Hello, I'm Jeremy Rivera, on the Unscripted Small Business Podcast. I'm here with Mason, who's going to introduce himself and his company and tell us a little bit about his expertise and why we should trust him as an expert in Ai Automation.

Gotcha. Thanks for having me on, Jeremy. My name is Mason McCumber. I'm originally from Houston, Texas, all the way out here in the crazy south. I'm a SEO and AI expert. And I've put together hundreds of tools for individuals and users to use and actually learn some of the new stuff for AI and for SEO.

What's it like being a cowboy in a desert that is changing every single day?

This is definitely interesting. More on reference to being in Texas. The landscape here, definitely people set in their ways and there's some challenges when dealing with clients and doing services and filling services. Trying to get everyone caught up to speed. So I love teaching about obviously all this SEO and AI stuff and I've been using this to help really leverage and help business owners level up.

What is the baseline for business owners at the moment when you're encountering them? Is it just they're generally aware that AI do stuff? Or do they have wrong preconceptions about what it can or can't do? Or are you finding they're coming to you and they've already plunked in a bunch of prompts into Claude and done a bunch of stuff like, what are you running into and why?

So what I'm running into is people not understanding agentic agents. So I'll give you this analogy that it really explains where AI is at and where AI is going ultimately and how you can use the bus. So where AI is and where AI is actually going is for the longest time, if you think about everything like a car, it's been super powerful, like the engine inside of a car for the longest time, even when we were on the GPT-4 models or your Claude Haiku 3.5.

The models have been extremely extremely strong assuming no hallucinations for this really long time. So if you were able to go back and forth like you're working on something with coding or you're working on something with SEO or you're just trying to formulate an idea you had that engine, but you were kind of the driver in the seat there.

So you were kind of communicating back and forth with these agents or not with the agents but with the chat. And that was your engine and you were kind of the driver. So if you think about getting from a destination you have an idea in mind.

So you have to have the idea which is point A to point B and you're kind of the driver behind the wheel taking the turns the lefts the right and just going exactly where you need to go to make it to your final destination. People aren't realizing that with these new tools that are out now they don't have to be the driver.

So the new way that people actually do and use AI is they use that super powerful engine, whether that's Claude, whether that's ChatGPT, whether that's any one of the million AI models that are out and have specific things on that. But the driver now is an agent, AI agent. So that's either OpenClaw. If you like a lobster, it's co-pilot. If you like the distro, it's ChatGPT and Codex, or it's a crab if you want to use the quad code to quad code tools. But that is the new way that
everything gets done. And then the user only has to think about the A to B and then give some minor instructions on how to take the turns and how to do the driving to make sure they don't drive across the grass or doing anything too crazy. But that's the simplest way that I can explain it. And people don't know necessarily that this is out there. They're still manning a lot of this.

They're still talking to ChatGPT to make the website and produce code and do certain things instead of letting it drive the car and then correcting it and saying, hey, don't drive over that person's grass or don't do these certain things if you can imagine the analogy. But they're a truck driver before and now you're more a kind of operator of the business and you're an operator of the truck. You're the driver dispatch instead of the driver. People are just having such a hard time stepping out of that truck.

Guilty — and it's not that I haven't had conversations where three-fourths of the words out of the guest's mouth was agentic and aging and agentic. I know that I can be doing more to automate it but it seems to me — is it the block? Between making that jump into leveraging that agented model, not knowing of a specific platform to jump into that modifies your viewpoint of how to use that tool, because certainly the UX UI of Claude, ChatGPT, is very similar to — you go to Google, blank page, prompt, put in stuff, right? And then it saves your chats and then you come back to it. Is it just a lack of awareness or is it a — you can do it agentic directly through the Claude interface, but you have to do X, Y, and Z, which means that you have to have some sort of meta-knowledge that that is something that you could do.

Yeah, the being the dispatcher — there's still there is a good dispatcher. So there is a lot of skill in how you actually orchestrate your truck drivers to take routes, like tell it don't go down a certain path when you want it to do something. So you can send it on the back rocky roads or you can send it down a straight path and have kind of that goal there. But a lot of the problem is people don't know that the tools are out there. And that's what I put together — over hundreds of individual resources for not just AI tools. I'm not just gonna shell everyone AI, because they don't need AI for everything, but a lot of resources for general business and leveling up your business from my consulting time where I was working with a startup and working with a private equity startup. So I put a bunch of stuff out there. And then I continued to add on that with all of the new AI tools that people can now use. But there's a lot of people that don't know there's other ones out there.

