Books That Rule Podcast

Dane Golden: It is time for the Books That Rule podcast. My name is Dane Golden, your host, and I'm here with a special guest, Alberto Fernandez, and he's the author of CoThink This is a Book, how to Do Better with AI. Welcome, Alberto.

Alberto Fernandez: Thank you for having me.

Dane Golden: So, anyway, I wanted to ask you, why did you write CoThink what made you say we need to write this in a book?

Alberto Fernandez: absolutely. So I noticed that people, today use AI like a, toy. It all, it's almost like a bonafide search engine and, that produces no outcomes. CoThink is a different methodology that I actually patent in patented. And it, allows people to leverage the knowledge of AI in a different way by giving AI's persona rather than prompting and by, using different AIs for different things.

This could be applied for software creation. This could be applied for content creation. This could be applied for anything. And the basis of the methodology is simple. You have an orchestrating AI. That's the main persona you have. For instance, if, we were creating a software, you will have a coding AI and then you will have a QA AI and the coordinating AI will prompt those other AIs to actually do the work.

Dane Golden: Okay. So, going to have a variety of people on here. Some are really going to understand AI and some aren't. But let's try to decode as what you're saying as we go along. So, first of all, first thing I heard you saying is we're doing AI wrong, we're doing it wrong. We're we.

Alberto Fernandez: we're using it wrong. We're using it like a toy or like a search engine, and that's not going to give outcomes or value.

Dane Golden: It's not delivering value. And then what I heard you say, and this as people who work in tech, they understand this language, is that you had one AI you're going to assign to development and one to QA. So in a, development team, you have different, entire people usually assigned to doing the development, making, the product.

And then you have a quality assurance, a QA team that says. Guys, you thought this was working and it's not. And this is the normal process and it's frankly the normal process we do in creating any product, whether it's shoes or buildings or whatever. You have some check, and in this case it's called the QA, and tell us more.

Alberto Fernandez: Absolutely. So. one of the things, one of the recent projects that I had was, a security platform, that, basically manages, assets, an inventory of assets, and I was able to create this platform, within three days, which, Software engineers would've taken, you know, six to eight months to do by simply applying the methodology that I explained.

So I give the task to the coordinating AI in this case, in, in that particular case, I use ChatGPT as my coordinating AI, because it's the best at understanding.

Dane Golden: coordinating our AI. Are we going to say that's the project manager?

Alberto Fernandez: that is the project manager and that AI is in charge of giving the other AIs prompt. So I'm not prompting for the actual work, the AIs, that AI understands the business outcome that I want, and then that AI is going to be in charge of telling, Claude, which is another, large language model to do the code.

And it

Dane Golden: your

Alberto Fernandez: the prompt for Claude, and that's my developer.

Dane Golden: You've hired Claude to be your developer.

Alberto Fernandez: and then I have Gemini doing QA of the code that Claude outputs. But the coordinating AI is the one prompting. Gemini saying, Hey, go ahead and look at what Claudeoutput. And then we circle back around with the output of Gemini, we re-prompt Claude to continue doing the functions if it found something wrong.

So in that loop, I am able to execute what will take software engineers months to do. in days, like I mentioned, I created a new platform, for security and asset, management in, in three days.

Dane Golden: type of, platform or framework or computer language or development language were you using or was it

Alberto Fernandez: IUI used, I use Python because that's where I'm more comfortable with, and that's where the human can go back in and, check. But it could be whatever language the user is, most, familiar with. and I highly recommend using something where an actual human can verify, validate, debug, But, the methodology is simple.

and by the way, I'm giving the example of software engineering, but the same could be true on content creation. You could have, you know, a script writer, you could have, a producer and you could have different personas to create content or create, short, short, videos

Dane Golden: design? How about design? Nike shoes.

Alberto Fernandez: Absolutely the same, principle applies, and you know, each, each, AI has its strength and it has its weakness. I use ChatGPT as a coordinator in this example that I gave you because ChatGPT is more conversational. It understands intent, it understands, what a desired outcome would be, and it knows how to prompt the other AIs.

