How to Use AI to Dramatically Improve Your Quality

Quick question for you. Do you know your AI level? Not whether you're using AI, because I know pretty much you are, but how well you're actually using AI. We built a free quiz that tells you exactly where you stand across 5 different levels. You get a personalized results page that diagnoses your biggest gap, gives you a custom AI prompt matched to your level, and shows you the one move that will make the biggest difference right now. Take it for free at socialmediaexaminer.com/aiquiz26. That's socialmediaexaminer.com/aiquiz26. Welcome to the AI Explored podcast, helping you put AI to work. And now, here's your host, Michael Stelzner. Hello, hello, hello. Thank you so much for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzner, and this is the podcast for marketers, creators, and business owners who wanna know how to put AI to work. Today we'll explore how to use AI to improve the quality of your content, your products, or anything else. My special guest is an AI strategist and creator who helps small businesses scale with AI. His newsletter is focused on entrepreneurs and helping them employ AI. He coaches executives and helps businesses scale products with AI. Austin Marchese, welcome to the show. How you doing today?

Excited to be here. Excited to be here.

I'd love to hear a little bit about your journey. How in the world did you get into AI?

It's been a fun one, to say the least. So I've always been obsessed with just productivity automation, just finding ways to get more out of the day. It started really in college, primarily just because I had too much going on. So I was playing college basketball, I was studying mechanical engineering, I was running a tutoring business. I was pledging a fraternity, just a lot going on. And my first job out of college was at JP Morgan, where I was a software engineer, and we were building an internal tool that used AI and some basic natural language processing. And so that was kind of my first foray into the tool. Believe it or not, there was AI before ChatGPT. Then fast forward to 2022 when ChatGPT came onto scene. At that point, I was running an engineering team, and then I was also running a 15-person agency as my side hustle. And so again, I had a lot going on at this time and I was like, okay, everybody that I'm working with, this is like everything that I've always wanted. Everyone needs to be AI native today. This is the future. So eventually I left JP Morgan and I had to slow down my agency because one of my clients was this company called IYK. I became their COO and we raised over $30 million from investors like a16z. And that is really where I learned how to scale a business directly from people who help scale, like Facebook, Amazon, et cetera. And while I was there, this is when AI kind of took that next leap. Like, it was like, okay, this is cool. This feels like the future. Okay, now it can really start doing some awesome things. I just felt like I was on the sideline and I needed to get skin in the game. So eventually I transitioned out of that role in November of 2025. Really with no plan besides I need to create space for this technology. And then over the past 7, 8 months, I've been growing my YouTube channel. I've 10x'd my audience there. I've gotten partnerships directly with Anthropic. I've worked with 7 and 8-figure businesses helping them integrate it. And one of the cooler things actually today, I don't know if you're familiar with Alex Hormozi, but my buddy sends me a screen recording and he actually used one of my videos where I was talking about one of his products as an ad to help him sell the product. So I found out about that today. That was a cool little tidbit. So anyway, it has been a crazy 7, 8 months and a journey to get to this point.

Love it. Okay, so you've definitely seen all sides of this. You've seen it from the Bay Area, you've seen it from working in a big business, working in small businesses. What do you think are some of the biggest misconceptions that a lot of people might have today when it comes to utilizing AI?

So it is amazing and it can really do everything, but a lot of the skeptics think that it's like a magic wand, right? I say this, it doesn't do exactly what I wanted. Now it's either a scam or it's a bubble, it's hype. And the reality is that in order for you to get the most out of this, it's a tool and the best people really focus on Calibrating it to their exact needs. And that's how you actually go from this AI slop that everyone talks about and says that, oh, AI can only produce slop. No, you need to calibrate it, and then that's actually where it can produce high-quality results. And so another thing that a lot of people focus on when they're thinking about AI is doing more. Like, how can we increase the quantity? How can we do things faster, cheaper? That personally doesn't really interest me. What interests me is improving the quality of the product and using AI as a tool to help you differentiate yourself, not from the quantity, but from the quality side of the production.