And that's the main thing — like Claude is one. I could list off a million that you can go out there and you can go try. OpenClaw is a really good one and then there's a new one that people can also take a look at called Hermes. That's like the simple out of the box — this is the better one that most people should be using but this hasn't gone mainstream and this is like within maybe a month, maybe two. I think it came out like March early. So it's been out for probably 30 days.

So Hermes is kind of an agentic approach that helps you more intuitively set up these sustainable processes. I myself, I have, you know, like a laundry list of what I do after say a podcast interview — process transcript, generating different types of content. What are the guidelines? What are the processes that you're ensuring that are put into place? Because, you know, here's an example I told it to process the transcript and use it as a strict editor and never remove more than 2% of the content. And nine times out of 10, that works. But then I came back on the 10th one.

Like, I remember I mentioned this person and it's not in the interview article that it stood out. So I asked it, what percentage? It's like, actually there's only 50% of the original content here. “Sorry about that.“

What is your process or what do you tell business owners that are concerned about — let's call it the habit of ignoring instructions for the sake of efficiency or whatever reason.

Hallucination.

It didn't do what it said, it said it did something it didn't do. What are your controls or processes to minimize that? Because it's not hallucination, but it is in a similar vein. Yeah, it's like an aversion that it won't do the thing unless it thinks that you're watching it.

Yeah, it almost feels like aversion.

Yeah, so the way that you do this is you have kind of these persistent memory agents as well as like you staying on top of how you manage the context and shape all of the messages. So part of the reason that people had trouble with OpenClaw, if you've tried this one out or even any of the other AI agents — if you want to use Claude Code or you want to use Codex or any of these, there's a million that are out there now. But the driver in this analogy, the driver has a short attention span. It forgets so bad. Like you're telling it one thing and you could be feeding it, but there's not a consistent feedback loop. So it doesn't ever know.

The driver's up so high in the truck and it just drove over someone's grass, but it doesn't know it just drove over someone's grass. It doesn't really have like a way to figure out that you just messed up. Hey, this is the — here's your feedback.

Your dispatcher just told you, hey, you drove over someone's grass. Please don't do that. And so it has a hard time with that. So if you have good memory and able to keep up with kind of that feedback loop and constant improvement, you'll have a much better time.

I'm actually a little interesting on this though for being such a big AI show there. I personally do not use AI to make a lot of my content. And the reason for that is because of the fact that it'll do stuff like this, but also sometimes it lacks a little bit of the soul there. And I see very soon Google is gonna penalize people for lacking that soul. Like it's very hard for me and just for others if you think about like a bounce rate statistic. I'm gonna bounce if I end up reading too much AI content on your website. Like if it doesn't feel personal, it doesn't feel like it's any bit useful or anything there, you could have a million different things about a million different things. And it wouldn't inevitably matter to me if I land on your website and I'm like, I don't really care to learn this — this is so impersonable, this is like someone's copy and paste of their ChatGPT. Like that doesn't actually benefit me.

There's human insights versus instructions that ChatGPT gives you. I think people lose that soul sometimes and they don't even realize they've copy and pasted something from AI. It has no soul. Like it's very, very hard sometimes whenever you're just pushing a lot of AI content and these are like, it's a factory pretty much. They have it set up on an automation just to push random topics and random content and it's ranking. But it's slowly getting pulled in certain things there. So it's very difficult in that aspect.

But I think the better way to do it is like consistent memory management and using something like Obsidian as well as using some of the newer stuff that has built-in memory. So you got to do both. You got to keep storage of everything as well as have something to make sure that it keeps remembering it if it doesn't go and read the files. So that's my way.

Matt Brooks of SEOteric called it the high mind hegemon of AI? And it was like, they did a study and it was some set of creative writing prompts that for humans will result in very diverse writing coming out of it. They fed it to all of the different models and all of them ended up returning strikingly similar themes and tones. And of course that's an artifact of the language learning model. It's based off of math, it's based off of dictionaries, it's based off of the data that you based it off in the first place. So if they're all based off of a similar or same training model, then the output at the end of the day is going to have the same sameness.

