, are there a few specific types of, hey, this is one type of use case scenario that most people are not taking advantage? Are there a few pretty popular types that you end up teaching and training and consulting with companies on over and over again that you see? Yeah. I mean, from a lot in the scenario elements, prediction, emulation, I would say are like the core ones.

[00:11:48] So one of my favorite ones is basically decoding an executive or say an individual style of talking and engaging and coding and then building a custom GPT where now you can generate content in their voice very, very easily. Right. Another one is using a project in ChatGPT or these features exist across the other tools as well to do things like adapt content to your style of writing. Right. So there's a lot of different ways.

[00:12:15] I think for me, like where you start to go from basic to novice usage of ChatGPT and generative AI tools to more advanced is where you start tapping into the capabilities of custom GPTs, projects, and some of the other generative AI tools that are out there as well. So decoding other people, meaning that you can both, if you are working for a big company, you can now create content in the voice of the CEO? Yeah.

[00:12:39] On the flip side, you can also better understand how a prospect whose language you can decode, how to better communicate with them. I'm assuming those are two use case scenarios, correct? Yep, totally. So anytime that somebody has spoken publicly, any emails that they have, for example, you can actually take those, go into ChatGPT in a standard chat window and basically ask ChatGPT to help you decode some of the phrasing that they use, their word choices, how they structure

[00:13:06] their sentences, some of the different ways that they write content. Some people use smiley faces a lot. Others don't, right? And from there, you can actually get a style guide and you can use that style guide and then open or create a project in ChatGPT and basically add that as an instruction set. And now anytime that you want to write in the style of that person, then you can start to do that. And brands can do this as well. So like we've talked about individuals, but you can also do that for a brand, right?

[00:13:33] So when you're creating content on ChatGPT, you're essentially getting ChatGPT style out of that, right? And that's what we're seeing is more and more people using generative AI to create LinkedIn content and otherwise. And there's a certain generic element to that. So how can you actually start to personalize it more? You can actually do what I just talked about for an executive, for a brand, for yourself, for anybody else. So now when you create content, you're creating it in your voice versus in the voice of ChatGPT or open AI.

[00:14:01] And that helps to give it a bit of a distinctive element, helps make it more authentic than what you would get just from using ChatGPT, which also helps a lot in terms of how your readers are consuming that content. Yeah. You know, this reminds me, HubSpot just recently introduced something they called Loop. I don't know if you, it came out in a blog post, but basically the system they say of how to work together with AI. And that first step really is, in fact, using AI to help you develop that brand style, to analyze your voice.

[00:14:27] So that going forward, you make sure that that instruction set's included to everything sounds uniquely you. Yeah. So yeah. Yeah. You know, for those that are curious that are just getting started, or maybe you work for a company where you don't really have that is really just recording conversations, transcripts. I literally upload my book and said, this is like the voice and tone vocabulary I use in my written vernacular, right? Which makes it a lot easier. So the more content you have, I would assume the easier it is to be able to create those guidelines. Correct, Matt? Yeah, totally. Yeah.

[00:14:56] And what I've also found is that spoken word content transcripts, like for example, transcript from this podcast really helps to nail somebody's style and tone a lot more than say like an article that they've written, right? Because often that article will go through several different people. There'll be other folks that are editing it. So you get a lot more authenticity if you use spoken word content, audio content to actually create that style. Makes a heck of a lot of sense.

[00:15:21] So the other thing you mentioned as we were prepping was the ability to build systems that run entire workflows and even optimizing execution across channels, plan campaigns, analyze performance, generate insights. So the workflow thing is one that really stuck with me because as you mentioned that people, myself included a lot of the time are using AI in very, very isolated ways. Yeah. So when you mentioned workflow, and now I am assuming, and I've seen people speak about

[00:15:51] this of using tools like Zapier to basically help automate. And I think a lot of our listeners use Zapier, understand, make, and how those work. But can you sort of go through what, you know, the listener, if they were going to develop an AI workflow in marketing, for instance, what would that look like? What sort of tools education are they missing in order to be able to do that? Or are they able to do that? They just need to better understand how AI works. Yeah, I think, so I guess there's kind of two ways to think about this. The first one is kind of the low tech version.

[00:16:19] And the next is kind of the higher tech version of it. And there's steps along that spectrum, right? Sure. What I would encourage people to do is, you know, oftentimes we're kind of in this automatic mode of using tools and going through our day and kind of doing different things. But one of the things that we do with clients is we will actually sit down with them and break down a specific workflow and then figure out how now we can use and apply automation to that workflow, apply different ways of augmenting that process.