Love it. Let's dig in on, on the quality here a little bit, 'cause we're gonna dive deep in this today. When people focus on AI quality the way that we're gonna talk about today, what is waiting for them on the other side? What are the benefits that they will achieve if they pay close attention to what we're gonna talk about today?

In my eyes, as AI gets better, the cost to produce something approaches zero. So anybody can produce anything anywhere on the planet for a fraction of a cent. And so just the nature of how AI works, it produces an average output. And Naval has a quote, I think it was back in February, where he's saying there's no demand for average. And so waiting on the other side of like, what's the benefit of really focusing on quality is This is where the value will accrue because anybody, let's say in a content or marketing example, anybody can say, hey, can create me 100 short-form content video ideas, right? That's instant. There's no real value there. Now where the value is, is actually having 10 high-quality short-form content ideas because in the sea of, call it AI slop, where average content is free, you need a way to differentiate yourself. And so that's really the benefit here where it's like, as the increase in quantity goes up because of AI, there is actually an increased demand in quality because that then becomes less available because everyone's using these tools in the same way. But when you focus on quality, that's where you can differentiate yourself.

Got it. Okay. So what I'm hearing you say is because everyone is going to eventually be utilizing AI to create anything, You can possibly imagine, the only things that are going to stand out are the things that are exceptional, is really what I'm hearing you say, right? So what used to be exceptional before is now becoming the baseline normal because AI is pretty darn good out of the gate. So if you can actually increase the quality of whatever you're producing as a result of partnering or collaborating with AI now, and you know how to do this because not everyone will know how to do this, now all of a sudden you stand out. That gives you an unfair advantage. Does that affect Yeah, exactly.

And there's a concept that I like to think about called intellectual obesity, where people stop using their brain because AI does the thinking for them. And when they delegate the thinking to this tool, it'll naturally default to an average output. And a lot like when you stop working out, more and more people are going to stop flexing their brain and flexing their critical thinking muscle. And so when I say that there's actually going to be an increased demand in quality, that's because these real strong practitioners are delegating their thinking to AI entirely. And over the course of time, their output is going to regress to the mean. But the people that are focusing on quality, that's where they're going to be able to differentiate yourself. And this isn't just in content. This is in really any industry in a corporate setting, right? This can be a report that you send to your manager. This can be financial analysis. It's all about how you think about these things.

Yeah. And just to help people process this in their brain, when ChatGPT first started doing images, everybody thought it was the coolest thing in the world and they started creating these images. And then everybody was like, wow, this is the coolest thing ever. But then after a couple of months, you could tell that that image kind of looked like all the other images that everyone else was creating, right?

Oh yeah.

And then all of a sudden you're like, yeah, you didn't make that. That was done by AI, right?

And it hurts your, like, your personal brand in a sense that if, if you're just producing AI output, everyone's like, oh, Austin, that's just, he just sent me an AI report. Like he didn't even look at it. And this is really where everyone's zigging. You want to find a way to zag.

100%.

Okay.

So let's say we want to embrace this concept of producing AI quality in anything that we are making. And it could be something as simple as analyzing information and presenting a report to someone. Or it could be something as complicated as writing scripts for, for videos, or maybe even doing strategic consulting, you know, and helping you figure out like what's next for your business. Everybody that's listening probably wants quality. So how do we— what do we need to be considering before we start? Because obviously we know, we know that at first blush, it certainly appears as if the stuff coming out of AI is quality. But we're hypothesizing it's not good enough. So what is it that we need to be thinking about before we get into this process?

Well, and we'll get into like the real actionable ways to do this, but there's 3 steps that I try and follow whenever identifying where to spend the energy. And so that, that first step is actually, where does quality matter? So the reality is that not every task requires a high-quality output, like where good is good enough. And so Identify those tasks where you need that last 20%. Like, it's a classic 80/20 rule, right? Where are those tasks that quality really matters, that are force multipliers for whatever you're doing? And that's where you need to hone in. I call it Sauron's eye. Where do you have Sauron's eye in terms of where quality matters? I don't know if there's any Lord of the Rings fans out there.

Yeah, I know what you're talking about.

