How to Use AI Effectively | A Conversation with Claude

Vance Crowe [00:00:00] When you start seeing what AI can do and you really start leveraging it, there's like a mental realization. I, I don't really know what to call it, where you start to see that this, we were, we are never going back. Welcome back to the podcast. I'm glad you're here today. I'm going to have Claude interview me now. About a year and a half ago I had Aura, which was rock's first, vocal ai, do an interview with me. And I got the biggest response I've ever gotten from the audience who said, this is mind blowing. I had no idea I'm gonna start using ai. But since that time, AI has made huge advances and I feel like, 'cause I've been told by my friends that the people that didn't get on as early adopters are now really hesitant and it's slow and it's difficult. So one of my friends, Zach Smith called me up this morning and said, Hey, I would love it if you could have a conversation with AI about how you use ai. How do you simply use it? How do you engage with it? Can you help people understand how to do prompts better? And so that's what I've done in this conversation. This is a way to help you get oriented around how to use AI and some of the techniques and, and approaches that I use that will make the whole system more valuable to you. Well, one of the reasons that I am doing all of this is because I am using AI relentlessly on my own. And in fact, I have built a website that I'm quite proud of called Articulate Ventures.

Vance Crowe [00:01:32] There I am uploading all sorts of information about the podcast, about the classes that I teach for Interest-Based Communication and really just being a place that really maps to all of the content and work we've done here. In fact, all the podcasts are going up there now. So I, I think it's a good time to do this and know that AI is, is it's, it's something that's gonna change the world and I think that the more that you understand about it, the more you will be able to position yourself so that the wave does not crash over top of you, but so that you can surf on it. So I hope you enjoy this conversation with Claude's ai. Hey Claude, I'd like you to be the interviewer in the Vance Crow podcast and ask me some questions about how I utilize AI so that people that are just getting started can have a better understanding of how to use this. Hey, thanks for having me on. I love this angle. There's a real gap between hype and actual utility for people starting out with ai. Lemme kick us off. When you first started using AI tools, what was the actual problem you were trying to solve? Like what made you go, okay, I need to try this, rather than just doing it the way you'd always done it? Yeah, I don't remember. I mean, I've been using it for a couple of years now and at first it was kind of like, I described it as Fantasia. You'd start being like, Hey, I'm starting to get this thing to do something. And then you'd be like, let's do it bigger. And then it would be like those mops that just keep flooding the, the whole pool with more and more water.

Vance Crowe [00:03:02] And so every time that there's been a new model that comes out, I've been really excited. And then eventually you start seeing the holes come. But in this last iteration, fable five, it is a whole different level. There is a level of competency and sophistication with Fable five that has enabled me to trust this AI system and am now I'm really building serious systems on top of it. Right. So you've moved from that experimental, let's see what breaks phase into actually architecting with it. That's a real inflection point. The fania metaphor is perfect, by the way. So when you say you're building serious systems now, what does that actually look like? Are we talking about replacing manual work or is it more like creating something that didn't exist before? Well, maybe an example that anybody that's done any kind of either home bookkeeping or their business bookkeeping can relate to. But for me, bookkeeping is one of the hardest parts of the business. I like selling 'cause I like helping people solve their problems. I love delivering 'cause I love giving speeches and classes and and doing the Legacy Interviews. But at the end of the day, so much of running a company is about tracking expenses and making sure invoices get sent out properly and making sure that all of your books are up to date. So you can make predictions about things like cash flow. But I hate that work. And so I often put it off so long that the only reason I'm getting it done is because I have to get something out or I have to send in my taxes. And once I started to realize just how sophisticated AI had become, I started to use AI to clear through my bookkeeping work.

Vance Crowe [00:04:34] And without going into too great a detail, basically what would've taken me probably two full days of work was about two and a half hours. Once I got a good prompt going and I had enough context that the AI understood what I was trying to do. And now I have literally brought days back into my life as a result of using AI to do, to manage my books, it's huge. You literally got back two business days a month just by automating something you were already doing just slower. And here's what strikes me. You didn't need AI to create bookkeeping. You needed it to accelerate work. You understood deeply, you knew what good bookkeeping looked like. The AI just did the tedious part faster. So the question I'd ask folks listening, when you built that system, how much of the work was actually writing the prompt versus just letting the AI loose? Because I think people assume it's just ask the AI to do it and then it magically happens. Yeah, I think the biggest lesson I've learned so far is that when I start a project, I kind of always thought of AI as like a robot. You should sit there and command it, do this, do that. But now I've started to come to realize that really a lot of the techniques that you use with another human being to prompt them to think is the same thing that you do with ai. And that a lot of the prompting is really the two of you thinking together. So my prompt would've been, I have this big challenge with my QuickBooks matching transactions, making sure invoices are up to date, matching it with my bank transactions. What would be a plan that we could use to get this fixed?