[00:16:48] So for example, if you're creating content, you may go through the process of, okay, first we need to identify the different topics that we want to write about, right? And then next we actually have to create drafts of that content. Then we want to transform that content into something that sounds like what we do as a brand or individuals to make it more authentic, right? And then we want to actually test that against an audience or see whether or not that's going to resonate. So if you're just to take those various different steps, you can now create a series of rather

[00:17:14] than doing everything in one chat on chat GPT, for example, you can actually start to create a series of projects or custom GPTs where you're actually taking information from that and putting it into these specialized versions of the tool that will help you get much better outputs and actually accelerate your workflow a lot more. And actually, when you go through that process of breaking down the workflow, that's also how you can start to think about using tools like If This Then That or Make, for example,

[00:17:41] or N8N, a lot of automation tools that basically string each one of these different steps in a workflow together. And so that's a super useful process that any individual, any team can go through to really start to understand how they can make better use of generative AI and how they can then add things like decision intelligence to what they're doing, to gather outside data to inform the content that they're creating and also testing that potentially against a synthetic audience. Gotcha.

[00:18:09] So I think that that workflow you just went through, a lot of people, probably including myself for some things, are doing each one of those things individually, right? With individual conversations, what have you. So, and correct me if I'm wrong, and I think we should also cover custom GPTs versus projects because I find myself using projects a lot more than custom GPTs. However, I do understand that you can call custom GPTs in a conversation by using App Mark or whatever it is and finding the name of that GPT and bringing it into the conversation.

[00:18:36] So I'm curious then for those that are listening, I have found after using custom GPTs for a while that after a while it shouldn't forget, but it does. And the quality sort of goes down and I've become a big fan of projects. And I've heard other people on podcasts talk more about projects than GPTs. So why don't we first cover that? And then if someone wants to create their first workflow and with projects, I'm assuming you can't create these custom workflows because you can't call on other GPTs. So that sort of negates that.

[00:19:05] But I'd like to cover sort of your take on GPTs versus workflows. And then how can someone listening, literally while they're listening, create their first elementary workflow, maybe tying together two custom GPTs? I think that would be great for the audience. Sure. So as you said, why don't we start with projects? So one of the things I really like about projects, I think it's a great feature of ChatGPT. One of the things I think is awesome about projects is the fact that you can essentially

[00:19:33] use it as a, think of it as a folder pretty much, where you're putting a bunch of different files where within those files, it's a knowledge base for the tasks that you want to complete. And all of the files that you've created or curated are specifically there to help adjust and inform the tasks that you're doing. Right. And in addition to that, what you also do within projects is you create an instruction set. And that instruction set essentially tells ChatGPT what you're trying to achieve. So I've actually created a couple of these.

[00:20:02] One of my favorites and one of the most simple ones is actually what I call a LinkedIn post drafter. And what I've actually done with this one is I've gone into the, to ChatGPT and I've actually gone out and found what are the best practices for creating LinkedIn content? You know, what do you need to start with? How's the structure of the more popular ones? And I found a bunch of articles about that. And I went into ChatGPT and I said, synthesize all this information into the top 10 tips that you would give anybody for creating LinkedIn content.

[00:20:31] And then with that, I actually asked ChatGPT to create an instruction set for me, which it did. And now I've kind of got a step-by-step workflow that I've created as an instruction set that I'm using in ChatGPT anytime I want to create a LinkedIn post, right? And I don't have to come up with a complicated prompt in order to create this output. I can do something simple. So for those who are watching, what I've got up on my screen here is the instruction window that talks a little bit about what we're doing.

[00:20:57] And, you know, initially it's, I want you to act as a story-driven content strategist and script writer. And from there, there's information on what the structure of the content is. You know, we want a strong hook. We want to also have an evolution from context, conflict to resolution in each one of our posts. We've got a couple of do nots, for example, not using but or therefore, a couple of different instructions on how we want to kind of create that content, right? So your instructions can be long or short.

[00:21:25] And then what you can also do is then add files to that project that are essentially a knowledge base. And then what you can do is then add a post about your company or your product, et cetera. And so anytime you prompt a ChatGPT project, what it's doing is it's going to that instruction set. It's reading the instructions there. And then based off that instruction set, it is then looking at the files, pulling in for relevant information, and then creating that output.

[00:21:49] So very simply, I can now, within my project, I can now say, create a post about coffee. And it's going to follow that instruction set. It's going to give me a post that includes a strong hook that moves us from conflict all the way through to resolution. And I've got something very strong, a post right away, right? So that's like a simple ChatGPT project that you can create. And what we've got for those who are watching, and I'll just describe it, is I said, create a post about coffee.