But like, where is that task? Identify where quality matters. And what's a force multiplier where going good to great is really going to be a differentiator. Throughout this, I'm just going to be referencing to how I've used it. And so for me, when I first left my role as COO and I wanted to just really open up space for AI, I decided to start creating YouTube content. And the most important thing for YouTube content is your title and thumbnail. That's your packaging. And so this was something where I was like, okay, this is the most important thing For me when I'm trying to grow an audience. So I'm going to really dial in this video packaging. But again, for somebody, for somebody listening or watching, this could be the weekly report they send to their manager. This could be analyzing their sales calls. This could be— there's, there's a bunch of different routes. But first, it's where does quality matter? Then step 2 is figure out who will be on the receiving end of whatever you're producing. And so The goal here is really to identify, in order for you to improve the quality, like, what's the best way to actually get a better quality output and to get better at something? It's getting feedback, receiving feedback, and then iterating and improving. So we want to figure out is who would be providing that feedback and identifying the recipient of what you're creating. So for me, my audience is 30 to 50-year-old ambitious professionals. That's the audience that I'm speaking to. So, okay, that's who I want. Now, how do I actually create something such that I can get feedback from that individual? And a simple way to do this is you can just have Claude or ChatGPT or whatever your AI tool is, have you ask you questions about who your audience is. So you could say something like, interview me about the audience I'm speaking to and ask questions to get a deep understanding of what they like and what they don't like. And the idea here is we're trying to outline the person who's receiving the product or receiving the outcome so that they can later on in the process provide you feedback and then you can proactively change it. So for me, I have this exact persona, but I also like taking it a step further where I try and identify a specific person that I know will be receiving it. So 30 to 50-year-olds. I have friends who are 30 to 50-year-old ambitious professionals. So one persona that I think about is a person, Darren. He's my friend, but he watches my YouTube channel. And so what does Darren like? What are some of his thoughts and how could I document him in terms of what type of content he likes so that I can later get his feedback before I actually create a video idea?

Real quick question here on a couple of things here. The first question is, where does quality matter? And where can you get that extra 20%, that 80/20 rule that can move the needle? And then you said, who is it that I am producing whatever I'm doing for? And then you talked about interviewing me, and we're going to talk about Darren in just a second. But what I wanted to do is back up just a little bit. When you and I were prepping, you said a couple things about it's a process and this concept of going from abstract to concrete. I just want to, like, make sure we don't skip over that before we keep going. Do you remember what we were talking about?

Yeah. So none of this stuff is going to work right off the rip. And so you really need to understand that it's an iterative process of improving your system within how you actually collaborate with AI. So nothing's going to be great the first time. You have to understand that you have to iterate through it. And a lot of these concepts, It's all new to everybody. We're all going through this together. And so one of the best ways to actually learn these concepts is bridging abstract concepts to concrete concepts. So this is when I'm talking about my YouTube packaging. This is my way of bridging this abstract concept of having an audience to a concrete one, which is, okay, my YouTube thumbnail audiences. So whatever you're learning, and in this case it's about improving quality, it's so important to speed up the time that it takes from an abstract concept to get to a concrete concept in whatever you're working on, because that's the most important thing. It's all about your system and the beauty of AI, it's extremely customizable. And so the second you start applying it to your day-to-day, that's when the light bulb turns on and you'll start learning and understanding the real value of what we're covering.

Perfect. Okay, so let's get back to this archetype, um, Darren character that you were, you were sharing the story on, and let's unravel that a little bit.

Yeah. So the goal here is getting a deep understanding of the recipient. And Darren is a friend of mine, but why this is so beneficial is I'm able to pull training data from conversations I've had with Darren, text combos. Hey, do you like this? Do you like that? And that helps me morph and create a persona in my AI system on what Darren does and doesn't like. And I'll cover this more later, but by having a real person, Darren, I can continuously ask him for his feedback and then continuously tweak this AI version of Darren so that it better reflects him so that I can make a better decision on what my YouTube videos are going to be. So an example of this could be if you're sending a weekly report to your manager, and this is something I know the Anthropic team does internally, they'll create a persona for their manager because they have all the feedback that the manager's already provided them in the past. So now before they submit anything to their manager, they're able to run it by them so that they can improve the quality before submitting it.