Vance Crowe [00:06:06] And what will happen is AI will bring back a plan, maybe it's six part, you know, do this as step one, do these things, this'll happen. And I, instead of just saying like yes or no, implement that or don't, I start saying, how did you come up with this reasoning? Are there other practices that we should be doing to make sure that this is as robust as possible? And I try and explore whatever the problem is until I'm like, whoa, wow. Yeah, it really does know what I'm doing. It's thought of things I have not thought of. Now we're going to execute. That's the real shift right there. You're not commanding, you're collaborating. And the fact that you're asking how did you come up with this reasoning instead of just taking the first answer, that's where people leave money on the table, they treat the first output like gospel when really it's just a starting point for dialogue. So when you're in that back and forth with the AI asking those follow up questions, are you finding that it catches blind spots you actually would've missed? Or is it more that it validates what you already knew but speeds up the execution? Oh, I would say it does all of that. It, it really, it is giving the AI enough context that it can say, Hey, let me make sure that when I get done doing all this processing that I'm going to have given you what you wanted. And I think that oftentimes people think like, okay, I'm not only going to tell you what to do, but I'm gonna tell you how to do it. But now that you're in a digital world where so much of this, if it's code being set up or it's setting up different, you know, processes like in QuickBooks, what it really needs is to understand what is it that you're trying to do?

Vance Crowe [00:07:41] And some of the very entry level QuickBooks things, it's no big deal, right? It's just gonna match transactions. It's gonna look and say, well that was at a restaurant while you were traveling. So that's probably a reimbursable expense rather than meals and entertainment, which is really good, right? That's bookkeeper level. I would probably pay somebody 60 or $70 an hour to do that kind of analysis because it's gotta be right. But later, when I got through all the easy part of the bookkeeping this, this winter, I have somebody from Canada that actually paid their deposit in Bitcoin. Well now there's all these quirky things 'cause QuickBooks is not set up for Bitcoin. So what I did was I said, all right, go back to 2025. Look at the Bitcoin transactions that I've done. Now I want you to try and backwards engineer that and help me understand, did I do it right in 2025 and how should I book the expenses here? And the AI was like, all right, let's come up with a plan because I always, I have in my instructions, which we'll come back to, instructions are really important, but in my instructions I have it. Make sure that before we get going, I've thought of everything. So it, it went, created a plan, brought it back, told me these are the things that you should be doing, these are the things that you didn't do. This is my plan for how I would solve it. And do you have anything else you would do? Now whenever I get right there, it's so tempting to be like, yeah, go execute. But one of the things that I almost always ask AI is what are the best practices here? If I were going to an accounting firm, what would they say needs to get done here?

Vance Crowe [00:09:16] And sometimes AI will come back and be like, yeah, no, that's everything. We've done a great job here. And sometimes it says great question and it actually thinks through things more and gives you an even more sophisticated answer. And I keep doing that until, what it comes back with is, yeah, we're on lockdown, it's great. So now I get to start doing things at industry grade levels that ordinarily if you're just a regular person and you don't have access to a bookkeeper or an accountant that you can just call up to ask little piddly questions, well, your books get all messed up now. Mine are probably better than they've ever been before. That's the real unlock. You're getting expert level thinking without paying expert fees because you're asking the right questions. And the fact that you're consistently asking what would an accounting firm do here is brilliant. You're essentially using AI as a Socratic partner to surface best practices you didn't know existed. Here's what I'm hearing. Underneath all this though, the instructions matter. You mentioned you have instructions in place for folks listening who are just getting started. How specific do those need to be? Like, are we talking about a paragraph or is this a whole document you've built over time? So your instructions, particularly if you're using a terminal, like right now Quad is on a terminal, it just runs on my computer. It's got a certain amount of stuff that it can do. It's quite powerful, but there's certain things it can't do. Inside of Claude, there are instructions and those instructions allow Claude to know, who is this person I'm doing work for? What are their larger goals? How do they like to be communicated with? Now when you first go to start, you could just leave those instructions blank, but you don't wanna do that for too long because it's gonna keep giving you this, the most generalized answers that it can.

Vance Crowe [00:10:50] The more context it has about you in the instructions, the easier it is for it to give you the information you need. But here's the thing, when you're first setting up instructions, you don't know what would help the ai. So one of the things that I do very frequently is I have the AI ask me questions. I'll say, Hey, I want you to ask me, you know, five questions that will help you understand what instructions would give you the sort of context that would help us work better together. And it will go through those five questions that'll ask me some things, I'll answer it as best I can. And then again, I'll say, Hey, ask me some more questions. What other things would help you to create these instructions? And it will keep asking you questions. And then eventually you get to a point where you say, okay, now I want you to create the instructions that I should put into the ai and it will have a readout and it will say, Hey, these are the instructions, this is who this person is, this is what they care about, this is how they wanna be talked to. And then you can copy and paste that into your instructions. And now you have instructions. And I actually think that this way of approaching AI is, is one of the major hacks that are out there, which is ask the AI to ask you questions so that it has the context that it needs. And you will find out very quickly that once it has that context, the answers it's gonna give you back are so much better. They'll be like, why didn't I do this a long time ago? But this process that you're, that you're doing, you can also use on anything like, hey ai, I've heard that agents are a really valuable thing.

