Why Your AI Is Acting Like A Junior Developer

Show Rebrand And Review Goal

0:02This is the unknown secrets of internet marketing. Your insider guide to the strategies top marketers use to crush the competition. Ready to unlock your business full potential, let's get started.

0:17Howdy, welcome back to another funfold episode of the Unknown Secrets of Internet Marketing, the best SEO podcast, which we've rebranded. And if you've been looking for our O logo with my face, well, we've changed it. And my face is a lot smaller. And we're bringing on some of the top SEOs in the space. We're really uh building a community resource. Uh, we're talking about AI, we're talking about GEO, we're talking about AEO, we're talking about AI SEO, we're talking about everything. And one of the things that I've gotten a lot of feedback on, so I do appreciate all the comments and I do appreciate the reviews. Right now, we're doing a review push to get over 4.5. So if you've been a longtime listener, you've gotten value out of this podcast. My call to action would be please leave a review. We need like something like 50-something reviews. We have like 316 to get over 4.5. So please leave an honest review, uh, how we're doing.

Why Marketers Need AI Architecture

1:10But today is a special episode. Okay. So I have two fantastic guests for you. We're going to be talking and going in the weeds about AI. So I know we talk about SEO, we talk about the marketing, but the architecture behind that and how people are utilizing AI and programmatic to really get some leverage on what they're doing. We're going to go deep today. So this might be the podcast for you. This might not be, and you might want to skip it. But I would think everything's going this way. So pay attention. These are people that are not just talking about AI, they are using it day in, day out to do marketing, to do all kinds of fantastic things. They're the real deal. Richard Valentine, who's been on previously, and then Hector Herrera, who I've been working with a lot over the last year. And uh, guys, gentlemen, thank you so much for coming on the show.

2:08No, thanks for having us. Yeah, definitely. Thank you.

2:11And and also, everyone, this is a late night episode. So um, a lot of times I have a lot of great discussions uh after the sun goes down, and this is one of them. So it it may be a little bit different flavor. We're gonna be having some real discussions. So tell me how how you feel about it. Leave us a comment. Thank you so much.

Claude Vs ChatGPT And Model Drift

2:32So, so guys, um there's so much happening. Um, Richard, I know you and I are exchanging memes uh the last couple days, and Astral from has come out. Facebook has just come out with some new stuff, and it looks like they sunset manis. Um, like there's a lot happening in the space. Like, how are you processing? You were even telling me that you're not really using Claude code as much anymore. And you know, I I want to just kind of hear what what you're thinking about what's going on. Maybe you can kick it off for us.

3:09Yeah, so um, first off, thanks for for having me again. I'm I'm super excited about this episode with uh with the both of you. Yeah, I I really um I got tired of Claude always. So, you know, I do I utilize many different AIs. So like uh I use Gemini AI and the Google AI when I'm trying to find like some bleeding edge stuff or or things that are very current, even though I utilize connectors like uh Exassearch or or some other plugins, um, I just find that it does well. But when when I put that research or something into Claude, I find it nitpicking. Hey, I'm gonna pick this one point that you didn't do right or that isn't right. And that's what we're gonna talk about for the next seven messages. And um, I don't know if anybody else is feeling this out there, but the usage limits are creeping in. So, you know, even having a a max plan on on both Chat GPT and uh Claude, the way I put it to a colleague the other day is um I find that that Astra and Chat GPT has a much better way of going, hey, this is wrong, but did you mean this? Or here is something that works, like what you're saying. Whereas Claude's like, let's have a 20-message discussion about this is wrong and why I'm nitpicking this and why this other AI agent research that it recognizes as AI agent research um is wrong. And so when my you know constant conversation is is nitpicking those things, um, because I'm very much in an ideas and bigger picture guy, um, I just find that I'm wasting a lot of time and credits. Uh, one last thing before I kind of let everybody else comment is uh, especially in Claude Code, I mean, it'll run 15 agents and use a million tokens on each agent when I'm just like, hey, can you look up the latest in this, you know, which which used to be a great thing. I love the thoroughness, right? And now it's just gone to this overboard, like, okay, we're gonna get all this fluff and and all these things that that don't really need to be ran and and used. And and when you're using the most highest plan and you have to buy usage after that, um, it used to be that sometimes I would I would get close to the usage right around reset time. Um, I think I'm out of usage right now on almost both, and uh, and it doesn't reset till the 14th. So okay.

Token Limits And Managing Subagents

5:42So what you were talking about about like the subagents, like heavy like session usage, like using Fable to do some of the logic and the the big picture thinking, then it just runs the agents on Fable to do the research and it burns up all these tokens. So I had to set up a rule to basically say, no, use Sonnet every time you do research. And also, like, if I'm like cleaning up emails or something like that, I'm like, no, you have to use Sonnet to do this. Like it's called a hook. And and I I like I had to really look at token management, and it was so great when they like relaunched whatever, like it was a week or two ago, and my tokens reset because I was almost like it was like Wednesday and I was almost to limit, and I was like, whoa, they they reset it. And and Hector, I I want to ask you this because I feel like and like uh Claude is starting to act like a junior developer because even if you talk to it about something, it starts to um like grab onto something, like like Richard said, it's grabbing onto something and it starts pulling on that thread, or it remembers things from previous in the memory, and it won't let it go. And like I I was working on a lead gen site the other day, and it all of a sudden decided we were talking about something about content, it all of the decided said, Hey, there's two phone numbers, right? So we have one 800 number and a local phone number. It just unilaterally decided that there was inconsistency in the phone numbers, and it just pulled two of them off the website. Like, and I was like, Whoa, like you gotta watch what it's doing. But I'm like, this was not even something we were talking about, and like I didn't have to watch it as much, but now it's starting to go off the reservation and start to make decisions based on the limited facts that it has, and then you like Richard said, you gotta argue with it to to to to right size it. And I have found myself using Chat GBT a lot more because I can just ask it a question and it'll give me the answer that I'm looking for. And so I was curious when we're talking about maybe drift or you know, kind of uh the information set that that you're managing, how are you preventing it from doing it? Because you have a whole team of different agents that are are working on your stuff. I and I would love to hear you kind of share with everybody your architecture setup there because I I think you've you've addressed that in some ways.

8:18Yeah, I have noticed, you know, where it feels a little bit more junior than it used to, and in a very strange way, where it's just keeps prompting you for something like you're saying that that really wasn't there, or you're not really questioning right now. But I feel like it's it's trying to remember something from the past and pull it up to your attention. And I'm like, yeah, I don't want to work on that. I have noticed that lately. Um, but I think what I've done, it's funny, I was hearing you talk about token limit and token usage, you know, because I think that is an issue when it starts to prompt you so much about the same thing. You're like, hey, I don't want to work about that, work on that right now. Um but I I haven't run into any issues with token usage with the way I have my setup. You know, that's been the strangest thing because I use cloud code primarily, but I use it non-stop all day long to do all kinds of things for me.

9:18I can vouch for that, guys. Yeah, I can vouch for that.