So what's the advantage to either creating an AI persona of our boss or an AI persona of our ideal customer? Like, help everybody understand why in the world they would even do something like that, just because it's kind of a Radical concept.

Yeah, it is. It's a little crazy. So the benefit here is that it, right, going back to the higher-level goal of improving the quality of the final product that you produce, think of, and we've all gone through this. So like my time at JP Morgan, more corporate setting, right? You submit a report to your boss. The boss is like, I don't like this. I don't like this. You go back, you improve it. You send it again. I like this, but I don't like this. Okay, you go back, you improve it, you send it again. So these feedback cycles take time, energy, and they also impact how the customer, in this case my boss, thinks about you in terms of the product that you're able to create. And so the goal here is you actually remove the human-to-human feedback cycle and you introduce a human-to-AI clone feedback cycle. Where you're able to iterate as many times as you feel is needed to improve the quality. And then at that point, it's like, okay, here's when I actually give it to my boss. And hopefully it's already caught 80, 90, maybe even 100% of the feedback before it's actually shared. And so now you go from, oh, Austin's pretty good, to every time Austin gives me a report, it's exactly what I wanted.

So it's almost like instead of human to human, it's human to clone. Back to the original human and then to the recipient, right? It's like that extra layer of quality assurance almost, right? But it's instant. You don't have to wait for your boss to get back to you 3 days later. You don't have to wait until you submit your next report next week. It happens instantly. The benefit there is obviously you can increase the quality of whatever it is you're producing, be more higher value to your clients, your boss, whoever that is, right?

Exactly. And one of the things that I like to try and do is just think about how this works in a pre-AI world. Like, you know, what would you have done in a pre-AI world? Well, you create a report and then you review it yourself. You look through it, you, you analyze it, you think through it. So this is a way you're not doing anything different. You're just augmenting your current workflow, which is your process of reviewing the product that you're creating. So it's nothing different than what you're already doing. You've just found a way to enhance what you're currently doing to create a better output.

Are you wondering how your fellow marketers are using AI? Well, we just dropped our 3rd annual AI Marketing Industry Report. For 3 years, we've been tracking all the trends. What are the platforms everybody's using? How are they actually leveraging AI? What do they plan to do in the future with AI? For example, today, 73% of your peers are using AI every single day. You can grab the report for free right now by heading over to socialmediaexaminer.com/AIreport2026. Again, socialmediaexaminer.com/AIreport2026. Okay, let's talk about this focus group thing, because I think that you're on to something here, which is kind of cool.

Yeah. So To this point, right, where we've mentioned a single person, Darren, but the reality is that you may have multiple stakeholders. And so this is where I'll use a Claude skill and we'll, we'll dive into this, but I call it an internal focus group. What a focus group is in the pre-AI world is a process of testing a concept to a small group before it goes to a broader group, right? When these massive brands like Nike, they have a concept Before they do a massive national rollout, they have a small group of people where they'll run the concept by them, see what they think, A/B test some things, get some feedback, and then that's their way to have a more informed opinion about their, their broader product. And so it's the same exact concept here, just we're creating AI clones instead of, instead of human people. So Darren could be one person on the focus group. My friend Johnny could be another. I have some other people who are part of my audience who I've now interviewed so that I have a clear set of people that provide feedback on all of my YouTube concepts. So anything that I post has been ran through my internal focus group the same way a massive brand would do this pre-AI. But now we all have access to the same tooling. It's just about creating the proper system So that you can actually do it. And by creating this focus group, you can also like structure, all right, this is a YouTube video. I want these 4 people to review it. Oh, this is a report. I want these 3 people to review it. And it gives you flexibility in terms of how you want and receive feedback. And this whole process, it is trippy. It's almost like AI time traveling, but it's so valuable. Like, if you look at my YouTube growth chart, like I could point to when I implemented this. It was like in April, and since then my subscribers have 10x'd and it's literally changed my life.