Yeah.

Yeah, I actually want to put in something on this. This is something big people aren't realizing. They're like, well, I don't use Claude. You can even use one of the models in China right now that are like absolutely have no affiliation to anything in the United States. But how AI is being trained is they're stealing data from other AI companies. So AI is just bad in general about stealing data, but they're stealing it from other AI companies to train their model. So how they train Kimi — one of the really popular models right now for coding and other things — is they just sent out like a million ChatGPT prompts to learn everything it knows about coding. So it is ChatGPT. Like it is the exact same model. ChatGPT is Claude, Claude is ChatGPT. They're all the same.

Yes, there's like a one-off jump and there'll be an improvement where they don't post all of their workings in the background. Like Grok is coming out with Grok 5 here soon. And it'll be a big jump, but at the same time, once it gets published and users can use it and you can use it with the API, they'll just send a million prompts, learn everything that they need to know what the new secret sauce is, and they'll use that to train their AI model. And it just happens so quick and that's why this cycle is so interesting. It doesn't feel like really anyone has their… their secret sauce ends up getting leaked.

And that's what makes all of this content so similar is you have to understand that they're training their model with other people's models. So there's absolutely no way that you can truly have something that is unique, even if you use something that no one's ever heard of and use some sort of AI model that's in a far off distant place in Norway or something. It's still being trained with ChatGPT and with Claude. So it gives you stuff from ChatGPT and Claude. It's simple as that.

Years with this pace of change, how frequently are you having to revise your sets of suggestions for your consultees?

All all of the time. It's a constant battle and I'm encouraging more people to use it but it's definitely something that you're still having to baby it and monitor it. Anyone that has something that's kind of set it and forget it — it's doing well. Like, there are some people that have some models that are tuned for some specific things. But if you want it to actually work and be a growing and long-term employee, I am one of the people that is just constantly monitoring. It is like an employee. The best way I can describe it is it's a teenager on steroids. Like, it's hopped up on G-Force and it's ready to go. But at the same time, like, it's a teenager at heart.

It goes in the wrong places, it touches certain things it's not supposed to touch, no matter what you do, it's a teenager hopped up on something and it's a little troublesome but if you want to really maintain it and keep up with it, you've got to be on top of it.

So I'm adding a segment to the show, if you've been listening for a while. I'm going to do something new. I'm having you answer a question from my previous guest. And then after that, come up with a question for me to ask my next guest. So the question was from Jordan Fernandez who does Tucson SEO.

Looking at the lifeline, lifetime of the business and saying “HEY, if you're going to start a business, you need to think about how it's going to end!” Are you going to pass it to your kids? Are you aiming to sell it? Or is this a lifestyle business? And you're kind of setting up your job as a business for the long term. But his question for you is, what's your biggest pain point in your business? What is the biggest obstacle that's preventing you from getting to where you're trying to go?

I think the biggest thing, and this is actually one of the reasons I'm using some of these AI tools, is managing people and having it truly understand how things are working. Like people — ultimately, AI is great, but there's still some level of human element that are going into these businesses. Cause I have like a team of blog writers and we have people that are consistently working in the SEO, we can use AI and get it to help us and do all of the fun stuff with that. But managing those people and keeping everything up to date and constantly up to date with what are all the tasks that we have right now? What are all the things we could be working on? Depending on how many people you work with, it becomes more and more difficult at scale to make sure everybody kind of understands exactly what priorities are and where they should be on certain things.

And I'm using AI and some of the meeting recorder devices like Plaud so that now I just talk to it on the back of my phone and it has a general context of what things should be worked on, how important certain things are. It's just continued to be trained and then it assigns tasks with priorities and tries to manage it that way. It's very difficult for a human to remember — in one specific stretch you'll have like two days I had 60 tasks that I got to get done. To try and task switch, task switching is terrible for your brain. Humans do it at an exorbitant rate and there's a lot of psychology on this that task switching will inevitably just hurt you so bad and you have to manage people at that some point.

So I strongly recommend people if you were to find one use case for it sometime soon, get it to manage some of your project management stuff. That is a huge help and it'll understand it because it can just task switch better than we can. So even if it forgets 5%, we forget at a much higher rate. So it's worth it long term for a lot of people to set it up as like a project manager. That's going to be the best use case for everybody. If you have a team and you're managing a team, please set this up and use it in that way. I think anybody that does this, they will have much greater success than trying to give AI every single thing that you do in your business. If you just give AI every single thing that you do in your business, you're going to have a hard time. But managing people, this is where I'm using it and it is challenging. It's still challenging even with all the help, right?