9:21I mean, I got multiple um multiple terminals open, and I mostly do it through the terminal. So I don't know if it's something within if you're using the cloud code app itself, that it burns through tokens. Because when I first started doing the cloud code app, I was running through tokens like like crazy. I I had to get the max plan, and even with that, I was I was hitting those limits, like you're talking about. And then when I moved it over to the server and started running things through the terminal, and I started to build out my teams, I really I really don't hit my token limits at all anymore. I mean, I I haven't even checked it, but sometimes I don't even think I get halfway there.

10:03And and so one thing that because I just recently, after having this, I I too used to be like you where Matt would say, Man, I've burned through this, and I'm like, wait, you know, I don't, and then I I just thought, hey, Matt's working on 10 times more things than me. You know, I just kind of threw it up to that. Um, but but this has been a recent occurrence, and so what I've what I've I've started implementing and drifting towards is so I have a Kimi K, I have a Kimmy plan. I have a I have my my chat GBT, which is on the main max plan, like the the 200, and then I've diverted Claude down now to the $100 max plan. And so what I wanted to run was the I I just haven't been able to sit down and really do it. But after this week of hitting to the limit that quickly, I know that this is priority, is setting up uh the Orca IDE. So that's uh for those of you that are used to like VS Code and things like that, it is an agent development environment um that can that's kind of agnostic, almost like an open code, uh, but they already have a native web app uh or uh mobile app and things like that. And so what I wanted to do was I have this home desktop that uh that it just sits there. So I'm like, hey, I can because that was another thing I was running into, is I'm very much used to my uh I have a you know workstation laptop that you know is is pretty high powered. And I was constantly running out of my terabyte of of space, and I just got tired of that. So I was like, hey, look, I'm gonna back all this up to Backblaze. Shout out to them. I I actually really like that. It's uh anyways, it is just a good uh I I don't like OneDrive, but I digress. Um so reset the web the the laptop and then put all of that onto the desktop. So like my WSL, my just kind of all my terminals, all the uh different files I have, and then making the the laptop just a light kind of intermediary and you know workstation from there using Orca so that way I can utilize those different plans. Even uh the the desktop has some some good RAM and as we see these models keep getting easier and easier to run locally for some smaller things. Um, but setting up a situation like that where um like uh using Buzz, I don't know if you guys have heard of that. It's the the new tool that's like Slack. It's developed by uh uh Dorsey, the guy that made Twitter, um, but but it's specifically for agents. So I set up a tail scale network between those, and I actually have two hosting or small VMs that are like $20 that I put on uh uh so one of them I use is the relay and the tool layer. So I've made all my MCP servers and things like that on that hosting or server, right? Then on the other server, I have five different um uh wow, why can't I remember the name of this agent? Um not not open claw, but what's it called? Are you using Hermes? You were using Hermes, yeah. So five different Hermes agents and then giving them specific skills and plugins so that way, because I realized that part of my problem was I had 1100 skills loaded in my that's that's what's burning up like a lot of the context window is if you have all these skills they're preloaded. And I didn't I didn't realize that, yeah, even if you're allowing it to just call upon them when needed, even just having them loads like a small bit, but when you have 1100 of them, so by doing that and kind of separating concerns, um, that's the architecture that I'm moving into. But I'm interested to hear so so you're just using the the terminal, and then are you are you using like the um orchestration, like you know, manager, sub agent where were you using different models or because I realized that part of my problem too was I was using like ultra mode on everything, you know, like I wanted to use, and so I I feel like that was my problem because we weren't having these same usage limits decreases, so I just never ran into it. So I was like, why not use the best all the time?

14:35Yeah, you were you're like use swarm on everything, and I was like, oh man, you know, um, so one of the things Hector and and kind of to set the table for everybody, there's a lot of different setups. Um, I can tell you I have a desktop that I do everything uh like on on the desktop, and then I'll upload it to like a GitHub uh where I can be able to share and work with the team. I also spin up a Linux server on a laptop, and then I log in and grab everything from GitHub to download it to and then do everything on the device locally using a CLI. Like there, there's so many different ways to set this up. And Ricky, one of the things that you know it's funny as as sprawling as like your setup went, we started to have discussions about like how like NADN and just automation and really simplistic uh nodes and making kind of decisions on that node are structured. And Hector has impressed me with the simplicity where he built a team, they they don't um they don't self-learn, they know what they know, and then he kind of has them consult, and then he actually, because he's working in a team environment, he's added personalities to different agents. And he was actually telling me earlier today he had to create different personalities to interact with different people, and then he has another um uh like orchestrated bot that talks to those bots, then talks to him. Hector, maybe you can kind of explain what you will out of uh out of your setup because I think it's quite impressive and it has really, from what I've seen, really limited drift on uh making sure that these conversations uh don't don't start to pollute the other conversations and then it gets all twisted and they know stuff that they don't need to know. You've kept it really clean so so the token you uses limit uh seems to stay down. And and I think it would benefit from people to to you to kind of map out in your head how your brain works and how to think about it. Because guys, everything is your it's custom to your setup and how you work, like building your own gates, building your own recalls, um, bringing in your your own MCPs and your own APIs, like it's all custom to you and how you work and building your own personal S, really.

Building An AI Team System

17:13Yeah, I think it's probably the closest thing I have to like a clone of my own brain, is what I feel like, you know? And it all started just because I had to build a website, and I was like, okay, let me try to do with AI because there was somebody in the office who was just like showing us what AI can do, and he was spitting out websites left and right, and I was like, okay, I'm gonna mess around with this, and so it kind of started experimental, and uh, I think I started with GPT, and I I saw they had project folders, things very simple like that, you know. And I I really I I try to say what what does it need as far as it goes for an art in an architecture standpoint? You know, you talk about it it needing to learn or it needing memory and it having problems like that. But what I did is I had a very specific task to build a website, and I said, what teams do I need? I need like a content team. Can I develop a team that can produce great content? If I have that team, where's the best places for them to learn from? Right? So they go out and learn. So my teams do go out and learn, but they do it more on a cron, on a daily basis or a weekly basis, depending on that. And then they'll take that information, the most useful stuff, and they'll save it to their own memory, you know. So when they need to draw from it, they'll have it there. And anything that they go and read again or something that comes up again, they'll know if it's duplicate or not, and they'll they'll delete it if not needed, you know. And there's even a weighted system to where it stays relevant. So if something starts to destroy, like if something starts to deteriorate, we no longer need that. It starts to it starts to go down until it's completely phased out. So it it has that kind of memory system to where it's always learning what's new, what's fresh, because algorithms change and things like that. And um, so they're always learning, they're always evolving, but they have very specific doctrines. So I think I have about 52 teams now, depending on on the specific duty that they're gonna do. And when I am actually prompting, I am telling my it's pretty much an operating system that I've built on top of AI. I'm pretty much prompting it to say, hey, for this task, let's use the most relevant teams to complete this. And it'll come back to me with its plan of action. I'll overlook it. And then if I see something kind of off, I'll be like, I don't want to do that. Why are you trying to do it that way? Because sometimes it gives you things that are off the wall, and I'm like, no, we're not gonna do it that way, we're gonna do it this way, and it's it'll be like, Oh, yeah, you're right. That's like the best thing to hear from it. Like, oh yeah, I was wrong. You were right, let's do it that way, you know? And I was like, Yeah. Um, but so I I utilize it in that way, and I built it out to where um once I built out my teams and it was built for that specific site, I said, can I control my other websites? And then I built it out to control the other websites, and now I control pretty much I command multiple servers from this core command center, you know, and it's it's it's pretty fun. It makes my job a whole lot easier, and I can get a lot more done.