Well, and that's how I found you. So let me ask you this. How do we decide who the heck goes into this group and how do we delineate between the different people in the group? Because this sounds like kind of an important thing to think through, right?

The key is you want them to be the recipients, the audience. And I have For me, I have different archetypes who I'm speaking to. So some are founders, some are day job builders, they have a 9 to 5, but they're not necessarily looking for their, their next job. I have technical people. So I have different personas that consume my product, which is YouTube videos. And I have other products, but that's just the one we're talking about. So when picking your focus group, Just think who is on the other side of the product. And a lot of the times, even if it's one archetype, maybe there's a different type of buyer, like a risk-averse buyer or a price-sensitive buyer. So no matter what, and no matter how dialed in you are, there are different people who you have to speak to. And so just think deeper about that. And in some cases, maybe it's, these are broad segments, But it's really about who would be providing me feedback or who would be making the decision to buy or not buy or to review or approve whatever I'm creating.

Well, and it's funny because if I think about the audience that's listening to this podcast, there's kind of 3 different audiences. I've got the marketer who works for a business that he or she does not own. I've got the entrepreneur, which might be a consultant, agency owner, small business owner. And, you know, and then I've got the creator, right, which is the person who is creating videos or writers or podcasters, you know, and they're different audiences. So I could just give them a name, each of them a name, right, and create some kind of background on them. But how do I actually like build them out, right? Because there probably needs to be some sort of data behind the scenes for each of these characters, right?

Yeah, definitely. And it's exactly right in terms of the 3 audiences. And I'm sure a lot of the times when you're debating between a guest or a title or a piece of content, it's like, how do I not alienate one while speaking to them all? It's like, yeah, I mean, that's the art of marketing, right? It's like your zodiac signs. It's like, oh, you're speaking to everyone at once, but you're not Speaking to anybody. It's beautiful. Anyway, so how do you actually do this? So should we just dive into the—

Well, let's talk about, let's talk about the knowledge a little bit, like the knowledge base stuff, because I do think people need to like wrap their heads around like that part of it, right? Because it's pretty obvious, like anybody who's a marketer understands how to create like a basic description of the different audience. But isn't it true that there's some sort of a base of information that's going to go behind that's going to inform them like you talked about? In your case with Darren, you've got text messages and maybe transcripts from calls you've had with him and stuff like that. So any kind of like tips on how to create that knowledge base?

Yeah, it all comes down to your data pipelines. So where are you capturing these? There's, there's different nodes that you can capture data from throughout your day-to-day. And so let's say in my case, I have Darren. And one of the things when you're picking these people, If possible, picking somebody who you have either access to or you have a robust historical dataset, because at the end of the day, the people on your focus group are only as effective as the data that's behind them. If it's like wishy-washy data, then you're not going to really get precise responses, which is how you go from good to great. So for Darren, I have access to him. So do you have access to a person who maybe you have a good relationship with. So let's say you're trying to clone your manager. You have access to them. You probably have a ton of email communication with them.

You got calls that are recorded, transcribed, right?

I think that's a whole other set. But the big— the most valuable thing for me for call transcripts has nothing to do with the automatic to-do list items after the call, the follow-ups. It's everything to do with just capturing raw information because I don't know what I'm going to need. when I'm going to need it, and it gives me optionality in the future. And so if it's your manager, you have calls with them, you have the recordings, you have email back and forth, you have Slack messages. So you have this whole data set. So just think about where have you interacted with this person and how can you capture that information effectively. And there's a, there's a lot of quick wins to actually do that. And if it is on the social side of things, there's also things like comments, Engagement, direct messages. So it's actually quite simple. And this is a challenge that a lot of companies have fixed in a way. So it's all about identifying the pipeline, capturing the data, and then bringing it into your system.

Love it. Okay, let's get into how to set something like this up. Presumably, really what we're talking about here is we're either setting up the equivalent of an AI Darren, right, or AI boss, or we're setting up a board of different people, advisors that are people we respect, And I've heard people on the show also pick famous people like Seth Godin, for example, right? Or somebody who's written books and there's so much knowledge on them out there. You want them on your board because maybe they're super smart strategists or whatever. I would imagine you could also pull public data in this, in these kinds of situations, right?