[00:22:18] And the drafted post is, I used to think coffee was just fuel. Double shot, slam the keyboard, grind through the day. Turns out coffee is a mirror. And it goes on from there, right? So we've got a very good post right off the bat because we've created that instruction set. We have a knowledge base that we're pulling on, and it's going to retrieve that information at any time it generates an output for us, right? Right. So let's take a step back. So I think that listeners probably understand that whenever you create a project, then the same goes in Claude.

[00:22:45] And I think the Claude example is a great one because if you start a new conversation in Claude, it has no idea what you've talked about in other conversations. So the concept of a project, of putting all these documents so you don't need to inform it, right? These are my brand style guidelines, et cetera, et cetera. I think that that concept is a very powerful one that I'm hoping my listeners understand. When you say create an instruction set for that project, is this a conversation within the project or is this some sort of global instruction set as if you were creating a GPT specifically for that project?

[00:23:15] I think that's where some might be confused. Yeah, it is a specific instruction set that you have created that essentially tells ChatGPT what you're trying to achieve with that project, right? So best practice is that you're isolating these different tasks that you're trying to do. And then you're creating the instruction set related to that task that you're adding into ChatGPT. So let me ask you something. We could do the same thing within a custom GPT. So why the project and not the custom GPT?

[00:23:44] Because it sounds like you're doing the same thing, right? Yeah, it's a little bit. So for me, I use projects for simple straight. I use projects for two things, simple, straightforward tasks. And the second piece where actually the knowledge base is smaller. And the second piece of how I use them is to retrieve information that I refer to often, right? So if I'm working on something for a client, I'll put all of my project files that for that client into that project. And then I can go and ask questions of that project.

[00:24:11] So I'll say, okay, when was the last time I sent them a scope of work, for example? Or when's the next date that I should send an invoice out? Or questions like that. So I kind of use projects for those two things. Custom GPTs for me are actually, you have a larger knowledge base and your instruction set is slightly more complicated. One of the things to the point that you were talking about earlier, one of the places where custom GPT fails, custom GPTs fail, is actually in how people have written the instructions.

[00:24:40] So I can actually go through best practices for that because there are ways that you can actually get very highly performant custom GPTs just by how you write the instructions and the files that you use to instruct how the system works. Gotcha. That makes a lot of sense. So it sounds like you mentioned GPTs are going to be the more powerful. So it sounds like a lot of the work, a lot of client work is going to be in the projects. It would then you isolate the custom GPTs to more of that advanced workflow type of work.

[00:25:08] I mean, how can you give us some examples? That was a great example on how you use the projects. Where would you begin using the custom GPTs? Yeah. So one of the ones that I've created is an executive thought emulator. This is a custom GPT that I built within the platform. And essentially what you can do is I'll just show quickly for folks that are watching. What you'll see here is actually a very, very complex set of instructions, set of guidelines.

[00:25:36] I've actually broken it up into a lot of different elements. So when you're creating these custom GPTs, often you have two options to create them. You've got a create window where essentially you're having a conversation with chat GPT about the custom GPT that you want to create. And that's actually where a lot of custom GPTs fail or break. Because essentially you're going to tell chat GPT, I want to create an executive thought content emulator and it'll write those instructions for you and put them into the custom GPT.

[00:26:05] But when you follow up and test it and you want to refine it and you want to say, no, actually I want it this way or that way. Then chat GPT tends to overwrite the initial custom instructions. And there's always a place that it breaks in the more complex, in the multi-step workflow. Right. Gotcha. So the best thing to do is actually, instead of, you can kick off your custom GPT in create by say, hey, I want to create an executive thought leader emulator. And it will create a name for your custom GPT, give you a photo, all those sorts of things.

[00:26:35] But actually what you should do is quickly switch over to the configure tab. And the configure tab in custom GPTs allows you to then edit the instruction set, upload files that you want to use, et cetera. And so what you want to do is in the instruction set, you want to keep it super simple. You want to say something like, okay, I'm creating this. This is the purpose of this custom GPT. But from there, I want you to then check this file that I've created. And that file will be like instructions for custom GPT.

[00:27:04] And in that file, you'll have if this, then that instruction. So if the user says this, then do this. If the user says this, then do this. Yeah. I was going to ask about the if then and how that comes into play. So this is an additional instruction set within a custom GPT that's located in a file. Yes, that's right. So what you do is you tell chat GPT to go and look at the file for instructions. And in that file, you'll have a series of different basically if this, then that's that you're

[00:27:33] essentially telling this is for an ESG report. So the ESG reporting is actually a very complex workflow, right? You've got various different standards of reporting. You've got different information you want to pull on, et cetera. And if you put that just in the instruction set in the custom GPT and not in a file, then chat GPT gets confused with what it's supposed to do, right? But if you put it in a file and you tell the custom GPT to refer to the file, then every prompt it's reading through the file, it's identifying what it should do based off the user's instructions

[00:28:02] and then executing that based on that. And this is also where you can now refer to other files. So in this instruction file, you might say, anytime the user asks you to create a piece of content about for this executive, refer to styleguide.docx, right? Which is the short for Word files, right? Question, Matt. Can you have a Google Doc, Google Sheet URL instead of a file name as part of the instructions?