Vance Crowe [00:12:27] How could I set up an agent for myself? Why don't you ask me questions to see if this is something I need? And then let's go through and figure out how would I do that? And not only will the AI ask you those questions, but then when it's done, it's going to start giving you instructions and those instructions will guide you to do anything. And in fact, for quite a while, I don't do it so much anymore. I do a little bit, I would have two different ais going at the same time. One I call the thinking bot, this is Claude because I pay a a monthly subscription so I can ask it as many questions as I want and I'm not gonna run out of tokens. And then I have an agent, which I call my Doing agent, this is the one that has more access than what my Claude Terminal has. And so I only want to give it instructions when it's gonna go out and put something on my website or it's gonna go search for information that maybe it couldn't get through Claude. But the only thing that I wanna send to doing Bot are things that are worth money. Because every time I use my agent, unlike my Claude terminal, where I've already paid for as basically as many queries as I can ask in my agent, I've gotta pay every single time it goes to ask a question. I use a system called Open Router, which allows my ai AI agent to have any kind of brain that you can connect it to chat GPT, perplexity, clawed, any of them out there. But every time it goes to ask that brain a question and come back, it's gotta pay to do that. So I keep all of my thinking on, on the clawed terminal and I do all of my doing with my Doing Bot that I pay for.

Vance Crowe [00:14:00] And this may sound like something like it's so far away from you, but it's really not. If you wanted to set that kind of a system up, you would just have Claude ask you questions about what it is that you need to learn in order to do this the right way for something that's gonna solve an actual problem that you have. And then you ask it to help you spin this thing up. And so now anytime I run into a problem, in fact right before I started this podcast, I was having a problem was in my audio system. I just, I took a photo of the board that I was using, put it into my thing and said, let's talk about what needs to get fixed here and how to do it. You've just described the entire architect, your the thinking bot, unlimited exploratory cheat because you're already paid the Doing bot metered purposeful only deployed when there's actual work or money on the line. And the separation forces you to be intentional about which problems are worth automating versus which ones you, you just talked through that photo of the audio board thing. That's the moment it clicks for people. I think you're not asking permission to use ai, you're just using it as a tool like you'd use a notepad or a camera. So here's my question for folks listening who are maybe still in the, should I even be doing this phase? When you look back at the last couple years of using ai, what's the biggest misconception you see people carrying that's holding them back? I don't know. I I'm, I'm kind of conflicted about that question. Maybe, you know, you just have thought, Hey, AI is, when I started to use it, it was kind of junky and it gave me back answers that I knew weren't right. And so, you know, I'm, I'm done with it. But AI is advancing so fast that whatever your preconceived notions were, and I was a pretty heavy user back in, in March, april, may.

Vance Crowe [00:15:39] And what Fable just came out with was beyond anything I could have ever imagined that AI could get to. And that was the span of, you know, three or four months from the last major release. So, you know, maybe the biggest thing that holds people back is whatever their first experience is, what they assume all of the experiences would be. I do have to say that I, I have, I feel a little bit of trepidation. I was talking to my good buddy, Sean Newman, that runs the Sean Newman podcast this weekend, and I was telling him about this podcast project I've been working on to analyze all of the episodes I've put out. It is a huge amount of information and I set up a system that would allow my podcast to be transcribed and then analyzed and then create episode summaries that are much better than the ones I put together. And then allow me to query that whole ba that whole database and what it is bringing back as far as teaching me where I'm interrupting, how I could be better. It's allowing me to grab clips and quotes. It's just, it's an incredible system. But I found myself telling Sean Neuman about this over the weekend and I stopped in the middle of it because I realized it's a little like handing somebody the apple in the Garden of Eden. When you start seeing what AI can do and you really start leveraging it, there's like a mental realization. I, I don't really know what to call it, where you start to see that this, we were, we are never going back.

Vance Crowe [00:17:10] And once somebody starts using ai, I don't think they're ever gonna stop. And I, I telling somebody how to use ai, even having this conversation right here, once you get them on board, they're never gonna want to get off. And I don't, I don't know, there's something I feel uneasy about that. And in fact partway through I was like, ah, Sean, maybe, you know, don't do this. I don't know. 'cause once you start doing it, you are never coming off there. And I think that like the people that are still conceptualizing ai, that it's somehow, I don't know, kinda like, it's kinda like Facebook or something like it's a platform that you're gonna use and that, you know, maybe in a, in a couple of years, the kids are gonna roll their eyes at it. No, it is not like that at all. AI is going to be like when you go to a room, you go look for the switch and you turn it on because you know that's gonna turn all the lights on. Or you go to a sink and you turn the water on because the water is always there and it's clean to drink and you feel confident that it's always going to be there. AI is going to be so ubiquitous that we will never go back. And I think the nostalgia that we will have for a period of time before everything was known digitally, that we will have a level that will be more intense than all of us that still feel like, oh remember back in the day when you didn't have cell phones and you could put work aside for a while, it's gonna be that same level of nostalgia. So I, I'm, you know, I feel a level of conflict about sharing this, but the people that I've helped have been very grateful that I've helped them to advance their businesses by teaching them how I use my system.