We've chatted about the internal focus group. Those are like your customers. And then another thing that you can do is I like calling it like create your, your board of advisors. And so you can like ask the board. And so those are more people, your Hormozis, your Seth Goodens, like, and it's all the same concept. So everything that we're covering, it extends to that. For the sake of this, we'll focus on that internal focus group. So how do you actually do this? So there's a couple of ways. One of my core beliefs is it's bringing abstract to concrete. So how do you do that? Well, meet you where you currently are. So what tools are you currently using and work this into it? So if you're using just chat, Let's say you're using ChatGPT Chat or, or you're using Claude. They have a functionality called Claude Projects. And so that allows you to upload context manually. And so what you can do is let's say you have all of this data about Darren, your manager, you can create a project, upload it all there, have conversations within that specific project. And so that's, call it level 1, and that's where I started. And if that's where you're at, 80% good. Start there, bridge abstract to concrete, just get going on that.

Real quick, before we go to the next level, it's one thing to load data into a Claude project, but we also need to give it a set of instructions, right? So what kind of instructions are we giving it to act as this kind of advisor or whatever role that's going to be?

My favorite thing is skills. And so this is a concept that's throughout really every toolset. Claude, ChatGPT, all of them have their Their concept of a skill. The simplest way to put it is it's a reusable prompt. So instead of you having to type out a whole prompt every time, you create a skill that does the exact task that you're looking to do. And so what I I would do, and you can say something like, "I want to create a skill called internal focus group that runs an output by the audience profiles I've created." And so what that would do is like it would create a slash internal dash focus dash group skill, which. Will take an output and then run it by the audience or the recipients of that specific output. So that's like high level what a skill allows. Now that's just like very basic instructions. But one of the, one of my favorite things to do is I rarely will write any prompt, any skill. In fact, rarely, I will never write any skill or prompt from hand. I'll have Claude or ChatGPT interview me to identify exactly what I want to do. And so based on the conversation that we're having today, you could say, hey, interview me to create an internal focus group skill where I want to take an output, have an audience set, review my output, and provide me feedback. Ask me any questions to help create this skill and things that I might not be thinking of. And it'll go ask you questions and it'll help you tweak this skill. For the exact things that you want. So maybe you want it, the output of it to be feedback individualized in a chart, and it ranks, it scores it on a score of 0 to 10 across all the people. You can format it specifically in this skill. And the beauty of this is every time you run it, it'll create the same output or same structured output and do the same thing every time. And that allows you to create a foundation that can You can build on over time.

And I just want to double down, folks. I have stopped writing prompts as well, other than the initial prompt that I'm writing to get the skill kind of made. All you got to do is just try to explain in as much way as possible. And I like to speak verbally. For those that have heard the podcast before know that I use WhisperFlow, but Claude also has a pretty good AI voice thing built into it, so you can just talk to it. And when you ask it to ask you questions, Ideally one at a time, right? But maybe not one at a time. That's going to be the key because it's going to write a prompt better than any way you can write a prompt. And nowadays I'm like not even really concerned about looking at the meat of the prompt. I'm just testing the prompt to see if it gets it right. Are you finding you're doing the exact same thing?

Yeah, exactly. You run water through the, through the process to identify where the holes are and the best person or the best thing at improving a skill is AI. So you let AI do it. You just tell the features you want and it'll create it. You don't have to worry about the implementation.

Okay. So at this point, we've talked about Claude Projects is a great place to upload context. We've talked about Claude Skills, which technically live above projects. Where are we putting the information about all the different members? Because that part is still open in my brain.