[00:28:28] You can, but it's much more reliable if you actually use the files. And those can be PDFs. They don't have to be Word documents. They can be PDFs. They can be Markdown files. They can be simple text files. As long as basically what you're doing with the files is you're kind of segregating out specific elements of the workflow so that when ChatGPT goes through the instructions and it says, okay, now I'm supposed to write in the style of this as a executive, I'm going to refer to the style guide. Or actually, I need to write a paragraph about this product.

[00:28:57] I'm going to refer to the product guide that's in here, right? So your instructions actually refer to those different sort of supporting documents to execute the workflow. Now, you could theoretically do this exact same thing in a project, correct? Or is this something unique to GPTs? No, you could actually do the same thing. But I have found that the reliability is better with custom GPTs than it is with projects in terms of those more complex instructions. Gotcha. Okay. That makes a lot of sense. And I'm sure it's still a common question.

[00:29:25] I'm curious because GPT-5, when it rolled out that weekend, it sort of failed me and I flipped the switch on Claude. I'm using both now. If we have Claude users, obviously you have projects in Claude. Anything differently you would instruct them in terms of how they would use Claude versus ChatGPT in this situation? I think, well, the basic rules kind of apply here are the basic best practices, which is that when you have complex workflows, create a separate instructions document and tell Claude

[00:29:54] or ChatGPT to refer to that instruction set before executing. And in that instruction set, you'll have if this, then that kind of instructions, right? That seems to work a lot better. And that's where ChatGPT or Anthropic, because the biggest challenge is you were saying earlier, it doesn't follow the instructions after a while, right? And so by doing so, you actually get more adherence to the instruction set that you have created. Gotcha.

[00:30:18] And would you say that in terms of reading the instruction set, if then, can you put those in simple language or is there a specific context, a specific language you need to use in order to get the GPT to follow the if that instructions? It's very simple. So like my instructions in the custom GPT are, I basically give a context on the role. You're a communication strategist. Then I'll say something like, you know, your role is to align tone, vocabulary, emotional

[00:30:47] undercurrents and rhetoric to this style. Please refer to the knowledge base for additional instructions and guidance. It's basically three sentences. That's the simple initial instruction set. But then to your question on the if this, then that, I actually asked ChatGPT to create the if this, then that instructions for me based off, I'm basically, this is what I'm trying to achieve. Here's kind of the rules, so to speak, of what I'm trying to deliver. Create an if this, then that set of instructions and it'll give it to me. I read it through. I review it. I make sure that the files it's referencing are accurate.

[00:31:16] Then I basically put that into a Word document or a PDF. So a simple example of that is, and of course at the top, it'll say rules for file use. If this, then that. If the user asks for ESG metrics, performance results, or year-over-year comparison, then use, and then we've got the file name that's there, right? And then we've got another, if the user asks for ESG baseline comparisons to 2023, then use this file, right?

[00:31:41] And so it kind of basically is directing ChatGPT different places to refer to information based off the if this, then that rule set that we have created. Gotcha. And I think this is another one of those use case scenarios outside of simple content creation, which is have ChatGPT or work with ChatGPT to create those instructions. You don't have an ideal customer profile, use ChatGPT to create one, right? You don't have brand style guidelines, use ChatGPT to help you create one. So, and this is a great example, right? Of actually using it.

[00:32:10] And then you know when the output it gives you that if then, that, you know, the context and the language is going to work within ChatGPT, right? Yeah, totally. And I think the other thing that you brought up, which is super important is chat hygiene is what I would call it, which is what I see a lot of users doing is they have extended conversations in one single chat window. They don't open a fresh chat basically. And what happens in that scenario is there's something called context window, which is

[00:32:37] essentially you can actually think of it literally like a window looking at a screen and information on that screen kind of moving and add the longer you chat with a single chat in ChatGPT, the longer that context window needs to be to go back to your original prompt or to information that you've given it before. So the limits of what we have from a computational perspective means that actually ChatGPT cannot take in all that information. So the longer your chat is, the less adherent it's going to be to your instructions or what

[00:33:07] you're trying to achieve and the more confused it gets. So like a good rule of thumb is anytime things start to go sideways in a chat, open a new chat and that's going to serve you better because it's actually going to be fresh. It doesn't, it's not polluted with other information that you've provided and you'll have a much better outputs, much better experience in doing that. And therefore to the novice users like myself, we'd say, oh, ChatGPT seems to forget. But what you just explained is it's not forgetting. It's actually that context window. Exactly. Yeah.