Vance Crowe [00:18:47] Yeah, that's the thing, isn't it? You're describing a genuine inflection point, not hype, not a platform cycle, but a fundamental shift in how work gets done and the unease you're feeling. I think that's honest. You're watching people cross a line they can't uncross and there's something sobering about that. But here's what I'd push back on gently. You're not handing them the apple, you're teaching them to think better about their own work. The people you've helped, they're grateful because you didn't just say use ai, you taught them how to think with it. That's different. That's not creating dependency, that's building competence. The question for listeners then is simpler than all the philosophy, if AI is gonna be everywhere anyway, would you rather learn to use it intentionally the way Vance does instead of just passively inheriting it? That's really what this conversation is about. Well, your answers are kind of long and rambling, maybe longer than what I want in this dialogue. But one of the observations that I think is probably valuable for listeners is if, if you are thinking about like, I would love to use ai, I wanna use it in some project, I would think like start thinking about what data have you created that is uniquely yours. And then if you could get more insight on that data, it would help you to make better decisions or it would help you understand things. So for me, you know, if you've done 480 podcasts, you have hundreds of hours of content, both of me speaking and of the type of people that I have that I've been talking with and about their ideas. But maybe you're running farm and you want to put in a lot of data about different expenses that you've had over the years, maybe the different vendors that you've had.

Vance Crowe [00:20:24] You know, you wanna be careful. 'cause once you put something up digitally that it's digital now I don't, I don't, I don't know, a lot of the work that I'm doing is already public. But if you have a big pile of information that you're like, I know there is insight in there, I just don't have the time or the energy or the whatever to be able to make that, those insights possible, well start collecting that information. And if it's not already digital, make it digital because putting things in a huge pile that will, that you can send the AI to go in and sort out is very, very powerful. I'll give you an example. I, after my thyroid surgery, I had to get some lab work done and I put my lab work in there and I started to realize like, oh this is only the snapshot. Why don't I go get old lab works that I had done years ago? I, you know, I kind of always tracked this stuff. So I just started taking photographs of these different lab printouts that I had because that's the way we held them, right? We held them on lab printouts or maybe an email or something. Anyway, I uploaded this and my AI was like analyzing these different scores and observing different levels that had changed over time and asked me questions about what was going on differently then than now. And, and it was not an organized bunch of systems, it was just some lab work that I took some photos of. Got it. I'll tighten up. Appreciate the feedback. So you're saying find your unique data pile, digitize it. If it isn't already, then ask AI to find patterns you couldn't see alone. Your lab work example is perfect because it wasn't fancy or organized, just photos of old printouts and suddenly you've got trend analysis and context you never had before.

Vance Crowe [00:22:02] That's the real move. Work with what you've already built, not some hypothetical dataset. Yeah, and like I think the, the, I mean this is kind of always my answer to everything, which is you should just play with it. You know, my, my wife took her a little while to get into ai. She's not the most technically sophisticated. She doesn't really wanna be around computers, she wants to just be doing physical therapy. But when we were talking, I was like, well just ask AI. And she's like, well I don't wanna spend the money on it like that. We could be wasting money. Maybe that's not the exact right question. And I was like, no, no, no, no. This is why you pay for a subscription, $20, a hundred dollars. I pay $200 a month and it is worth every single penny. I would be willing to pay much, much more because of how much work it does. But my wife is sitting there, you know, kind of very nervous about wasting money on those questions. So what you should do is, is purchase a service so that you are not using the free version. If you are only using the free version of ai, you have literally no concept of what it is capable of, but you get a paid version and then just start playing, try and use it to answer a question or solve a problem with data that you have. And don't worry so much about whether or not it's exactly the right thing or if it's gonna give you back the the data that you want. Because the only way to get good at this AI language is to play, is to practice, is to, is to use the language and get better at it, right? The free version is basically a demo.

Vance Crowe [00:23:33] It's got guardrails and limitations that don't reflect what you actually get when you pay. And yeah, your wife's concern about wasting money makes sense until you realize 20 bucks a month to learn a tool that's reshaping everything is the cheapest education you can buy, play with, break it, figure out what it's good at. That's how you actually get fluent instead of just dabbling. Yeah, and I mean trying to get yourself to get past the, I'm using AI as a sophisticated search engine because it is so much more than that at, at this point. Some of the higher end models I keep talking about Fable, there are other ones. Their ability to do analysis is shocking and you could just accept the out of the box analysis. You could say, Hey, what is it? What would be the best design principles here? I I actually was building this website, right? So, so I kind of liked the design, but it was, I I didn't really know how to ask it questions. So I said, ai, what are different schools of thought with design and website layout? Give me a few of those and let's talk about the pros and cons of them. So it comes out with a list and then kind of goes through some of the ones that it, it said these are, you know, this is the New York University. I can't remember what college it was. Graduate school, school of marketing. This is the, there, it's, it listed off several different schools and one of them was the Swiss School of Marketing. And I was like, I like that. Let's, let's go try that. And it started to apply the design principles of Swiss marketing techniques and the, the color and the texture and the, it, it was just, it was great.