Yeah. So you can bring it into the project. So if you wanted to create, call it a single file that said, Darren and manager, and that just has all of the information there. That is doable. What I like to do, and so Projects is good, the layer above that, I'll use Claude Code or Claude Cowork. And why I like doing this is because it has access to my local files. And so I'll, you can create a project, call it, I'll just call it Internal Focus Group is the project name. And within that, you can then have a folder that says, Darren and manager and each individual person. And what that allows for you to do is you can— let's say you have a call transcript, you can upload that to Claude and say, hey, I want you to ingest this or add this to my internal focus group. And I'll look at it, I'll be like, okay, I'm going to bring it into the manager dataset because that's where the manager is. And that's one of the simplest ways to do it. And there are more complicated structures. Where if you just ask any AI, tell me how to create an LLM knowledge base in this project, what that does is it'll structure the information in a raw folder and then a wiki folder. So the transcript with your boss, that would go into the raw. It's like the raw transcript. And then it has a wiki folder where AI processes that information, takes away any learnings. So when the next time you ask the internal focus group, it'll go to the wiki because that's already been processed and it makes the AI's job a lot easier. And then if it needs to, it'll go to the raw transcript to pull an exact quote if needed.

Okay, let me help people process this a little bit. I'm not using Claude Code, but I am using Claude Cowork. And basically what Austin is saying here is you can create a little hard drive folder on your computer and say it only has access to that. What I'm hearing you say is you can copy and paste the transcript into Claude proper, and you can say, store this file and analyze it, and it will automatically— do you have to tell it to create the wiki, or is it just doing that for you?

You'd have to structure the project. Let's say you want to build this system. I would start with, I want to make my system into an LLM knowledge base. Tell me how to do it. And it'll look at your system and it'll tell you, LLM knowledge base was popularized by Andrea Carpathi. He's a big AI influencer. I think he's with Claude now.

I think he was with Google or something like that, but you might be right.

Yeah. He's one of the GOATs. So anyway, it has training data on what that term is, and that'll give you the best direction specific to what your environment is.

Got it. Okay. But the good news here, the key thing here is that the one advantage that Claude Cowork has and/or Claude Code is that you can store all this stuff on your local computer, right? And because it's on your local computer, you can add documents in there and you You can figure out how to do all this later, but it can actually work with the files on your local computer, which is really, really cool. So at this point, what else do we need to know to set all this stuff up? We've talked about projects and we talked about how data can live in projects. We've also talked about cowork/code where the data is now living on your hard drive. We've also talked about skills, which kind of lives above everything. How do we kind of connect all these dots together now?

So one added benefit of also having the files on your machine is that you now own the entire system. It's on your system. And so if you want to change from ChatGPT to Claude to Grok, that's totally up to an open source model. That's totally up to you. You own all the context, you own all the information.

Or if one of them is down, right?

Or if one of them is down, you have optionality and you go from, I call it renting intelligence to really owning intelligence. And this becomes your intellectual property that you're building. So Anyway, that's, that's a, a side benefit of all that. But once you have this set up, and we touched on it in the open, but it very much is a process. So you're, you're gonna set it up, you're gonna run it by your internal focus group, and then the feedback might not be exactly what you would expect, right? It's like, I know Darren, I know my manager, that wouldn't be the feedback. And so what you have to do is you have to Create a way to iterate and improve the system over time. And so that can be as simple as within your Claude project, or I'm not certain in Project if you'd be able to do this, but let's say in Cowork because it has access to your files and you're like, okay, that's not exactly what the feedback is. My manager would have said this, not this. And I'll be like, okay, great. I changed the feedback.

Here's how it works in Project. You can tell it to upgrade its memory or you can tell it to update the skill, but you cannot tell it to update the system instructions. So if you have a skill for each of your people, then you could do it in projects, right? So if each person represented a skill inside of a project, you could say update the Darren skill. You know what I mean? That's how you would do it. Yeah.

Yeah. So whatever your system is, the concept is iterating. And so what you can say is, hey, based on this conversation, update my skill, update my knowledge base, update— you just say update my project, let it figure out how to update it. Stop using the brain. Let's update my project so that it doesn't make the same mistake again, or the next time it provides feedback, that this is applied. That's just one feedback cycle. The more you run it, the more you're going to harden this system and the better it's going to get. And so for me, with Darin, I would have an idea, I would have the feedback, and then I would send him a text, yo, what do you think about this title? And he'd be like, eh, I don't like that. For this reason and this reason. I take a screen—

It's after you already know what his AI avatar said, right?