[00:33:34] And other people have this idea that they're training ChatGPT by having a longer chat. But actually what it's doing is it's just reading what you had previously given it. It's not, you're actually not training it. And the best way to actually train it to get those outputs is to create sort of instruction sets, use projects, use custom GPTs to start getting more out of it and move to those more advanced use cases. And on a side note, I'm wondering becoming now a heavy Cloud user, Cloud will just cut you off. Like, hey, please start a new conversation. I wonder if they do that to prevent that happening, right? Yeah, that's exactly why they do it.

[00:34:04] I think that's a really handy thing that I experienced that recently as well. And that's probably what they've seen in terms of user feedback experiences with that is people started complaining about Cloud not being as adherent to their instructions. So now they say, okay, we're going to stop here. Let's move to another one because our context window is starting to get filled up. All right. So for all you frustrated Cloud users, there you have it. So I want to move on to more of the workflow and even sort of agentic AI.

[00:34:31] And it really begins with a question and then it gets back to that question I asked you, which is if we want to reference a Google sheet instead of a file. And I mentioned that because let's say I want to do real-time analysis. I don't want to have to upload a new file each time to enable that analysis. So I'm curious, when we start thinking about it dynamically, we get into those workflows and agentic AI, how would we begin that process? Is it as simple as starting with dynamic information or using other tools or what else do we need to be doing? Yeah.

[00:35:01] So if you want to get, so basically when we talk about agents, essentially what we're talking about is process automation, right? Getting generative AI, essentially layering an automated process on top of generative AI so that you're using the power and the strength of generative AI to actually work through that entire process. So there's lots of popular different automations that's out there. One of the ones that I think is a really kick-ass use case is basically creating content based off of things that are in the news or that are relevant to what's happening in the zeitgeist, right?

[00:35:30] So you could actually every single day have something like a Google news alert, for example, get sent to an automation tool like make or if this, then that, or all of the different tools that are out there. N8N is another one. And based off that, have ChatGPT ingest that news information that you've got or information from that Google alert, and then give you a couple of ideas for different articles to create. And then from there, actually draft the article.

[00:35:56] So rather than you sitting in between all these different steps, you're essentially automating that entire process. So at the end of it, what you get is the output. So for example, you could actually call on a couple of different workflows that you've created to get that output without having to interject all the time. So basically we're using if this, then that as a great example, because it's so simple that anybody can use make.com. I'm a user, but it is a lot simpler, but sometimes it can be complex. Let's go with the if this, then that.

[00:36:24] So basically with the if this, then that, the output is a conversation within ChatGPT then so that the next time you log in, the information you requested is already there. Is that the best way to look at it? Yeah. I mean, and you can get, there's different ways of automating. So for example, in your inbox, anytime you get an email from Google, you forward it to if this, then that, to kick off that workflow, for example, and the output can also be sent to you, right? So you don't have to log in.

[00:36:50] You basically, you're automating the whole thing from start to finish in terms of the outputs. You're sending it to ChatGPT. And then when there's a new output from ChatGPT, you're sending yourself an email. Is that sort of the scenario? Interesting. Okay. Yeah. Very cool. I think a lot of listeners' minds are probably blown at the moment, but I do want to keep going. I want to make sure that we cover this because, man, we've covered a lot already. But as we were prepping for this interview, you also talked about building your own AI-powered

[00:37:17] microagency, one agent at a time, that you don't need a team when you have agents. With simple prompting patterns and a modular approach, you can assemble a flexible stack of AI-powered collaborators using tools like ChatGPT, projects, custom GPTs. Manus is another tool that you mentioned there. So I think what we've talked about so far, projects, custom GPTs, automations, I think that a lot of our listeners already can visualize where this is going. But Matt, take us there.

[00:37:45] Take us to the promised land, the holy grail of being able to create our own microagency and orchestrating all these things working together. Yeah. So essentially, it's basically the next step in that workflow breakdown that we talked about earlier. So if you're sitting down and you're breaking down a task that you do regularly and you're thinking, OK, what are the steps that are involved in that task? And then you create a custom GPT or an automation on if this, then that, or make or any of those other tools.