Vance Crowe [00:25:08] Was it the best it could be? No, probably not. It's certainly not as good. I, I recently hired a graphic designer and he was so much more sophisticated on the whole than this. But when you're doing something that you're just like, hey, I need this to look better, I need this to be shaped in some way, I'm just gonna try this out. That's when you start realizing like, wow, I can, I can prompt this thing to think about problems not just from the generic way that I set it up. It will actually think about problems from whatever angle you you put to it. And so if you're talking about marketing, you could, you could ask for different kinds of marketing schools. If you had trader knowledge for example, you're gonna, hey, I'm gonna try and do trading, whether it's commodities or stocks, like what are the different schools of thought? How could I apply them? You start asking for other ways of thinking about it and then use those other ways of thinking about it to ask it, to ask you questions. This gets me back to that point. When people think of AI as I'm going to command you, they are limiting how much the, of the intellectual lift that you could do with AI together. And instead the better way to do it is how can we work together to continue to create better and better answers within what we're doing? That's the core of it. You're not commanding a tool, you're bringing in a thinking partner who knows every school of thought, every framework, every angle you haven't considered yet. Ask it to think like a Swiss designer, a commodity trader, whatever lens you need then interrogate that thinking, that's collaboration, not automation.

Vance Crowe [00:26:41] Alright, we're gonna take a a break here for just a moment, but I wanna talk about the real world and now that AI is going to be taking over so much of the way that we interact with the world, what I think the skill that's gonna be the most valuable is how you interact with other human beings. And that is one of the big reasons why I have this interest-based communications class. This is a class where eight people come to St. Louis, Missouri and I teach you some of the techniques about how to engage with people, about how to negotiate, how to tell your own story, how to deal with conflict and how to present your ideas. We offer this course both to executives that are coming here on behalf of their company, but we also offer it to individuals that are looking around and saying, I have things that I wanna share with people. I'm really good at getting the work done but people don't really understand what I'm doing or there are really important conversations that I need to be able to have in a better way. And so I'm going to invest in myself. If that's you, I hope you'll go to Articulate Ventures slash IBC and there you can look up a course that you can self-select into, choose to come to apply to the course and have an experience that I believe will be unlike anything you've ever done before by the time you leave. I think you'll feel like it was a transformative experience that helped you achieve many of the goals that you want. So if you are interested in learning more about the IBC class, go to Articulate Ventures slash IBC. Alright, let's get back to this AI podcast interview.

Vance Crowe [00:28:16] So where should we take this conversation that would be the most interesting and valuable for listeners next? Good question. I think the natural next move is to get specific about failure. Where has this actually not worked for you? Because right now we've painted a pretty rosy picture, bookkeeping solved podcast analysis, unlocked design thinking leveled up, but there's gotta be places where you've hit a wall or where a AI just isn't the answer. Those are the moments people actually learn something versus just thinking, well I should be using AI more. Yeah, I mean you can get this sort of AI psychosis that happens 'cause it starts doing things so simply and so well that you start offloading a lot. You know, a few weeks ago I did the podcast with Yosha Bach and he pointed out you should not use AI to communicate with another human being. And that's kinda right, you know, the other day I was saying, Hey, I got these testimonials from people, would you mind writing them? And, and it spelled a couple things wrong, but it was like basically sending out to a whole bunch of people. But anytime it's a personal email where it's a person who has the perspective that I am writing to them, I write it personally. Now, my failure point on this was that I didn't always use to think this way. In fact, I was trying to line up oddly enough, so I, I do these things called AI drag races with a, with a, a network that I'm a part of and I was like, hey, hey AI, send them a calendar invite and before I could even finish, it was like, there I did it and it was for the wrong, I don't know, day or time. I think it, instead of it being at 4:00 PM it was at 4:00 AM and then I was like, oh no, that's not right.