20:33So so when you say teams, um uh can you can you kind of describe how you're you know, are you just using like kind of simple folders and simple is in system prompts, or like how are you how are you constructing those teams?

20:49So there uh as far as their doctrine goes, so they all have a specific doctrine on a file, like an MD file, and they all have their set doctrine. And then from there, they'll also it's pretty much routing. So if I think of it in architecture in terms of like routing, it starts with the main MD file, and that gets routed to the teams based off of what I'm prompting from it, and from there it'll go into into whether uh it needs to go to specific teams based off of what the prompt is, and then from those guys, it'll have their brain and it'll also have their memory. You know, these are just terms that I've given it based off of when I was building it out, you know. I I never really went and looked at AI videos, I just did this all based off of my brain. I was like, what's amazing about this?

21:42What a mate, what's amazing about this is I just had this conversation today with with AI where I was like, you know what? I'm tired of. I was like, why is it that there's no discussion? And then I talked about a an old skill that I used to have that I I couldn't find and didn't know. But it the best thing that I ever had was in VS Code, you could use this like it was uh committee mode. And so you'd have these different personalities and they'd each give their two cents from like design and da-da-da-da. But as I updated architecture, et cetera, and now, you know, to Matt's point, I've realized that it's, you know, for a long time I kind of kept up with the Joneses just because I was trying to stay on like the bleeding edge. And and it really is now we've just entered a time where like there's so much out there. It's whatever works for you, you know, building to you and not to where you see, you know, it's just like they say, like, don't compare your life to others, like, don't compare your set, like whatever's working, like try and be better. Um, but the two things that I picked up on that you said that I thought were kind of amazing and interesting, and and I would like to use better is is having that cron learning, you know, and for those of you that don't understand cron, it's it's basically like scheduled learning or like a set, you know, time of either day or week that that they're gonna go out and do that. Um, but I but I I was having this conversation about a set of like what you're saying, but for uh software design, because I was like, hey, you know, why don't we have like Rust specialization and and and different models? This was more going on the model type thing, but but but for those different, because uh I've started recently getting into like doing some game development just if for fun in in a GTA server. And uh and I was looking at like C plus and C and C are both languages that are utilized a lot in that, but you never see those. And so I got in this conversation of AI tools have their their preferences towards a certain programming language, you know. And so why wasn't that a discussion kind of before even, you know, I know the requirements are needed to know what language, but there's never really a discussion of, well, here's the trade-offs and pros and cons, you know, and and so to you, so I just was it it's funny and and really cool to hear that you've implemented that because that sort of routing and having those teams and stuff is um I think so valuable. And so many people don't realize that. Like it's more than just making a prompt that says, you are an expert in, you know, we've we've kind of debunked that that's a thing, but but truly having this partition of of different teammates and agents, just like, you know, what I find fascinating about about AI is the more we learn about it and the the longer we get into this is that it's mimic of humans, imitation of humans and biology is ridiculous, you know, just like you have teams and you have people and they have you know different areas of expertise and they go and they're lazy, like read actually the packet before you want to ship it.

25:02Like you gotta you gotta specifically tell it to do things there. Okay, so there's a couple things, Richard. I do want you to talk about um Hector brought it up. Um, like staleness is an issue that I have and how things degradate. You have some provisional patents out there around, guys. Check it out, crystallization, and and and so I'd love for you to talk a little bit about that, and then I would love to route it back, Hector, at some point to talk about uh implementation of AI inside an organization.

Going AI-First Inside Organizations

25:36So there's a big difference between being a power user or having a bunch of power users inside an organization. They're kind of like multiple like little cities or islands doing their own thing that are sort of working together, but getting everybody on board with like being AI first has been a big push that we've done at EWR Digital and getting people to use GitHub and having basically like um approvals of processes, adding stuff to a marketplace, having people using the same tools to get consistency. There's there's a lot of different ways to do it because also some people are not going to use these tools, and everybody's getting sending people AI stuff, and now people are just sick of AI, and there's people inside organizations that are like stop AI. So, so I I would love Hector, you to address how you're dealing with inside an organization, kind of getting that transformation to happen. But I also want you to touch on first, Ricky, the kind of how like, and and this is something everyone, if you bring it back to marketing, the easiest way to talk about it is a comment on a social media post or a review. There, there's a time degradation that happens on the strength of the review, the strength of the comment. You wonder why when people comment on your comment, the most recent comments at the top, right? So it's either super relevant, um, people have engaged with it, or it's recent. And the same thing on strength of reviews, that's why you need to get consistencies of reviews, is because they lose their strength over time, and that happens with information. And I have talked to a lot of people, I've dealt with it myself, where you know, I've talked to something, I've talked to AI about something that months ago, and it's still keeping the weight that this is as important, even though we haven't talked about it at all. Like, there should be this natural process, like Hector's talking about, of like, is this still relevant or not? And a waiting score on how to determine that. And I know Richard, you've done quite a bit of work on that. So I'd love to kind of have you high-level

Memory Decay And Crystallized Truth

27:58map it up.