I have a question because, you know, Claude — obviously, any particular chat that you're working on has a memory limit. In your experience so far, what are the platforms now that are kind of structured to help you load in your institutional knowledge?

This is something several previous guests — Matt Brooks of SEOteric talked about it. I know Mike Kalinka of Kalinka Group could use it to help develop business plans for his financial planning process— But the question is how to create more long-term memory within your AI project so that you're having outputs, because we talked about the other side of the problem. Like if you're using the generic pool, then you're going to get generic results. So obviously, the flip side of that is then I need to load in my stuff. But if any given chat can only remember so much stuff at a time, what's the best way to build out a sustainable long-term database — like a clean-up interview with the founder so that you've got the mission statement, the description of the service product or whatever. What's your two or three products that you recommend or processes you've been using to kind of hold that institutional knowledge for use in AI?

So one of the ways that I'm doing that, I've mentioned it, it's Obsidian. So it's a Markdown system. How familiar are you with Markdown? I know this is somewhat familiar to some people.

I would say I'm familiar with it, but for the audience's sake, let's not assume they know what that means.

Markdown is just a different way to store text files that will ensure that long-term everyone can have some consistency in the way that text is written. The way that we do it is we do it with Obsidian. Obsidian is a markdown platform that manages all of our files. So we make logs of transcripts. We make logs of every single thing that we're working on and that we're doing. So if you're ever getting on a meeting or you're ever getting on any specific thing, or even if you're just talking into your phone, it manages that and saves that to a markdown file. So all of my conversations are recorded. And they're kept in my markdown file. So that is the way that I manage it — it does a database search to go and find the specific relevant pieces and they all connect together.

So that is one thing and then as the AI does stuff — people have not always done this — you have to tell it to leave clear traces of everything it does. Cause AI a lot of times will just go. It just goes and then doesn't leave a good trace on how it got to the perfect destination. So repeatability is scalability. And that's the number one thing people are forgetting to realize. They're doing something and they're doing it once. And even if you get the perfect output from like ChatGPT online, I would still recommend using that in an AI agent because you can repeat that. The problem is right now if we get the perfect output from like some sort of AI model and we can't repeat that, you're just kind of battling that — chasing that one high that you got that one time you did one thing awesome. It made that perfect blog or whatever article for you. But long term that doesn't actually help you because it can never repeat that again.

So make it actually write down the things that it's doing — it can repeat those same steps. If you always take the same, like if you're going that destination, imagine point A to point B, you tell it, hey, go from point A to point B. If it wrote down it took a left turn, it took a right turn, it took a left turn, it took a right turn, it'll get to that same destination in roughly the same amount of time and roughly the same steps. But people are just saying go, and they don't watch the driver. They don't see where it goes. And then it takes a right turn one time and a left turn the other time. And then gets to the destination and you're like this is completely different than what I had before — well it took a different route and you didn't tell it to keep and maintain that.

So the markdown system is just a text file system where H1 they're saved with a pound, H2 are saved with two pounds and then H3 are saved with three pounds — generally that's enough. And then you ask it to just trace all of its exact steps on how it got to that final output. And you'll get something relatively consistent over and over and over again. You just don't have to try and keep monitoring it, but you have to feed it back and be on top of it too. It's not as set it and forget as people like to assume it is. It really isn't. No matter how awesome it sounds that people are saying, yeah, I set it up that one time and I've never touched it again. It really hasn't been that way for me.

Now I can quickly see for like a digital marketing business, like it's a very short hop, skip and a jump to, hey, I've been making blog content, I've been making marketing material and you can distribute it here and there. What's more of the thought process around like a hard business? Like you're delivering concrete walls to get put up in the real world to protect data centers. You're doing that thing in the real world. What are your, what's your low hanging roots of going into that organization and talking with the business owner, where's the soft underbelly that you're saying, hey, this thing, this is a very quick one, increase your efficiency to do this very simply.