[00:38:14] Then essentially, what you're now starting to think about is how do I create a team of these things to work together to execute a series of tasks that I need on a day-to-day or week-to-week basis, right? And that's really where you start to unlock your own productivity and your own agency in a way that we haven't been able to do before generative AI. So essentially, you're taking it to the next step, breaking down the workflow. Now thinking of those tasks, how you want to, what are the different tasks that you need

[00:38:40] to execute on a regular basis and actually starting to create a team of what you can refer to as agents or workflow automations to essentially help you work through any marketing, communication or other tasks that you're trying to deliver on. And Matt, I'm assuming here we can also use the power of Chachi PE to say, hey, I want to create a team of agents. Where do I start? And I'll probably ask you what your tasks are. And from there, you begin to develop these one automation, one GPT at a time, I'm assuming, correct? Yeah, totally.

[00:39:06] And I think one of the tools that I'm a big fan of and have become a big fan of over the course of the last several months, which will give you a glimpse of kind of where we're going with this idea of agentic workflows is Manus. And Manus is a Chinese company, but they're using Claude language models to produce their content. But what they've done is they've essentially added workflow automation to any prompt that you can give it. So you can actually execute very, very complex research tasks, very, very

[00:39:32] complex workflow tasks and get a bunch of outputs from that. Manus has two different modes. It's got chat mode and it's got agent mode, for example. And based on the complexity of what you're asking it to do, it will actually break down your prompt, define and develop a series of steps that it needs to execute in order to deliver on that output that you're looking for. And then execute those one at a time. And it does something called computer use.

[00:39:58] And what that means is it's essentially opening a virtual computer to execute some of that work. So it'll actually go off and do research for you, look at different websites and gather information in a way that is actually, from my perspective, better than what you would get with like a deep research. So one of the first things it does, break down that task, go out and do that research. And it will also use computer use to actually go deeper and provide you with an output.

[00:40:23] So like one prompt I've used is break down the business strategies of chat GPT, open AI, Google, Meta, all of these different AI labs and tell me what their marketing strategy, business strategies are, put that in a quadrant that I can visualize. And this is for an article I've been thinking about writing in terms of what are the different business strategies of these companies and how are they evolving? And so I can give that that's actually, you would take days to do that.

[00:40:50] You know, read all these articles, consume all that content, synthesize it all, come up with a proposition. And with Manus, it'll actually break down that task, find the information and then come back with a report for you that you can use to base your perspective off of your research off of. That is really fascinating. I think, ladies and gentlemen, this is a great example that Matt's talking about, how AI doesn't mean that we have no more work to do. It's not going to do everything. It actually enables us to do more.

[00:41:17] It actually enables us to unleash our creativity. And the time that AI saves us, we end up, I think, putting back into AI to be doing more advanced things like you just mentioned. So I think that's a great example. And I was going to ask you about the difference in deep research because in chat GPT or Claude, you can use deep research and make that part of a conversation or a prompt. But what you're saying is Manus just seems to be better at doing the deep research and then taking that information and then acting upon it. Correct? Yeah. Yeah, totally.

[00:41:45] They've actually done something called thinking in code with their platform, which is essentially they're using code to think out the different steps, which is super interesting. And so you actually get, from my perspective, I get much better results, much more relevant results when I'm doing those tasks on Manus than if I were to do it on chat GPT or Claude. So when we think of this team of agents, it's also basically what is your AI tool set as well, right? Because it doesn't always necessarily need to be one thing.

[00:42:12] Chat GPT could be your daily driver, but you may also want to use a Manus, for example, every now and then to do deeper research, deeper tasks that you're trying to execute, where it's one of those things that comes up on an ad hoc basis. You don't want to have a custom GPT creator for that. You just want to go off and execute that one time. Gotcha. And it's manus.ai, I assume? Manus.im. .im, okay. Yes. And they had basically, they were gated for a while in terms of you couldn't sign up easily, but now they've opened it up.

[00:42:42] Anybody can sign up and start using it. And I think it's one of the best AI tools that's out there. And you personally don't have any, being an American, I have to ask you, you don't have any security concerns about using a Chinese AI tool? No, I mean, I don't share anything confidential in the tool, anything sensitive or anything like that. Yeah. And when I'm using ChatGPT, I'm also, I'm using a Teams version. So all of my, and I'm not sharing my data. So like one of the things that people, like a very, very simple thing is you can go into

[00:43:11] settings in ChatGPT for those people that are using the free version. And there's something, there's an item called data controls. And you can go into data controls and actually tell ChatGPT not to use your data to train its models. So it's, it's an option called improve the model for everyone. If you're using the free version of ChatGPT, this is on by default. If you want to protect your information and data, just go into settings, find data controls, click on improve the model for everyone and toggle that to off.