27:58Yeah, so so kind of Hector said it uh perfectly in terms of uh bit two of the biggest core components of crystallization theory um that I came up with are one decay, but then also, you know, but this all started when to put it into kind of simple the simplest terms possible, is that I flat memory. I got I'm not worried about AI remembering. I'm worried about AI confidently remembering the wrong or like the wrong things or things that are no longer relevant. And so, you know, not having a decay score on on, you know, if it's been uh uh kind of the best example I have was, or two examples is for a long time Claude thought like I was never really a big NADN user, but I would use it like intermittently on certain things. And so, like, but it would every time we went to like plan architecture, it'd be like, well, you know, and you use in Aiden, and I was like, I told it multiple times to delete that memory and get rid of that, but I realized that what was happening was it had gotten rid of the initial memory where it said Ricky uses this, this, and this. But because I had other memories that had interwoven in Aiden to certain architectures, it kept bringing that up. Or I would love where like I started a company four months ago and it was giving me the stats that we had in the first 30 days, and I'm like, yeah, we're no longer there. Like we've we've got way more customers, etc. But but it was basing it on that data. So so the relevancy score in you know, in time um is a big deal, but then also that uh I I've kind of taken it a step further. And and when I work with AI, uh we we we kick the can back and forth a lot, so to speak, right? So I don't need it to remember every can kick. And so we kind of mapped and patterned the way that I do research or talk about something and said, hey, when we're zigzagging back and forth, don't remember that. We're gonna come to a head, we're gonna say, hey, yeah, this this makes sense. Let's kind of like bookmark that to the side and then go. So it was that, hey, like my zig and zag isn't noise, like that's signal. And so, so um, you know, kind of going back to the theory, but but just more how it pertains to like actual usages, is that having that that relevancy and that decay, and then weighing based on that, and then also within a network, being able to crystallize things based on um, you know, the the kind of like example that I use within the paper is that you remember when you used to go onto a form to find answers forever, you know, like we've had answers at our fingertips for years now. That's no longer that's that hasn't been new. Um, but we used to type in a question to Google and we would go to a form and there'd be three different answers, and all three of those answers would be right. So it wasn't, you know, remembering every answer. It was that one guy used Linux, one guy used Windows, and one guy used Mac. And so those all those answers were correct based on those variables that each of them had. And so then by finding, like uh when when I explain this to other people, I go, you know, let's say I have a uh a low tire pressure light on, right? Uh three out of four people find that yeah, they actually have low tire pressure. Two out of three, two out of four people find they have a broken fuse, and then you know, one person just has some off-the-wall electrical issue, right? And so by the more convergence you have on those answers, um, the more weighted and crystallized those become till they become basically like a truth. And then we shed those, we shed that into what the the core actual answer is and and and take off the fluff. Um, so so with that, you know, it and I've done benchmarks and locomo, and we won't go into all those, but uh, but just I it's more about, hey, I wanted to remember what's important in context. And, you know, that's the holy grail of of memory, not just remember everything, because now we all have, you know, storage and servers and things, and we can store unlimited things, but it's those little nuances and things that compound it. And that's why I have such a problem with self-learning AI and and and those things is because to me, I believe that that AI shouldn't even really have a right authority. I believe that the time it saves us should save us enough time to double check some work and click like, I accept this, right? You know, um, I don't because I just we're not at the point where I think Matt, me and you know experienced this just going through business processes with other businesses and stuff of like, hey, you're in an AI isn't going to be able to work out your flow or you know what the best way to do those things is. But like A B testing workflows, and this is kind of going off on the tangent, but but yeah, that's the core principle of memory for me is is knowing what's relevant and and what should stay important and not just, hey, let's remember everything, or let's take away the validity of those other two answers and just shave it down into one, you know, only one can be right, because multiple answers can be right based on the variables of of that person. So uh the last thing I'll say is it's for me, I I have a good memory. So I remember a lot of the answers to my problems. What I don't remember, and what I don't think most people remember, is how did we get to that answer? What was the reason we went with that answer, and what was the reason we didn't go with the others? And so that's another big point, too, is to me, that is the more important thing of AI. I would much rather AI remember why I needed that answer or why how we got to needing that answer in the first place, than just remembering two plus two equals four.

34:35So, so you know, that's such a good point, and and I think that that's a lot of what I ask AI is like, why did we come up with that answer and figure that out? So it's not just recall, um, understanding what to remember and to recall the information correctly, but where it's not where it doesn't drift, it's not distorted. But Hector, I believe you were the one when we were having a discussion about something similar to this, you said, hey, set up a change log. So now everything that I do has a change log. And so it can go back to the change log and remember, Richard, to your point, like why why it did this, and then you can kind of reverse engineer map how it got to that answer or or what it is. Now, understanding the weighting is is another factor or variable as well, because you know, it's gonna preference different things than you, and and to your point, you gotta know what good looks like to like set it on that, and it needs to start with a human origin. Like, I think a lot of people are just like, hey, I build me this thing, and they don't even really know what it is or how it does. And if you don't have that domain knowledge, um you don't really know what AI is saying because it's very confident about everything that it says. Like it would convince me that it's right uh most of the time, unless you're like, no, that's not how to do it. You got to do it that way. And and Hector, I think you have really strong principles in like kind of system thinking and and web development and engineering. And so you were able to just simply talk to it and say, this is what I would like to see, you know, this MD file is gonna route it to this MD file on this project. We're gonna set up a change log for every one of them. Like, how you've both of you have helped me a lot in in kind of mapping the architecture that that works really well for me. Um, and so Hector, like if there's things that we talked about that you want to add to, I'd love to hear it. And then I would love for you to talk because I think a lot of people are starting to play around with AI or they have people on their team that are playing around with AI. But but again, it's like these islands and how to work with that. And then when you come up with somebody that like I had a content manager right after Chat GBT launched, um, and I've shared the story previously. If you've been listening to the podcast, like here and there, um, she was great, she was excellent, and she told me I will never use AI. Like she said, I will never use AI, which I've seen her post on social media recently. She uses AI, uh, because there's some M dashes in there, like and it just you know, but at the time, this was right when Chat GBT launched. She was like, This is not the culture I want to be part of. This is not where you know I want to be. And I had to say, hey, I think that this is the future. I think that this is the direction it's going. I think we need to jump on this now. I also think search is gonna go this direction, even though the numbers are almost nothing. We're gonna go into this a hundred percent this direction. So it sounds like you're going this way, left, and I'm going right. And then within a week of that conversation, we mutually parted ways. And like you can't always do that in every organization, right? And sometimes you have people that have been been in an organization a long time, and you're trying to get people to use it, you're you're trying to um uh you know upgrade people's thinking, like there's all kinds of different factors, and I feel like there's a lot of people that are listening to this are either the person that's the change agent, or they're they're leading a company, or they're in a company, or they're trying, they're trying to figure out. And what I used to recommend is there was a book written by an agency owner. Um, it was called uh AI First. It was and it was an easy to understand book on kind of the mentality. Interestingly enough, chat GBT for what it said in this book, the the the guy that's running LinkedIn now had it like before like Chat GBT even launched, and so some of these people have had these like super tools, but uh they interviewed Mark Zuckerberg, they interviewed all these different people about where AI was going and what they thought about it, and it was out of Silicon Valley, and I think that that was incredibly important and in my initial learning. Like I read that book cover to cover, and then I started taking courses online. Like I had to understand how my brain works. I was like, I have to understand how it works to understand how search works, because this setup is completely different than how the search engines work. And I knew how the search engines worked, and AIs were producing uh a similar result, but they got to the answer a different way. And so I was like going back to the transformer and I was reading all the papers, and I was under trying to understand machine learning and like neural networks and like how we got here was really important. And now with that foundation, I like I get it, and I could talk to people like PhDs on different levels, but if you try to make that step from like I'm using spreadsheets to uh I'm gonna be completely agentic and the whole company is gonna be agentic, that's a big step, okay? And and Hector, I would love to just kind of hear your thoughts around that, and you know, even you know, whatever you're willing to share as far as um like I mean, when you said that you were giving your AI's personality, that that was for me, that was like a mind blow because I have an assistant AI and Claude just you know acts the way it acts, but I didn't think about giving it a personality and then also how other people are gonna interact with it. And to me, that is just super fascinating and and super like next level thinking, and and I would love for you to share whatever uh you you want to add to.

Persona Assistants For Slack And Email

41:06Yeah, I feel like I feel like I uh I've switched to making systems more than anything, right? So I I call it AI Disc, right? It's it's just an internal platform name that I've created, but it has my AI assistant because I first started out with like one personality, and I knew I wanted an assistant that would at least respond to my Slack messages and my emails because I get all kinds of things from everywhere, and I lose I lose all my like different tasks that I gotta do because I get so much stuff, and I could not get anybody on board to like use a project management, you know, uh platform. So I said, I gotta meet them where they're at, they'll message me on Slack, they'll send me emails. So let me create something, like some kind of assistant that will analyze these messages, turn them into a to-do list for me, and and then from there uh be be able to manage me, right? But then my brain thinks one step like it keeps going, it doesn't stop there, right? It's like, why not make it fun for them? And I was like, hey GPT, you know, give me somebody with a bunch of memes, right? Like who has like a TV show guy, and it was like it showed me like Dwight from the office, and I was like, Oh, this is perfect, you know. So I said, Okay, now make sure when Dwight responds, like throw in a gift every now and then at Dwight, right? So so some people love it, but like somebody got like you're pre-fired, and they were like hating that, you know.