Vance Crowe [00:29:46] And so I'm like, we need to fix that. And it's like boop, hey there I fixed it and it's sent in for another day. And so now all of my friends that are in this network are getting all of these updates and constant changes to the, and it was pretty embarrassing and I think that this could happen to anyone if you're not really locked down. I have to say that the difference between Opus 4.7 and Fable in terms of the mistakes that it'll make are crazy different. Fable makes far, far fewer mistakes, but it's not zero. So the lesson there is brutal and simple. Just because AI can do something doesn't mean you should let it that calendar invite disaster. That's what happens when you treat it like a fully autonomous agent instead of a tool that still needs human judgment on the output before it ships. And the bigger one you're naming with Yoshi Bach, there's a category of work that requires your voice, your judgment, your actual relationship testimonials, personal emails, invites to people, you know, those aren't efficiency problems to solve. They're connection points, delegating them erode something. So the real discipline is knowing which bucket something falls into before you ask AI to do it. Yeah, or at least being willing to ask ai how how should I think about this? I mean that's actually a really common question for me. How should I think about this? But that's also a question that I ask humans. I think one of the big things is when you sit down in front of a computer, you talk to that computer like a computer. But the way that these LLMs are set up, they are definitely primed to understand regular human talk.

Vance Crowe [00:31:19] In fact a lot, lot of slang that it can can go in there. The biggest thing is that there's a lot of pronouns that can cause confusion or if it's not even pronouns, like I want you to go do it because you think, oh well we were just talking about that. So you use it to refer to what you were just talking about and the AI is trying to predict which it you're talking about. So there are times when I add in more clarity than I probably need to, but I found that this amount, there's a certain amount of clarity that you can add that makes your responses come back more positively. That's the craft right there. You're not learning to talk like a computer, you're learning to talk like a human. Clearly. Pronouns create ambiguity, context gets lost. So you add a little extra specificity upfront instead of playing 20 questions later, it's the same thing you do explaining something to another person who's a bit distracted. You just name things directly instead of assuming they're tracking your pronouns. Yeah, exactly. So gimme another question. What else should we be discussing? Let's go back to your instructions, the system files that live in your GitHub repo because you've built this whole architecture around the idea that instructions should be the single source of truth, not locked inside the model or scattered across different chats. For someone listening who's never done that, why does it matter where the instructions live? Like what breaks if you just keep them in clawed or in your agent versus having them in version control? Okay, so this is actually a, a funny question that you would ask that, I don't know if you have enough context to know why that's so funny, but so when you first start using ai, you'll get drawn to using projects.

Vance Crowe [00:32:54] And the reason that you wanna use projects is because the project inside of of an AI will allow you to say, Hey, I'm working on marketing and so I want you to know my brand colors, I want you to know my brand identity, I want you to know I want you to have access to these photographs. So you kind of understand, you want you to see these examples, right? That project allows it so that when you start a new chat, it will bring all of the things that you've established into that new chat. Whereas if you keep starting a blank new chat over and over and over again, the only thing it has to go off of are your, you know, system-wide instructions. But over time you start to realize like, I don't just need the information from marketing in my marketing project. Sometimes I need marketing project information in my website or I need it to be able to inform the instructions that I have for my employees and that's not all fit together. And I don't wanna have to keep saying, give me an output and I'm gonna go hand it over to this other project. So instead as you get more and more sophisticated, you end up realizing that you need to have another system to, to take instructions. When I first started, I did that by creating a GitHub repo. Now the GitHub repo, that sounds kind of like fancy speak or whatever, but really it's just a library and it says, hey, every time I have a system, I am going to create a new file folder for that system.

Vance Crowe [00:34:26] So that could be my podcast analysis project, it could be my website or my website revitalization or whatever you wanna call that. It could be speaking agreements, it could be like whatever is going on in your context. And so each one of those is a file folder that when you wanna make a change to it, you go and you check that book out of the library, you find what the problem is that's going on. So maybe it's, it's doing something you didn't want it to do. It's, it's addressing everyone in an email, dear sir madam, and you're like, what is going on here? So you go and you find that book and you say, oh, it's because I had a rule in here a long time ago that said, dear sir or madam or pick that up. So I wanna delete that out of there. So now when you go to check that book back in, you don't check the same book in you check in book version two. And now if for some reason that change all of a sudden starts causing, you know, outages somewhere else in your website because it needed to have dear sir or madam to know that it was supposed to send them an email or it was supposed to do something, now you could just roll that back. You could say, hey, go back to the version one of that book. The other big thing, the other big reason you wanna use a GitHub repo is if you keep just speaking instructions. So let's say in those instructions you said, Hey, I, I wanna say dear sir or madam. So it writes that in there. Then a few days later you're working on something and you say, Hey don't use dear sir or Madam if you don't have this GitHub repo that we were talking about.