[00:43:39] And that's just a simple, easy, you know, security thing that anybody can do, but also, you know, a bit of peace of mind, right? In terms of how you're using these tools. That's the best practice for sure. I mean, just starting to use Claude, they didn't have that setting. Then after using it a few weeks, the setting came on. And yeah, I think with ChatGPT, correct me if I'm wrong, Matt, but at the beginning, you could not turn that off if you were on the free plan, but now you can. So you don't have to pay in order to get the privacy. Yeah, that's right. Yeah. All right.

[00:44:06] Well, you know, Matt, we've covered a heck of a lot in a short period of time. And I have a feeling we could go on and on and on forever on the conversation. It's really fascinating. You're truly an expert, but I do want to, you know, let you get on with your work. I think this is really a great juncture to really ask more about specifically what your company does, what types of clients that you help. And then if any of our listeners are interested, how they can best reach out to you. Yeah.

[00:44:30] So we work specifically with communications, marketing, and sales teams to help them adopt and scale the use of generative AI. And we do everything from training and enablement to helping them define their AI strategy to also helping build automations and agents as we talked about today. One of the things that I find is most organizations, as we've taught the theme of this conversation has been on the underutilization of generative AI. Most organizations are still underutilizing it.

[00:44:59] Most users don't know what's possible with these tools. And that's what we actually help to do. And the other part of this is one of my favorite quotes to come out of the last few years was from Ashwath Damodaran, who is a Dean of Finance for the New York University Stern School of Business. And he said, when everyone has AI, no one has AI. And it's a very important statement in terms of how companies should be thinking about generative AI with respect to delivering strategic differentiation.

[00:45:27] If you're using ChatGPT, just like anybody else, then it doesn't really give you that much of a competitive advantage. So what we do as part of the strategy elements of this is we actually help companies to define what their strategic differentiation is going to be when it comes to generative AI. What's the data? What are the competencies of their team that they deliver better than anybody else does? And how we can augment that using the technology. Fantastic. And I think, ladies and gentlemen, just by listening to this 30-minute or so interview, you got a

[00:45:56] feeling for not only what Matt can bring to your organization, but also just the potential out there for AI is just so amazing. And I think we probably only scratched the surface. And obviously, every company, every organization, every role within that company have different missions, have different brands, different culture, different legacies. And therefore, AI can take you in so many different directions. So I'm sure, Matt, you're excited about the future as much as I am. I want to thank you so much for coming on and sharing all of your knowledge and expertise.

[00:46:26] Are there any other... I mean, that was a fantastic quote that you had. Any other sort of parting advice you'd like to give our listeners? Well, there's actually one killer use case that we didn't get. If you don't mind, I'd love to talk about just really, really quickly. Absolutely, yeah. And that is using Code Interpreter within ChatGPT to actually help you game out different scenarios, do message testing, those sorts of things. So for example, one of the things I can go into ChatGPT and say is I can say, create 15

[00:46:51] different audience segments and simulate how receptive they would be to a cold brew coffee pop-up. And I'm going to specifically instruct ChatGPT to use Code Interpreter. And basically, Code Interpreter gets ChatGPT to actually start to code out these different

[00:47:17] scenarios and actually give you scale in terms of this scenario planning, message testing, et cetera. So now instead of just having one or two audiences that you're getting feedback on for message testing, concept development, et cetera, you're actually doing this at a much, much larger scale and in a way that gives you very, very relevant feedback. So what ChatGPT is going to do when we ask it to use Code Interpreter is it's going to open a coding window. It's then going to develop those 15 different audience segments.

[00:47:43] And it's actually going to test this idea of a pop-up cold brew coffee product and actually give us feedback on which audiences are going to like it, which audiences aren't going to like it, how they're going to respond. And that's a really, really great way of using synthetic audiences to actually think of how you're going to approach marketing, messaging, communication, all those different sorts of things. And that's a really cool use case and it's available for anybody. All you have to do is Prompt ChatGPT with use Code Interpreter and it will do that.

[00:48:10] So that's a handy way of getting more value out of the tool. Gotcha. That's a, we've heard, I mentioned we had a Steven Lewis on many moons ago who talked about creating these custom AI avatars. So it sounds like using Code Interpreter is a quick way to scale the creation of those. Any other outside of creating custom avatars, any other use cases for custom interpreter that you would recommend? Yeah. Anytime you need to scale something, use Code Interpreter. So say you want to come up with, you're doing an ad campaign where you want different messages for the audience.

[00:48:41] Basically what you can do is add, ask Code Interpreter, give it that first message or that first call to action and say, iterate on this call to action for 50 times, use Code Interpreter and it will do so. And it does it very, very quickly. And that's really the beauty of, of the platform. So like, it'll do things like even say corporate in creating those segments, which we talked about, it'll do things like corporate office worker, remote tech worker, health conscious gym goer, morning commuter at a transit hub.