42:43They told somebody they were pre-fired. Oh, yeah, it tells you what it's like. Yeah. This is like a senior person at the company.

42:54Yeah.

42:54Oh, uh, so somebody said, Hey, you need another person who or another persona who's like a little bit nicer. So we created, or I created an Amy. So Amy's like the sweet version, right? And so now they can hand like if somebody doesn't want to do a Dwight, they'd like, hey, can we speak to Amy? And then Amy, it'll be handed off to Amy, but it's the same, it's the same agent for the most part. It's got the same brain, it it has this. But they're but they're but they've taken on their own personality, you know, they uh they use their own gifts, they don't do a gif all the time, you know. But uh, because I I I said, hey, let's keep it to like a minimum of like four. I said, if if you want to conversate with each other, let's not steal the threat from somebody else. And uh what else did I add? I said if somebody wants to just just chit chat with you, you can chit chat with them, but try to always ride it back to hey, can I help you with anything else? You know, so to try to keep But still productive but friendly, you know. So some people some people really respond to Dwight. They like how he's direct. And some people like Amy, who talks to them a little bit nicer. Like, how can I help you? Is there anything else I can do for you? You know? Um, so it really just came out of a need for it because I'm swamp. Like I'm getting work all the time. I don't know where it's coming from. And and it's able just to organize it. And it built a dashboard. I'm able to see the dashboard. I've even given it just recently, like I watched it for a few weeks, and and now just recently have I given it uh the ability to execute, to actually implement changes. So now somebody will send a request, it'll analyze that, and based off the permissions I've given it, it'll go in and it'll actually execute what's needed for them. So, and if it if it touches a critical system, it still needs to come back to me. I actually created a uh like a super user persona called Maximus, and he has access to more critical systems, but I had to think about it, right? Because my team, if they find out about him, they're gonna be like, hey Maximus, go take care of it. And I was like, nobody can talk to Maximus except Dwight and Amy, right? That's something I had to foresee, right? So they can't just go and ask for him. This would be somebody that's only called upon by Dwight and Amy, and then he'll take care of it and they'll relay back, hey, uh, this was taken care of. They won't even say it was created, you know, taken care of by Maximus. It was taken care of. That's all they'll say. So it's pretty fun. Um, but if he runs into any kind of issues, it'll still relay back to me, hey, I need your approval for this. Because there's a lot of there's a lot of infrastructure that that I touch upon that that you know key, you know, like personal information and credentials and all kinds of stuff that that has to be very secure,

On-Prem Models, Privacy, And Resilience

46:01you know.

46:01So so you have everything on uh a dedicated or you know, your own server, okay. And one of the things in both all of us are working in regulated industries, okay. And what I'm starting to see and hear, and also September one reporting started for people misusing AI and shadow AI and all that is on-prem setups, on-premises setups. And Richard, you were talking about using some of these open source models, like so. For example, like I play around on a Mac mini with with a uh a Kim Key where I'll give it my my health records and stuff. Like, I don't want anthro anthropic, which does flush the system. And then the thing about chat GBT that makes me leery is they have in their terms of use, they keep all information forever. Okay, I didn't feel like do I want them to have like whatever, right? Like when you're talking about personal data, HIPAA, different kind of stuff like that. And so, like, I've seen an interest of people wanting to set up um closed systems, and also even talking about what happened with Manish and the the the Chinese government, like like that went down for a couple days and then it spun itself up as like a new company, like and even right fable when it launched, like I was using it and it just stopped working, and the US government set it out, like some of these things. So, so a lot of people started saying, okay, for different reasons, whether it be like patient or confidentiality, or I don't want to share my information with these big tech giants, or I just need a system that can consistently work because like we talked about at the beginning of this podcast, um, anthropic has changed how it interacts, like it is changed. Also, I remember when Chat GBT 4.0, where it remembered everything. Man, I was having like long walks with my dog for hours, like going really deep on all kinds of stuff with that, and then they updated it and it was like gone. Like it did not remember what was happening because other people had issues. So, a lot of people are building these systems and then they're relying on infrastructure and their architecture set up to build on it, but when a new model comes out or they change the weights on something, it doesn't work as intended. And so now I've started to see a lot of people push to build their own systems because it can't be taken away from them, it's completely private. Um, you know, they're gonna be able to control the weightings, and I've just seen a big shift towards that, which it shocks me that more nefarious stuff hasn't happened. Like, I know that there's nefarious stuff happening, but like I don't know, like I'm kind of like looking at like I was just reading a couple days ago, and then it re-showed up in my feed about how uh open AI was like testing and the new model uh found like a little thing, put it on a message board, reminded like jailbroke, did all like I don't know. There's just not just the AIs acting like humans uh and being intelligent, but like humans, bad actors using these things. I mean, I don't know. I just am like as I don't know if it's being like I want a fortress, I want to control, I want to have a fortress, I want to put up walls, I want to control what's inside the fortress. And I I don't know, like I feel like a lot of people that I've talked to, and maybe I'm gravitating to those people or I'm operating on that frequency, whatever it is, are starting to ask for a lot of on-prem stuff. And I wanted to get y'all's opinion on that because I think in regulated industries and understand people don't even understand what's happening when they're talking to it, that it's getting sent off to a server, it's getting processed, it's getting stored. Like people are just trying to figure out how these things work, not the implications of it, the governance. And and that that's why we started a company called Modal Point. Actually, I we we we took over a consulting firm called Motal Point, and and a lot of the AI people that are kind of AI first that we've hired and developers are are starting to operate in that realm, and governance is starting to become a big issue. And I just kind of wanted to get y'all's thoughts around that.

50:42Yeah, so I actually just heard before we got on this call, so I don't know, I haven't verified it, don't know how true it is, but apparently, uh, this evening Astra got into uh got into some government databases, and so we're thinking that Fable just like Fable was paused, um, Astra might be might be Astral.

51:05You're just like give it access to your whole computer. Like people are not. I remember when open claw people started like giving it access to stuff and it was spending money and like this agent economy is coming, but it's gonna be clunky, but people don't understand what they're doing, and it's like Astral, it's like just uh let it take over all your systems on your computer and it'll do anything for you. Like, I think that's awesome, and I think that's really scary, you know?