Lot for coding because Claude is the best at creating code. And I use Gemini at criticizing because Gemini has the library of Google to really understand what is good code and is what's not good code. so

yeah.

Dane Golden: Alberto Fernandez are we, so when we're saying. Using another AI. What, is that the language that people are using or are they saying use another model or use another platform? Or what is the, when we say, what do we mean, what should we be saying?

Alberto Fernandez: E even, if I was to, let's say I only have money for a subscription in one AI, I would still use the same principle. For its own prompting of having different, context windows. So my context window for planning and being the project manager is going to be different than the context window that we are going to have when creating code, and that's going to give us the ability to move much faster.

So have the AI prompt itself. So let's say I didn't have money to subscribe to three different AIs and I only had one. The principles still, holds a lot of value, and that's where AI fails most of the time because you have hallucinations, you have made up things, you have, a lot that could go wrong when you are.

Doing a project that is huge on a single context window of an AI, and then that runs out and you have to start all over again. And it's like amnesia. The AI, forgets everything you did and you start from fresh. this way we keep continuity even when that continuity ceases to exist. So you can start fresh,

Dane Golden: now, thethink, book that you've written. By Alberto Fernandez, that's you. You've written, you, you, gave me some principles that, that it, that people should follow and one is that AI fails when the user works without structure. So I'd like to follow some of the principles that you've talked about and explain why those are important.

So I have them in front of me. I don't know if you have them in front of you.

Alberto Fernandez: I, don't, but

Dane Golden: Okay. I'm going to prompt you. I'm going to prompt you then. This is my prompt. You're my AI. I'm prompting you. Number one, AI feels when the user works without structure. What does that mean? Mm-hmm.

Alberto Fernandez: So when you give AI ambiguity, AI does what, we call hallucination in, the AI world, which is, it makes things up, it fills in the blank with. Stuff. And it does it in such a elegant way that you actually believe It it, it's very convincing at lying to you. So you start having hallucinations.

Number two, you start creating garbage code that doesn't work. For instance, in, in software development, or, Outcomes that are not desired. And then if you have that monolithic, window of, context, you start getting a lot more garbage because of those hallucinations. So you have to use AI in small burst.

That

Dane Golden: Well, I'll tell you.

Alberto Fernandez: outcome is.

Dane Golden: I'll tell you, I ran into this problem because, I was using ChatGPT. I'm not an expert. I was using ChatGPT to modify code in the Google ads, YouTube targeting, scripting, structure, and. no, almost nothing about how to structure that, but I knew what I wanted. I don't know if what I called was really vibe coding or whatever you want to call it, but I was trying to get to a very quirky result, which I was having difficulty with, and I did not have any structure at all.

just some questions.

Alberto Fernandez: Just, transactional prompting and, that's the difference between structure and transactional prompting. In transactional prompting, you know what, you may get brilliant, outcomes, but that is not the norm. Transactional prompting where you ask a question, it gives you an answer and you continue, and, that sort will deal.

Very, rarely good results. What I'm saying is, instead of asking ChatGPT to give you, you know, that, that snippet of code for YouTube Ads, or Google ads, I would have asked ChatGPT to give me a prompt. To have another AI give me that outcome. So I would tell ChatGPT, okay, I want this outcome in Google Ads and I want to do it in a way that is organized and chat.

GPT will give you a prompt that is targeted and you can even tell it, Hey, do not hallucinate. I want this outcome. And

Dane Golden: Now, let me stop there. Let me stop you there. So number two, you say prompt engineering isn't enough. I think we're moving on to that. that what we're talking about here? Prompt engineering isn't enough.

Alberto Fernandez: yeah, absolutely. prompt engineering is the art, because I consider it an art. The art of being able to know how to ask AI things, but that's not enough. So in this example that was walking you through, I am prompting the AI to create a prompt for another AI.