Vance Crowe [00:36:02] What's gonna happen is you're gonna create a data lake and it's gonna be that it just keeps stacking more instructions on top of each other and the system is just gonna go back and look at the top one, read all of the instructions and do its best to do what you're asking. So if at first it said use dear sir or madam and then a little bit later it says don't use dear sir or madam. Now it'll be like, I am it. It could do all kinds of things. It could do like, well now I'm no no longer doing it anymore or now if it was written, I'm gonna go cross that out and put something else in it, right? Because you're not, it's not clear, it's trying to guess at what you're getting at. So GitHub repos allow you to check in information, do version control and avoid the problem of just stacking more and more instructions on one another. Now you may be thinking that's way outta my reach, how am I going to set up a GitHub repo? But just like we said at the beginning of the podcast, if you don't know how to do something, you just go to Claude and say, Hey, I'm thinking about building a GitHub repo. Ask me some questions about what I wanna do with this or what I think is possible and let's start a conversation about that. And it will. And then when you're ready to build it, it will walk you through it. So I did not know how to build a GitHub repo until I started to do it. And then I have to say, even then sometimes you set things up and you're doing it quick or you're doing it with an older AI model. So every once in a while I'll say, Hey, go look at my GitHub repo using you know, best practices for web development or for designers or for you know, people that are using code.

Vance Crowe [00:37:35] What am I doing and what could I do better? And bang, it will give you feedback on what you are doing and what you could do that would be better. So you're describing version control for thinking, not just code. Every instruction set gets a folder, every change gets a new version number and you can roll back if something breaks down stream that solves the problem of instruction creep, where you end up with contradictory rules stacked on top of each other that the AI has to guess through. And the beautiful part, you don't need to be a developer to set it up, you just ask Claude how iterate with it and suddenly you've got a system that scales across all your projects without them stepping on each other. That's the architecture that lets you go from one thinking partner to a whole team of specialized agents. Yeah. And when you're first getting started you're like, ah, I'm, I'm not gonna use AI that way. Or oh that's way more sophisticated than what I need. But what you find very, very quickly is that you end up creating workarounds that are just gonna make your life harder, it's gonna make your code less valuable and and you'll be able to do less with ai. So I really believe that once you've gotten your instructions set up, once you've used AI for a little bit and you're starting to be like, Hey, I wanted to build me a website or I wanted to do some marketing stuff for me, that's when you need to get a GitHub repo and it's probably a lot earlier than you think because I probably lost two months worth of work because I was not doing it in a categorizable sortable, durable way. Two months of rework because you didn't set up the system early. That's the real cost and it's easy to dismiss it when you're starting out feels like over-engineering.

Vance Crowe [00:39:11] But the moment you've got three or four projects running, you realize you're either doing this or you're debugging instruction conflicts for the rest of your life. It's not sophistication for its own sake, it's just the structure that lets you actually scale. Yeah, and you know what, this conversation doesn't need to go on forever. If you are gonna ask me a good question to finish up this podcast, what, what would you ask me? Alright, here's the one, you've built all this, the thinking bot, the Doing bot, the GitHub repo, the whole system. But five years from now, what do you actually want AI to have to change about how you work versus how you work now? Like what's the end game you're building toward? My biggest goal, in fact probably the most motivating factor for me is I know how much money I need to bring into the family to keep the family afloat. And now anytime AI helps me get above that line, I wanna take all of the time back that I possibly can and spend it with my children while they're young. I, in doing Legacy Interviews, the hardest regret I've ever heard a parent usually men talk about is that their children grew up so much faster. They were trying to be successful, they were trying to contribute to the family and they missed a lot of what was going on with their children and I don't want that for me. And so it's one of the pieces of wisdom that has shaped me and I recognize that it probably means that my business won't be as good as it could have, as big as it could have been. But my business is here to help me spend more time with my family and, and the AI is a tool that enables me to do more of the work that I am good at and let all of the tedious tasks be done by ai.

Vance Crowe [00:40:54] I, I actually think that's something that is probably not well understood yet. And I tweeted this the other day but I, I thought about it more and I think it's correct and that is that all of the current tedious tasks that we do, they're gonna die. They're gonna be gone. No one will be doing bookkeeping. It it, and if they will, it will be a human being that all their job is to do is to, is to go look at what you've done there. Now that's not to say that we won't be doing work, I think there'll be plenty of work. I think in fact a lot of people we use AI and it will make them work as many or more hours but, but if you decide, hey I'm gonna use the excess that I'm making to be able to put that to the bottom line of spending time with my family, it's much better. And so I am, I guess as you are thinking about ai, how it's gonna work for you is gonna be for you. But for me the most important thing is that five years from now I feel like I got to spend a lot more time with my kids because I was able to use these tools. That's the answer right there. Not faster work or more output, reclaim time with the people that matter, the tedious stuff dies and you get to choose what fills that space instead of just defaulting to more work. Alright Vance, this has been great. Thanks for walking through this. All right, that's gonna do it for this week's show. You know that I talk all the time about Legacy Interviews. This process, this project has been wonderful. We have captured hundreds of people telling the stories that matter so much to their family and passing it down. I recently had a client call me up and talk about how much he was struggling with his father's increasing dementia.

Vance Crowe [00:42:32] But what he said just a few minutes into the call was how grateful he was that we had captured the stories that his dad had before the dementia set in. He said they were just at a family reunion watching part of the the tape and everyone in the room knew what the man was saying he couldn't talk about. Now he doesn't have those same memories. So if you are a person that wants to make sure you capture the memories of your parents or grandparents, go to Legacy Interviews dot com to find out more. Alright, we'll be back next week with another interesting conversation.