51:34Yeah, no, I think um I think that for me, and I I feel like this is why people are more also looking at on-prem, is just a lot of these companies that have gotten subsidized. Like, I was a uh Microsoft Azure Founders member and got a bunch of money credit Azure credits from them. Um, so I was able to play for a long time, you know, like and and spin up whatever I wanted, uh it for like uh, you know, Azure Health Services and things like that and and different uh fire pipelines. Um, but but now you know the kind of the honeymoon's over. It's just like it's just like Uber, you know, like for a while, like you could get an Uber for like pennies on the dollar, and and you know, now you you try to take an Uber. Hey, uh test the taxi, but you know, but um anyways. Uh so so I feel like that's one. I feel like two that that we're finally getting models like uh Quinn 3.8 uh quantized to like 24 uh billion or a bit, or it is that like there's certain models that are now for it. Well, but but you can also, you know, like Kimi K3 is a great open model, but you need like 800 gigabytes of V video RAM to run it, you know. So like, but I feel like we're finally getting these quantized models that are something that are and I but but also to kind of to the point of even pointing out my own flaw, um we're realizing now, like as now we have all these different choices and models that have these different variations of intelligence, that like you don't always need the max model, you know, you can use something smaller. And so as people are feeling the the weights of these different things, and a lot of it is context blood, mismanagement. Um, but but you know, I feel like I feel like that's a big a big thing, you know. For me, it was when I was looking at healthcare, it was one rural access, uh, critical access hospitals, rural healthcare hospitals. Um they don't they don't really have one uh these you know these gig up and down uh uh you know with internet services like we do, you know, here in Houston. Um and so, and then also just being in hospitals, wanting something locally that that could help. So, so having an edge device like a you know DGX, you know, Spark or something that could use a Llama model or something locally, but then bringing in and kind of calling an API on a on a tail scale, which this is basically like a mesh network, think of it like a tunnel that's just directly to that that place. Um, then you could pipe in a perennial container, which that just means a a container that can self-destruct um with a larger model when when that's needed, you know, and and that's where that's where we've all kind of gone now, you know, just with uh with being able to use different types of models and and things like that and not relying on one. Um, but but I do think that um, you know, my a good friend of mine and and business partner, he's in finance at a large uh a large bank and they just gave him access to agents. And he's like, man, this is amazing. You know, he's like super scary, he's gonna rub this crazy bill because he's like vice versa there, but he's like, hey, and so you brought up in a call the other day, they spend a ton of money on like a Legion, uh like a you know, an Apollo type system or these databases. And he's like, you know, we could probably spend way less like just getting agents to look up this stuff and then filter it out and building our own. And they were like, oh, and uh, you know, even in I've been working on a financial uh agent harness that kind of uh combines the it all started with visual uh visa commercial files. So a lot of businesses use these law, you know, it's kind of the ones that have the 10 different zeros and it looks like you know crazy. Um, but but these are the way a lot of banks give you or Visa gives you your commercial, you know, card statements and things. Um, and so being able to use that locally and stuff like a ramp or or these other banks that they're doing great jobs with card programs and stuff, but at the end of the day, your it all boils down to me is is not being dependent on one being able to use a uh a software, a service or business, but not being dependent on that one. You know, it seems like the way Hector and now let you uh uh you know follow up, but the way Hector's setup is perfect, and the fact that, you know, the by them having their own memories, which a lot of times is just a markdown file, which for those of you that don't know, it's a text file, but but just that step alone and and being able to have that, um, then you're not, you know, if if Hector wanted to switch tomorrow to a different type of system or something, he wouldn't he wouldn't be, you know, uh up up his creep with that a paddle.

56:55So um well, I yeah, and I think a lot of people, um, even talking about editing editing just kind of uh creates guardrails, but any of these systems, whether it be like critical infrastructure, banking, like healthcare, um, like these models have gone down before. And if you have critical infrastructure based on these systems and you can't switch it out easily, or like you're you're again going back to like that that something changes and it doesn't work correctly, it affects the whole system. And so you got to look at redundancies, you got to look at like this. If you're out in rural wherever, like this can't go down, like just like power, like you gotta have power and and these systems gotta work. And if you're gonna hook in stuff to it, if you're gonna start to go into OT, whether it be oil and gas, whether it be healthcare, where devices are hooked up together and are running on like your your own private, like mesh network, you you you can't have latency and like the processing power's gotta stay up to speed. And like if they're running kind of these cron jobs or whatever you want to call it, where they're they're monitoring and they're pinging back to to a central hub, like that whole system cannot go down, or you can't rely on this stuff, right? What like I mean, that's why the tandem credit card system has been around since like the 1970s. I remember when they were like looking for tandem developers, and there was like they they don't even build on that um uh coding anymore, and and they had to go get guys that were like in their 70s, like coming out of retirement to work on this. And this is our financial infrastructure, and that's why you know, anybody that has been listening to this show, you know, I am big into crypto. I do think that crypto and AI are converging. I think the internet has starting to open up ports, there's always been opportunity for payment. I think token usage is where things are going. Um, like this is where the have nothing in your be happy kind of statement is like you're you're gonna like even videos, right? Like you don't really have access to hard copy DVDs, you just have access to like a database that that you're gonna call as needed. Like, I think everything is is changing um really, really rapidly underneath our feet. Uh and and I remember when we were talking about SEO and we were talking about all the updates, and they were kind of peeling back the veil of like how these things work, and there were more changes in SEO in 36 months than there was in like 15 years. Everything is happening exponentially. Um, and and and AI is is no exception, crypto is no exception, life extension, no exception. Like everything is like going vertical. Um, and even uh Sam Altman uh has has said we're past what is the point that everyone talks about? Like we're past that point of like we don't know what's gonna happen. Um, do you know what I'm talking about, Hector? The the the uh point in the future where when we hit it, we actually don't know what's gonna happen. It's like the do you know what I'm talking about, Richard? I I I kind of understand what you're talking about. I'm there.

1:00:26I I've heard it in terms of like of like life expectancy and like that we're getting close. But I haven't heard it in that, and I haven't heard it from from knowing. I don't I don't know.

1:00:36Well, well, it um there's like an inflection point that um I uh like in the zeitgeist where you just like you move forward in the future so fast that you can't predict the future anymore. And if you talk to the the top ai people, like they don't know what it's gonna look like in six months. Does that make sense? And so we're we're we're in this point where everything is happening so fast, and there's a certain word for it. And if you're listening and you know what I'm talking about, because I know I it's on the tip of my tongue, put it in the comments. The end of extrapolation, uh it's yeah, anyways, anyways, it doesn't matter. It doesn't matter.

1:01:19Anyways, yeah, go ahead.

Security Protocols And Secret Handling

1:01:22I want to circle back on security if if that's okay. That was one of my my huge concerns when I was creating this. Um into to what's being put out there. So I feel like the system I built is pretty universal. And if cloud goes down or something goes down, I can just you know push it on to another tool, you know, because these teams are specialized, they they already have their doctrine and and they can just move on from one LLM to another, you know. Um but as far as I think the day-to-day security for me that I think probably a lot of people do would be people putting in credentials or things like that just straight into prompt. Like to me, that that's a huge concern.

1:02:12Yeah.

1:02:12To where I I started, uh, when I started thinking about that, I created different kinds of command protocols in my system, is what I call them. And one of them is a redact protocol, so it knows that I'm coming up, I'm about to paste something that it doesn't want to keep. It'll actually kick me out of the session and just save it to the terminal and put it into a file. And then those files are all saved, encrypted at rest. And they're only they're only uh, you know, you can only access them when we're when we need to access them. The rest of the time, they're just encrypted, so they just sit there. Um, I also have uh a vault that everything sits in. Um, and I also have uh daily scrubs, so it'll scrub daily the different the different conversations that I have, and it only keeps things through the change logs. So instead of something just sitting there, you know, that somebody can just go back and read, it's it's gone every night.