Yeah, I know what the AI avatar said. I know what real Darren said. I take a screenshot. If it's the same thing, I'm like, bang, like we're there. If it's not the same thing, I'm like, okay, let's get it so it's the same thing. So I'll take a screenshot of that conversation, bring it in, and use that update prompt that I just mentioned. And I'll then update the data. I'll then run it again and be like, okay, now it has the exact feedback. After about 5 or 6 iterations, I stopped texting Darren like 4 months ago. He was like, yo, when are you going to text me? I'm like, I don't need to text you anymore. You've been outsourced. But, and so yeah, that iterative cycle is really just so important and it's not a magic wand. You're going to set it up. It's going to work good, but to get it to great, you want to bring it through these cycles. And step 1 was figuring out what, what's, what task is worth doing. If it's worth doing, it's worth doing right, and it's worth iterating and really improving it. And so that's where the juice is, that iterative cycle.

One of the questions that I have, and a lot of people might have, is there's a bazillion models just on Claude. You've got Fable, you've got Opus, you've got Haiku, and I'm probably missing Sonnet, you know, and different levels and stuff. For this kind of stuff, does it matter? And if it does, talk to us whether— because some of these are reasoning and some of them aren't, right? Like, what's your thoughts on this?

Yeah, this might be a little controversial, but I think decision fatigue is like the new cigarette. I think there's just so many decisions that you have in your day-to-day, especially in the AI world. Should I use this tool? Should I use this tool? Should I use this model? There's so many decisions. I try and simplify it and remove any decisions that I don't think are force multipliers or massive game changers, especially ones where the answer isn't the same day-to-day. Like one day you have access to Fable, the next day you don't. The next day you do. And it's, I find people really get like bogged down on a lot of these things. And so I'm of the opinion, if you don't have any token constraints and you find that you're never running into like token limits, just use the best available model. And if that's Fable, if that's Opus, like it doesn't really, really matter, at least in my eyes. What matters to me is removing that decision so that they can focus on the things that they own, which is improving the workflow, improving the training data, because the models will change, the models will get better. That information and that system that you have set up is on you to create. And that's where I want people focusing their energy.

Austin Marchese, this has been really, really incredible. If people want to follow you on the socials, I'm assuming it's YouTube, but where do you want to send them? And then if they're interested in possibly working with you in any kind of capacity, like you want to send them anywhere for that?

Yeah, yeah. So it's austin.marchese on socials, on YouTube, Instagram. Feel free to DM me. I love chatting with anybody. And yeah, if you're interested in kind of learning more, potentially working with me, I do actually have a custom plugin on buildpartner.ai that walks through like exactly how to set up systems like this. You like, it'll give you the exact prompts and stuff. So you can go check that out or you can DM me on Instagram and I'd love to collaborate, figure something out.

Austin, thanks again for sharing your insights with us today.

Yes, thank you for having me.

Hey, if you missed anything, we took all the notes for you over at socialmediaexaminer.com/a121. Be sure to follow this show on your favorite podcasting app. And if you've been a listener for a little while, we would love a review. Also, let your friends know about this show. You can tag me on Facebook, LinkedIn, Instagram, and/or X. And do check out my other show, The Social Media Marketing Podcast. This brings us to the end of the AI Explored podcast. I'm your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day and may AI help you become more successful. The AI Explored podcast is a production of Social Media Examiner. Before you go, if you're tired of figuring out AI all on your own and testing tools that don't seem to work the way you want them to work and constantly rewriting scripts and prompts, Because they don't sound like you. The AI Business Society was built for you. Every week we cut through all the noise. Every training is led by expert practitioners, and you're surrounded by marketers who are just like you. And right now you can join for $497, which is our special sale price, and your rate is locked in for life. And you got 30 days to ask for a refund if it's not for you. So what have you got to lose? Check it out at socialmediaexaminer.com. And I'll catch you on the next episode.