[00:49:10] All of these different permutations that really create a ton of dimensionality for any marketing and comms team that you'd have to spend, have to have had to spend tens of thousands of dollars on before to be able to message test that with an audience. And then I'm assuming you then take the results from that Code Interpreter and are you dividing that into now 50 different custom GPTs or what do you do with the information afterwards? Yeah. You could take those. Exactly. You know, you're, you're bang on. You could take all those audience segments that are being created. You can then build on those audience segments.

[00:49:40] Say you've got five that you think would appeal, your product will appeal to. You can then create a project or a custom GPT to emulate that audience and go into that project. Anytime you've got product updates, campaigns that you're launching, anything like that to test what way can I position this that will most likely resonate with this audience segment that I've created. Excellent. Hey, Matt, before I forget, if this, then that, does it output to both custom GPTs, conversations and projects?

[00:50:10] Does it do all three or are there limitations when you want to start using automation tools in terms of which types of conversations you have in chat GPT? Yeah. From what I know, it's using the chat GPT, open AI APIs. Okay. So you can't use it with projects or custom GPTs. Okay. But if this, then that, you know, obviously make if this, then that the different automation platforms have different capabilities. But from what I understand, they're using essentially the API to work through that. Gotcha. So that you're going to have to, you're going to have to put that into perspective when

[00:50:38] creating, take those instruction sets from the GPT and project and files and put them in a conversation and then have the if this, then that right to that conversation would be the way to handle that. Correct? Yeah. Or what you do is within your if this, then that workflow, you create a set of instructions and similar to how you would do for a custom GPT. And as part of the workflow, you're sending the instruction set to open AI, which is basically what's happening with the chat GPT, except we've got a nice handy, you know, interface that we're working with.

[00:51:08] So just create that instruction set, send the instruction set with your prompt that you've automated and you'll get a similar result basically. Got it. So this is, I think, one really simple use case scenario. You have your RSS feed. And then when you have a new blog post, when, you know, 15 minutes after you published that blog post, boom, the, all the content you need to promote that in social media is there. I'm assuming that's a really easy way to get started. Yeah, totally. Different permutations of content for exactly that scenario are a great way to start.

[00:51:38] You know, think of a simple task, how you would automate that, all the different, you know, mind map it, all the different outputs from that. And you can automate that entire thing using one of these automation tools. Awesome. Matt, I think our listeners have a lot of work to do now. You've given them a lot of homework. Hopefully everybody's brain is overflowing with ideas now. One last question. Let's spell out your company name. So if people want to reach out to you and your company, they know where to go. You mentioned sequencer. There's a lot of different ways to spell it. You could spell out your URL. That'd be awesome. Yeah.

[00:52:04] S-E-Q-U-E-N-C-R dot A-I. We basically left the E off sequencer because, you know, tech companies, we have to spell everything differently, right? So yeah. All right. There you go. All right. Well, Matt, you've been just a wealth of wealth of everything today. So thank you so much for your time. And yeah, depending on how this goes, you know, a lot of my listeners might say, hey, we want Matt on again. So I'm sure we'll be in touch in the near future. Happy to. Thanks, Neil, for having me. All right, everybody. I hope you got as much out of that conversation with Matt as I did.

[00:52:34] I mean, we covered a hell of a lot of ground from understanding why most of us are all underutilizing AI to the difference between projects and custom GPTs to creating those instruction sets with if this, then that logic and even building our own AI microagency. I think the big takeaway here is that AI is not going to replace us. It is going to enable us to do more. And I am a firm believer in that.

[00:52:59] If you're still just using chat GPT for basic content creation, you are leaving a lot on the table. So here's your homework. Pick one workflow that you do regularly, break it down into steps and see where you can start building a custom GPT or project around it. Start small, but start. If you want to connect with Matt, check out his company at sequencer.ai. That is S-E-Q-U-E-N-C-R dot A-I.

[00:53:25] And hey, if you found value in this episode, make sure to subscribe, leave a review, and share it with someone who needs to hear this. Until next time, make it a great virtual day. This has been your digital marketing coach, Neil Schaefer, signing off. You've been listening to your digital marketing coach. Questions, comments, requests, links, go to podcast.neilschafer.com.

[00:53:52] Get the show notes to this and 200 plus podcast episodes at neilschafer.com to tap into the 400 plus blog posts that Neil has published to support your business. While you're there, check out Neil's digital first group coaching membership community. If you or your business needs a little helping hand, see you next time on your digital marketing coaching coach.