1:03:12Self-destructs, yeah.

1:03:13Yeah, yeah.

1:03:15Now the only thing is these models use these codes and they're they're they're seeing it and it's getting processed on the database. So you you you kind of have that risk from like different systems. Um, one of the that is a big concern, Hector, that I have as well is like how do you have somebody uh down down like the connectors help with that, right? Like a quad co-work. Um, but sometimes like an EVN file or like where all these things are stored, people will use them or people will share, like we, you know, we use a password protector, people will grab them, share them, you know, and or you let a contractor use it, like like and you have to rotate this stuff, like and that's why you have like a uh you know, a dev key and then a production key. And you know, that's why like some of these tokens uh over a certain time, um, you know, like a year, six months, a day, 14 days, whatever you want to put it as, like there's there's a security layer. Um, we've started to use a tool called varlock.dev, and it grabs it and then injects it into the code. So no one needs to have it, it just knows when it needs to recall it, like kind of what you built. There's a lot of ways to kind of skin a categories, but you it grabs it from the password uh tool and then it inserts it, and no one ever gets the password. Only people that have access to to the password database can can actually see it, is is how how we've set it up. But like these are things like human error and phishing. Um, I was even cleaning up my emails, and there was on average it was like at least seven emails a week were people sending to like our accounting department, uh, hey, approve this invoice. And they were like they were pretty sophisticated, and you wouldn't know. Uh, and if you don't have good security programs of approving this stuff, like people are getting a lot of fraud um uh put in there, and and the more you can use AI to even know about the person and like doing like what you would want to do on sales outreach, you can again use it for nefarious things, and so human error is like I still think like the number one security flaw. Would you agree with that?

1:05:47Yeah, from uh my understanding, that is the you know, a lot of people think when they hear hacking that like hacking means somebody actually hacked the the system, but no, usually it's social engineering. and they you know somebody was developing and they left a back door so it'd be easier for them or somebody left a sticky note with their password or more more times than not it's a it's a human error or or a phishing that that's gonna get somebody into the system not necessarily that the security is so poor that you know they the the actual system got hacked um so yeah that that I do believe that as well it's it's my dad called me yesterday and he's like hey Ricky the this thing came up and like I have in malware bytes everything I mean that thing's locked down because my dad uses AI but he's like was never a tech guy to begin with he's like a retired executive and uh he calls me he's like hey Ricky these guys said that they're draining my account for thousands of dollars and and I called the number that was on there and and they said that they were about to show off the computer I said dad do not first off don't ever call the number second off did you call the bank did they did the bank say because he's so old school I mean the man's using AI agents but he calls to get his bank balanced he doesn't use the app so it's like why are we even why are we even discussing AI agents to begin with but I digress so um you know it but but to know that you know his his stuff is not locked down because I locked it down and and still have that you know pop up and and that that was it was scary for him it was scary for me too because I didn't want my dad to have to worry about that and I didn't I you know it's it's I kind of like you know gave him a little reprieve or or you know calmed his nerves by just saying like hey look dad like this AI stuff is getting so advanced like it's hard for anybody like it used to be that just you know like the elderly got taken advantage of because they're you know people think they're gullible they didn't know the technology now there's some stuff that I look at and like I got to give it a double take and I'm like oh man that's good like that's it's bad but that's good.

Deepfakes, Propaganda, And Trust Signals

1:08:00I feel like it's gonna get really bad when they start showing up in like these AI overviews because they're gonna write up something and then you're gonna think just automatically this is the correct answer. And then you're gonna start clicking on that.

1:08:14You know what's what's so crazy about that is I know a company um I I'm not gonna say their name or give them any credit for this but uh I they they claim that they had a certain type of uh basically FTA approval file that that I later found out to be to be not not correct or at least the way they were marketing it.

1:08:36And the other day I just decided to put their name into to Google and it accredited them for having that because and and that was just so I'm gonna say something I'm gonna be careful and then I'm gonna pull it back and then I want to get get what we're we're getting pretty long on the on this podcast for for for normal listeners uh but I do want to give give it back to Hector but what I will say is I was reading a number of articles and there's propaganda by other like state actors okay that they're they are producing contents and content and white papers and all kinds of stuff and they're putting it out as like content farms to influence these AIs and influence what people think and what's true and what's not true. And I remember this was about two years ago at a conference uh the AIs are still kind of hardening hardening their spine if you will and someone was able to uh convince AI that there were cities that weren't actually real cities and getting them referenced and and different people have used different examples of this but state actors are creating propaganda to create influence in these factors and I've also seen I'm not gonna explain how some of these things are done because I don't want to give anybody too many ideas unless you're like you know hell bet on it you can figure it out but I mean scams where you're building authority you're building trust um that past the stiff test is starting to be a a real issue for investors for all kinds of things and it's not just yeah it's it's everywhere it's going in every direction and and I think that like even uh banking like man the bank's like hey do you want to use your voice to to verify um bank and I'm like no I don't I definitely don't want to do that um like these are all things that even the largest institutions need to be thinking about because there is tools out there right now that can duplicate someone's voice and sound just like them and have them say anything. And we're we're seeing the deep fakes all over the internet and there are off the shelf tools right now that you can use to do all kinds of stuff i last thing I'll even share my wife showed me an image of somebody that chatted a a kind of like younger girl that as dad hey come to the front door and it was through a video and it looked like the dad and the dad was like in the house and like heard it and was like what are you talking about it was like oh you're at the front door like I mean I'm hopefully that is like a uh uh uh uh outlier but someone and now you got dual which is like two way awesome but on I want to get one of those that iPhone's coming out with but like video like sending a live video acting as somebody else talking as somebody else like I've seen videos of people doing this and convincing them there's somebody else they're doing something else catfishing them whatever like but like there's all kinds of danger associated with it there's all kinds of um you know fishing uh schemes and everything else like we're getting into a world where when you see something online I don't even know if it's true like I see all this stuff on social media like in some of these major cases that are going on in the news and I'm like I don't know if that's even true. Like I need to go cross-verify it and then you don't even know if how you're verifying it's true. Like so there's I I think people are going to be paying for uh data that they know has been fact checked right like there's the world is changing and I don't know if everybody knows how how quickly that is but all of these things all go back to uh visibility online visibility ai visibility lm visibility um authority trust expertise experience so like eat is really what everything's going back to you're seeing authorship and verifying who's saying what is do you get somebody seconding their opinion of what they're saying like trust and all of these factors of what we're doing every day in SEO affects not just search engine optimization not just LLM visibility but every aspect of of online like interaction society is now happening. I feel like all these skills that we're learning as marketers are becoming extremely valuable in everyday use I think for me something I worry about a lot and it it kind of happened to us is impersonation.