Dane Golden: Your prompting prompts.

Alberto Fernandez: Yeah. And now I'm creating a prompt that is going to surpass what a human could, create as a prompt.

And that is where the secret sauce comes in, because that's where now the outcome that you desired will be attained versus going back and forward 300 times with a single chat.

Dane Golden: And, said number three. AI performs best when each model has a job. Now, by model do you mean, are you defining Claude as a different model than Gemini or what?

Alberto Fernandez: Yeah. Or it, could be Claude or it could be, you know, Claude Sonnet

Dane Golden: when we

Alberto Fernandez: it could be

Dane Golden: when we say

Alberto Fernandez: is the model,

Dane Golden: we mean Gemini is a model ChatGPT is a model.

Alberto Fernandez: So Gemini is an AI engine, A model is Gemini three that just came out. ChatGPT, has ChatGPT 5.1 that just came out, this month, but it has 5.0 and 4.0,

Dane Golden: model is like a version.

Alberto Fernandez: Yeah, it's like a version, within the large language models.

Dane Golden: Okay. And so why does it, why does AI perform best when each model has a job? I, I say a different job. I is what we're saying.

Alberto Fernandez: Yeah. because now you remove the ambiguity where, you know, the hallucinations in AI happen. It's usually that ambiguity. If you make AI's world small. It becomes predictable. I'm a father of five. I have five kids. some of my kids have special needs and, I'm using with AI the same strategies that I use with my kids, which is keep things targeted, keep things small.

And then you get the desired outcome. If I tell my son with autism to go off and do something, I am not, and it's ambiguous, I'm not going to get any results. But if I tell him specific instructions, step one, go to the kitchen and grab a butter knife. Step two, go grab the butter. Step three, go get the bread and put it on the butter.

I get better results than if I say, Hey, go to the kitchen and grab a snack. So I use the same principle. With AI that I do in my personal life, essentially.

Dane Golden: and so, so when we say, you said here is breaking the roles into strategist, researcher, analyst, critic, fixer, teacher, executor. Those are all different jobs that we might have in an organization, and you're creating a different, different, using a different model for each.

Alberto Fernandez: Absolutely. and like I said, each model has its strength and its weakness. I did a i, I did a LinkedIn post earlier this week that outlined what I believe my personal beliefs, because I don't hold the whole truth, but my personal beliefs of what. Model it has what strength. and it's been a great read.

I've had a lot of good feedback from a lot of peers. but, each model has its strength. Each model has its weaknesses. And if you play to those strengths and weaknesses and assign those tasks to, to each of those models, you're going to get better outcome than you would otherwise.

Dane Golden: How do you know, for instance, what is Claude good at that you happen to know that Gemini is not that ChatGPT is.

Alberto Fernandez: I would say Claude is best for creating code and for, like if you're a mathematician and are doing mathematical equations, Claude is the best at, strategic thinking at creating code, at doing, engineering task, if you will. where I, believe ChatGPT is very conversational ChatGPT sometimes almost feels too real and too human.

And then Gemini is fantastic at having data. Gemini has an amazing. context window that is just changing the whole industry. So I, when I am, approaching a project, I'm looking at my stock and I'm saying, okay, what fits well for what task? for instance, I will give Gemini always the teacher, documenter, creating documentation, creating, creating criti critiques or, QAing software like I did in that example. And I will give Claude the more tactical, jobs such as, software engineering, and content creation. Claude will be my producer. and, many other tasks, cla will be that strategic aspect, and ChatGPT is. For me, as always, my coordinator because I feel that it can have conversational, and, understand sentiment of what you're going for and what the desired outcome is.

Dane Golden: And we're, talking to Alberto Fernandez, he's the author of CoThink and we're talking about how he uses different AI models for different tasks and we're going to tease that. CoThink is also a. You've launched and I want you to tell me a little bit further on that people can log into, correct.

I'm so far, I'm

correct.