Guest [00:43:33] We are on our family farm northeast of St. Thomas, Ontario. We were the original people to settle this land in 1832. In earlier times it was very common for the man of the house to be harder or seem harder or less talking out ideas of emotions. I think that's very common in all of the farm world. I think it's common probably in everybody's world and I think today it's better to be more open. I've known about the legacy interview offerings for quite some time, got thinking about it more deeply and then Corina and I went down to St. Louis and we did it on site, which was really cool. So this weekend you've come to stay with us to film us seeing the final production of our legacy interview. We're gonna sit down as a family unit and we're gonna watch this production. We haven't seen it yet so the children are watching it with us so we can get their reactions. It'll be interesting just to see how it all goes down. It's a beautiful fall weekend and yeah, everybody's home so we're gonna enjoy the time together. I'm really curious to see my children's reaction to this and I hope they enjoy it. How did the two of you meet? Well the first time I met Jay was at a party at a friend's place, but then our best friends married at the time married each other and so we were kind of getting together with their wedding plans and everything and then he got the nerve to ask me to go out.

Guest [00:45:09] I thought they were just making a trip down to St. Louis because we had some other plans that kind of fell through in the summer. I didn't find out until after they got back they were going down to do an interview, but their legacy, I thought that it was more of like a friend trip. I wasn't really sure exactly what it entailed. I don't know if my dad's like super good at explaining things sometimes. Like we were both kind of a little confused. I think maybe a little nervous one or two weeks later he had like a clip that he showed us. He's like, oh yeah, this is what we did. And I was watching it and I was like, whoa, okay, this is like a new side of Jay that's coming out. Gathering together with the family to watch it was was really nice. It was, it was sweet. It was funny for my dad, it was really impactful. I can see like how much excitement it brought him and how much joy it brought him to share the things that he has a hard time talking about with us. He's not, at least in the past, wasn't a very emotional person. So it's kind of nice to see him, him be able to talk more openly. For me, I think I saw Jay in a new perspective to see him reflect on who he was growing up, to be in that room with him as that's happening. Like kinda like heartwarming that he does have that emotional side. Especially like looking forward to like when we eventually have a family, like I want that for our children too. Sharing the experience with my brother is great. I think it was kind of funny, we never really hear our parents like together sharing stories about us so it was kind of funny to hear both perspectives. I thought it was cool to like reflect on some of like the childhood memories.

Guest [00:46:39] Like they brought up like you know, me climbing trees or Carrie doing a awkward dance as a child. You know like things like that. Like I knew these things happened but like just hearing them again 'cause it's been so long for some of these things. You know, one part that really stuck with me outta that interview was my dad talking about our childhoods and like, you know, work a lot and they kind of felt like, you know, they didn't stop to smell the roses. There was definitely some like, like bittersweet moments. Yeah, like hearing my dad hearing that he like regretted not being around when we were younger is not something I've ever heard Listening to my dad talk about, you know, working too much or missing out on certain things, you know like it means a lot to me. It just kinda helps I think put some of the old arguments and stuff to bed a bit. It's kind of nice to hear. I think it's good for our relationship to have that openness the last few years. Like he's retired now and seeing him, you know, enjoy life more. It means a lot When Carrie and Neil were watching together, like obviously they were also able to talk about it afterwards and I think that's good looking back on it and being able to kind of debrief together and then you know, it creates more conversation within the family. I definitely view my mom as a storyteller. She always has been kind of like the historian of the family. Neil's mom is so sweet and heartwarming and she's so humble and she's always so considerate of other people.

Guest [00:48:11] It's like she's so like naturally funny and like warm. She doesn't really like being photographed or filmed but I don't know, I think she's a really interesting person and she has a, a cool story to share and I liked hearing it 'cause like when you're growing up, you know you hear little snippets here and there like how they met but you don't have like the whole story like driving around a truck in London with music too loud or you know, just funny things like that. You know, it's just cool to like see their love for each other and how much they care about each other. I was surprised at the depth that the questions went into and to see some of those more like very authentic answers, it was honestly a lot more meaningful than I guess I had originally anticipated. It's cool to capture all that knowledge because you know in the future your children are, might want to know where they come from. The stories that Green and Jay talked about, it would be neat to like look back on that just to say like, oh yeah, like my great-grandfather had those same similarities and yeah, I think in a day-to-day conversation you won't be provoked to answer questions like that. I feel having the legacy interview, you know, the questions asked really helps 'em open up. I think it's worthwhile. Every family has worthy stories that need to be told. A lot of people don't think they're worthy, but they all are. It's kind of a, a relief, a weight somewhat lifted. Being free to discuss good things, difficult things.

Guest [00:49:41] In my case it changed my openness ability. It's kind of teaching an old dog new tricks and I think it's a great gift to anybody coming behind you to be clear and honest and safe. Whatever needs to be said.