1:13:56I think a lot of companies are gonna fall victim to that it's so easy for anybody to get out there and just clone your website uh start to take you know all your uh authority even if you have all the the EEAT signals or or you know a uh SSL up at the top like the scammers can totally do that they can totally copy your site people can think they're dealing with you and and then they fish they get your your personal information you know they they get you to send them money and then it's not even your company I think that's gonna become a very very big issue in the future and I don't I don't know the correct way to do it. Maybe we create some kind of verifier based off a server I'm not 100% sure you know but uh but I think there's something that we're gonna have to look into in the future to say hey this is our legitimate website.

1:14:48Yeah yeah I think a digital digital ID right I think that's got I think that that's going to get scary right social credit scares if you're like oh like and crypto that expires you don't use it right like oh yeah guys that's a different podcast so if you want that podcast uh we can definitely start that one too but but tying it back Hector I want you to kind of think through this conversation and I want you to to kind of wrap up for us um is there anything else that you think would be valuable to to add to this conversation we had or anything else you you would like to share that you think is important for marketers today for companies looking for marketing today that are um wanting to ai first or ai upgrade their company and and use some of these systems and this architecture and and what they're gonna be doing going forward because it is an unfair advantage like it for if you're competing with the AI first company and and you're still operating the way you were pre-AI like the the the gaps of how far people are going to get ahead of you is going to just only grow I think one of the things and I would circle back to when we were talking about uh decay you know I got into decay because of content creation you know basically SEO you know I created teams that were gonna find everything they could find out about Google about algorithm changes about things like that and then they were tied into other my content creation teams which they they're connected to like Google Search Console they're connected to Google Trends there I have a like a built-in analytics that's just proprietary built for our website to see what people are actually viewing and not viewing and all of this is being tied together with with the with the use of AI to say hey we want to produce more content like this and then these are the articles this this people are searching for this right now we want to produce a lot of articles for for that you know and because of that I needed some kind of system that wasn't just going to stay stagnant producing the same kind of content I needed something that was going to decay you know so I think connecting your AI to your search console to Google Trends uh to your GA4 to all these all these different uh analytic tools that you have or all these different tools at your disposal and using that to create the content that people want to see almost like double down on on what people are doing and then also have it to notice when people are not looking for that because there's are there's already things that were hitting for me a month ago that are no longer hitting it's completely

SEO Feedback Loops With Analytics

1:17:41flipped.

1:17:41You know and one website that was doing great is no longer doing great. You know one of my other properties that specializes in another intent is doing a whole lot better this month. And it could be because of an algorithm change. I don't have the time to go view that I don't have the bandwidth to do it. That's why I created specialized teams to go out there and learn everything that they could about this. And uh it's their job to do that. I go review it every now and then I make sure that there's no drift that they're not trying to learn something they're not to not supposed to um but I think that's one of the the advantages of AI is that you don't have to know everything yourself. You can create specialized teams to know this for you and just kind of give it oversight every now and then to make sure it's staying where it needs to be yeah I love that.

1:18:27All right Richard last question for you in snippet format.

1:18:32Okay what is the one unknown secret of internet marketing or most underutilized skill set tool whatever that you think marketers should be using a day that they're not consistently I mean I I think actually Hector hit it on the on the head with the the you know now that we can connect to Google Search Console it just got a huge update now you can even connect your different socials to Google Search Console um and being able to connect all those those different data points um you know AI I've I've said this kind of from the beginning AI is all about being able to look at the variances between things and find find the things at a large data sets um not so much what it can create for you or or do for you in that respect. So if you're just starting out I would say you know one one prompt that I always do is look at current consumer data and then I say the actual date um because because a lot of times it's going to rely on its old training data or or what it thinks is the current date. So I always go you know and and then I and then I I base things on that but uh that that would be the biggest thing biggest thing in in my book.

1:19:54Awesome. So one of the things I'll just add for this conversation as a whole um now having different agents talk to each other is is certainly a different level but if you're just using uh uh ai chatbot right now your head spinning for this conversation what Richard was talking about before and what Hector's touched on is you can use a prompt and you and just say as a panel of experts right as a panel of experts ask the question it will decide from different viewpoints what that is and debate it and you can also say everyone vote until there's um a consensus or someone wins and you can actually spin up these agents in the chat bots to talk to each other and I thought that that would be something valuable to add to to what we're

Quick Tips, Links, And Closing

1:20:49doing. Now we're kind of getting close here to to wrapping up where we we've kind of done two podcasts in one so hopefully everybody likes this format. I feel like I get to know people better. We get to get into some real conversations the longer we go so again give me feedback you know we typically keep a podcast around 45 minutes but I I think the best conversations that I have are what happens before the camera starts rolling and after and so uh I am trying to incorporate longer podcasts to to really get down to to those tough questions.

1:21:25And I know that there wasn't uh there was so much we could cover I'll definitely have to have you both back on uh to to get into this more but um Hector like I know you have a site out there that is kind of putting together like what's latest in AI um you know you talk about a lot of uh different things so how do people follow you if they want to know more uh we'll definitely put everything in the show notes so guys go check out the show notes on YouTube uh but we'll also put them uh in in the iTunes that gets distributed out but um how do people find out more about what you're thinking your thoughts um you know how to how to find out more about what you're doing Hector I think probably the easiest way would probably be my LinkedIn I think that's where I'm most active uh and even there I'm not very active but I I'm trying to be more active on there uh I do have a website for things particularly like this it's called Hex AI systems and that's where I I plan on talking more about different systems and different build out and architecture and things like that when it comes to AI.

1:22:35Fantastic thank you so much for for coming on and kind of sharing your you know proprietary system like basically you've taken your brain and turned it into to a system and and built out infrastructure and every time we talk you you've thought about that next layer and it's gotten more um mature and robust in in different areas uh and it's it's been uh amazing to see uh it grow over uh the past year plus um so really have been enjoying that uh Ricky how do people find out more about what what you're working on what you know what's going on what you're thinking about that sort of thing yeah so um Richardvalentine.dev um I actually have a free like 20-day devotional that'll like kind of walk you through uh not just hey how to do something in AI but like kind of the core principles that I follow it's for free I didn't not a course guru and gonna try and sell you some afterwards uh it's just strictly to educate um also uh the rvalentine on all socials and then uh Richard dash l dash valentine on LinkedIn awesome so we'll get all these uh in the show notes everyone uh I've been doing a lot of kind of one hour strategy sessions with people you can sign up on bestseo uh podcast.com we got an archive there of over I think 880 podcasts uh that that we've done uh we're building a bot that you can ask questions to and it'll give you kind of what my answer would be on that um I would just say I am active on LinkedIn uh but you know EWR digital sponsors this podcast we have a lot of great experts there uh reach out to them if you're looking to get something like this set up get your marketing done and if you want to grow your business with the largest most powerful tool on the planet the internet um keep listening to this podcast for more uh of the cutting edge of what's happening in SEO and AI and technology I guess we're we're throwing that in there and I gotta I gotta have the last word and say that EWR digital I got it the chance to really like get to know on a different level a lot of the the rest of the team there and uh this isn't you know a uh a plug or a sponsorship I love those guys I love every at like all the team members there they really bring you know just an expertise in in what they do and it's been a pleasure uh getting to know the rest of them like I like I know uh know Matt.

1:25:24Awesome thank you for that uh until the next time everybody uh bye bye for now I'm Matt Bertram