Alberto Fernandez: no, you're correct.

Dane Golden: talk about that, but want to get to your other points. Number four, drift is real and constant. What do we mean by drift?

Alberto Fernandez: a again, drift is those hallucinations is where you're asking the AI to do something and is, you know, out there doing something absolutely different or, making things up. and the, dangers part with hallucinations in drift is, It sounds convincing. It sounds like it. It's real, it sounds factual, and, it makes you feel like, okay, we're, good here.

And it's not, and that, that poses a, level of danger if you're blindly trusting AI. I saw an article, this week of a, front page newspaper in Pakistan that just copied and paste from AI, and they even copied the prompt that says, Hey, would you like me to make this? Snappier and make it nicer for you.

and that is unfortunate, the side effect of having the smart large language models, which is people stop thinking and they, start just copying and pasting blindly. And, that could lead to disastrous outcomes if, we don't become smarter. so

notice I'm not

saying.

Dane Golden: think, that leads you to number five, validation matters more than generation. And I'm going to jump in here real quick and say, there's a saying if you write novels that great novels are never written, they're only rewritten. And I think that's what you're getting out here, is that no matter what you're doing, checking your work, retrying, redoing is extremely important.

Alberto Fernandez: absolutely. And that's why again, in my model or in the example I gave you when I filled out your form, I talked about, hey, you have a teacher that is an AI that is teaching and learning and documenting. You have a QA or a critique that is the one that is critiquing and QA and you don't want a first pass, from any output that an AI has to go out.

To the world. it could lead to disastrous outcome. And in, in this particular one, in Pakistan, it was just funny because it was in the front, in, the front page of an English newspaper that a lot of people read. But it could, have disastrous outcomes if, you know, in, in other industries, if people just blindly follow ai.

So. Again, as smart asis are having that, validation and you could do it with AI itself, having that validation and then that human validation that the outcome that you wanted is the outcome you got, is, just necessary.

Dane Golden: Now a, number six, AI works best with clarity, boundaries and constraints. What does that mean?

Alberto Fernandez: Well, that's the example, that I gave you earlier, of if you give AI ambiguity, it's going to, give you garbage. It's not going to perform. and, again, being a father of, kids with special needs, I, use that every day of my life. And, that's how I landed here. That's, you know, that's what got me to, got me to, thinking of this method because I noticed that AI first has amnesia, one prompt and a new chat.

Forgets about the other. second, it, if you give it ambiguous instructions, it gives you garbage back. Third, if you do not, clarify or give it clarity on what the desired outcome is, you are going to not get that desired outcome. So being very structured and organized in the way that you interact with AI is paramount.

Dane Golden: Seven, the future belongs to people who orchestrate, not just prompt. What does that mean?

Alberto Fernandez: That means that, I've had a lot of friends, being afraid that AI is taking their jobs and I don't think AI is taking anybody's job. I think AI is a, a. Augmentation of humans and it augments us whether it's good or bad. If, you're a smart person, it will augment your smartness. If you're, a not so smart person, it will augment your not so smartness.

I'm trying to be as, as as possible. because I don't know, I don't know. your podcast and

Dane Golden: You don't know who I am. You don't know what I've done. You don't know how smart I am or if I'm smart at all. I.

Alberto Fernandez: So, but that's the reality. It's just AI is going to put a, flashlight on top of who we are as humans, and it's going to highlight our greatness. or how not great we are. and people that learn how to operate with that level of knowledge that we're never going to have in our heads. I'm never going to be able to write software in 10 different languages.

AI can, I am never going to have all the knowledge it does. AI does. So I don't want to compete with AI. I want to leverage it to augment what I know. and that's what, you know what I meant by that point.

Dane Golden: And, asked you what the top features and benefits of CoThink your service is, and you said a predictable system. So tell me about that number one.

Alberto Fernandez: yeah, absolutely. So, I created CoThink as a service. So the, book talks about the methodology, not the service, the

Dane Golden: Hold that right up there to the screen. Hold that right up. Close to the camera. Closer. Closer, yeah. Good.