Automation is another thing people aren't taking advantage of. So not everything has to be AI. Like it just as simple as that, like a lot of the certain business tasks, I ended up shifting from just being used in AI to automations at the end. If there's a super repeatable task that always works in a certain way, you can, if you have your server that's running, OpenClaw or running some sort of agent set up in it — it's one of the platforms like Zapier or Make, it's just a connector of tools. You can set it up locally on that computer where you're already running your agents and then you use that as automation. Like a lot of these people have some sort of repeated task — whenever they get an email by this person they forward it, or something along those lines, whenever you get a file uploaded to this specific Google Drive send out an email and do certain things. But there's a lot of repeated task and effort and the more you can minimize the amount of time you're task switching and like juggling between too many different things. If you can use your creative mind and post really nice content and do certain things, I think that's going to ultimately be better for people. But if there's a repeated task that you have to keep doing, it doesn't have to be AI.

Everyone's putting AI in everything. AI is not for everything. Make AI, make automations and that's how you win. Cause no matter what we say and what we do, I really have a hard time believing that everyone has it figured out that they never have to manage any of the outputs that they're getting by AI. For everything you see on Twitter in X, everyone's got it figured out, they've never had any problems, it always is perfect every single time, and I failed to realize that is the reality.

Let's see, I'm going to give you a soapbox to stand on and shout at the people about what do people need to do with AI that they're just not adopting, not understanding about.

What people aren't using is for certain, repeated tasks in the day to day business. And I'll give you some examples. So the main one that I'm telling people to use AI for is if you're doing a presentation and you want to use PowerPoint or you want to use slides, you can use it for this. Ask it to make a website. Ask it to make a self-contained HTML website that performs like a PowerPoint. And ask it to style it how you want it, add images, but that's how you get it to make a PowerPoint. So it's really good at making code. So that goes and makes your PowerPoint. That's something that people are doing all the time. They're making pitch decks, they're making everything else. It can make a really, really nice looking PowerPoint or slide for your presentation or if you're doing a meeting.

And I've used it literally just yesterday. We were performing some SEO statistics. We used it there. I spent 10 minutes on it. I gave it a prompt.

I said, I wanted it dark themes, Scandinavian minimalist. I gave it all the data files and you can just export it from your Google Search Console. I said, hey, make a report. How did everything go? Then it made that and made like a nice little website. All you do is you just click through it with your key and it goes as well as like Excel.

So another project that I'm working on right now is called ZapTime.ai. It's a time tracking and monitoring. Excel is so big for like accountants, lawyers, or really anybody that manages large amounts of data. You know we definitely do. We're looking at like SERP analysis for certain things or, if you're just in Excel often and you're trying to make formulas, please, please for the love of God use AI to make your formulas. Like there's a lot of accountants and people that I've talked to. And the reason that we're starting this new project called ZapTime.ai is to help people manage their time and see where they can use AI in their life. It's something new that's coming out. I haven't posted it very much publicly, but that is one of the big things. Like if you're using Excel ever, please use your functions with AI. That's one of the big ones, but like some of those Microsoft suite things that you're doing on a daily basis, you're making PowerPoints, you're doing Excel. All of those can be done with AI and people aren't using that.

I love it. Give a shout out where people can connect with you, what you want them to know that you're working on next. It sounds like you've got that specific program. Is there anything else you're putting out — a book, if you've got a video series — how can people connect and find you, Mason?

So people can find me out on LinkedIn. There's a lot of stuff coming actually right now. We're hosting, here in Texas, one of the largest meetings for OpenClaw and entrepreneurship is happening. So people are coming all the way from colleges in Houston, as well as like general people on May 2nd. If you're anywhere inside of Texas and you have the ability to make it out, there's going to be a meeting with hundreds of people all going over OpenClaw, getting it set up and kind of using it for certain applications and certain uses. And that's all posted out on my LinkedIn.

Just LinkedIn and then the username is Mason McCumber, all one — M-A-S-O-N-M-C-C-U-M-B-E-R. So you can check out that. I'm always posting consistently on LinkedIn. I try and get to the blogs whenever I can, but working with clients, it's always on the back burner, but I'm always trying to stay up to date and post stuff on LinkedIn. This is one good place to find me and there'll be more updates about ZapTime and how people can use that coming out here soon. And then all of the resources that I was talking about for AI tools and certain things that are coming out, you can find that on our website and our blog at buildwithmm.com.