Alberto Fernandez: we go.

Dane Golden: Thank you so much.

Alberto Fernandez: So, no problem. So the book talks about the methodology, not the service. And I want to split those two things into, separate realms because it's important.

The service that I created is just a way to have a single place where you can orchestrate that, where you can log on. Put an API key for Claude, put an API key for ChatGPT, put an API key for Gemini, and automatically have those interactions happen, without you having to jump from one to another. So it's just the service that I created in coding is' just a way of making it easier for, people to leverage the methodology, but there's no secret sauce.

You could, have the same outcome by simply just using the method and doing it yourself. So, I really haven't pushed it too much because I wanted this to be more of a methodology than, the service itself. But the service is just a facilitator.

Dane Golden: tell me what role, tell me how role-based orchestration improves accuracy.

Alberto Fernandez: a again, it, if you give AI very explicit, instructions, you get good outcome. If I tell Claude, I want you to create. A piece of software that tells me when I'm going to die. It's going to create hallucinations if I tell it, create a piece of software that based on data inputs and, analyzing normal, life cycle of humans.

And, these are the datas that you're going to input. And, you know, you're going to understand if they're smokers, their health history and all of that, then you can give a, give a outcome of, you know, a person's longevity. Then we will get better outcomes. So it's all about how you frame it. And when you give small targeted instructions to AI, you get good results.

When you give long, big instructions, you. Are prone to hallucination. So that's the real rule of thumb here.

Dane Golden: And Alberto Fernandez, author of CoThink, which is a practical method and framework for working with AI with consistency, structure, and accuracy. I always like to ask this question, what is the question I should have asked but didn't.

Alberto Fernandez: To be quite honest, you have been very thorough, about, you've been very, you've been very thorough. I think, the, biggest question that you asked is what makes this different than anybody just using ai? And the answer is simple. it's not, it's just a different methodology on how to orchestrate with ai.

Dane Golden: And his name is Alberto Fernandez. He's the author of CoThink I know it's on Amazon. Where else can people find it? And your service.

Alberto Fernandez: It's only in Amazon right now. I, again, this is my first time venturing to write a book. and, my wife has pointed out that I have misspellings and, grammar issues. I tried for the sake of authenticity not to use AI to write it. So actually it was written 100% AI free, which is ironic.

It's an AI book. So, this is my first time publishing a book. I decided to do it after I had so many coworkers, friends tell me, Hey, how did you get to that outcome? and did that so easily when, you know, we struggled for months to accomplish the same thing. And, had a friend who reached out to me and said, Hey, can you just write it in a white paper and give it to me?

And it started with me writing a white paper to a friend and then I'm like, let me just turn it into a book because I think it'll be beneficial for other people out there to understand how I'm doing this. because it might be different. May not always be the best method, and everybody has their own methodology.

But I decided to turn that white paper into a book. I started writing it, about three months ago. I recently became unemployed and had the ability to finish it. And, you know, I was doing consulting and I had a couple of, Consulting jobs that I was doing, but I was able to set my own time and I was able to sit down and finally finish it.

and I took advantage of that time. So, it started with a friend asking me to write a white paper to him because he was intrigued in how I got some outcomes that he couldn't get.

Dane Golden: Fantastic. And it's Alberto Fernandez is his name, and, on LinkedIn he is Alberto Fernandez one and the, bookend Service CoThink Pro, but also CoThink is on Amazon by Alberto Fernandez. Alberto, I am very grateful for you being on the Books That Rule Podcast with me, Dane Golden. Thank you.

Alberto Fernandez: Dang. Thank you very much for having me, and it's been a pleasure. It's been amazingly fun. I, hope to do this again, at some time. when I create CoThink version two.

Dane Golden: I can't wait for CoThink version two. My name is Dane Golden. This has been the Books That Rule podcast. We'll see you next week.

​