1111: How to Get Better Results from AI to Amplify Your Productivity with Gianluca Mauro

Pete Mockaitis
Gianluca, welcome!

Gianluca Mauro
Hey, thank you for having me.

Pete Mockaitis
Well, I’m excited to be chatting with you about AI. You are a genuine expert. You’ve been researching and studying this stuff way before even normal folks had heard of this ChatGPT business. So great to have you. And tell us, any super surprising discoveries you’ve made along the way as you’re researching and teaching this stuff?

Gianluca Mauro
Well, first of all, I think something that is interesting to think about is when ChatGPT came out three years ago, it was the “Oh, my God” moment for most people, right? But AI has been out there for quite some time in different shapes and forms and with different levels of usefulness, let’s say. And I think the first “Oh, my God” moment for me was when I realized that, basically, every industry and every professional could find a use for AI.

And I’ll tell you probably what was the most interesting, or strangest maybe, project I worked on. I worked on an AI project to control the quality of diapers in a factory. So, yes, you can use AI for pretty much everything.

Pete Mockaitis
Well, now I just can’t let that go. How does AI help do quality control for diapers?

Gianluca Mauro
Well, so are you ready to go on a journey on how a diaper factory production unit works?

Pete Mockaitis
I imagine AI might be able to analyze rapid photographic imagery of diapers as they come off of the line to assess quickly potential for defects and fix the issue more quickly upstream prior to them being packaged and having to be thrown away. But I’m totally making that up.

Gianluca Mauro
You got it.

Pete Mockaitis
Oh, I feel like a genius. Yes!

Gianluca Mauro
Oh, my God, you are. This is extremely accurate. Extremely accurate. We had this issue that, you know, they basically have, a diaper is basically two layers of elastic material with something that is absorbing in the middle. And then if you pull this elastic material too much, it breaks, especially when you’re cutting it into shape.

Pete Mockaitis
Been there.

Gianluca Mauro
Yeah, exactly. So, if you had kids, you know that that’s not fun. So, we were looking at all these pictures in the factory as they were cut into shape to try to understand, well, what was the ideal size of those big elastic rolls and try to basically optimize productivity. So, that was quite a crazy moment because, think about this.

I did this project maybe seven or eight years ago, so three or four years before ChatGPT came out. It was not obvious for anybody or for any company that they might have a use case for AI.

So, imagine me when I went and pitched a diaper production company, “Hey, maybe you should look into AI to minimize the mistakes, the defects that come out of your factory.” It was not obvious at all. It was quite interesting to find actually amazing use cases in that context as well.

Pete Mockaitis
Well, that’s very, very intriguing. Well, so we’re talking about everyday professionals utilizing ChatGPT or other AI tools to be more productive. You’ve got a LinkedIn Learning Course on exactly that. So that’s pretty handy. Could you maybe start us off by sharing what’s perhaps a fundamental misconception or mindset shift that helps make all of this stuff make sense?

Because I imagine we could spend all day talking about, “Oh, here’s a really cool prompt,” or, “Oh, here’s a fun little tactic,” “Here’s a nifty little thing you might try.” But could you maybe set the stage for us on a more principled foundational level to help us scaffold the rest?

Gianluca Mauro
Absolutely. And I think the most important thing for everybody listening, you need to understand that, in order to really get value from AI, the number one thing you should focus on is your mindset and changing your habits. This is not anymore about necessarily getting the right tools. Most tools are pretty good today, not perfect, but, you know, they’re pretty solid, especially compared to three years ago. And it’s not even about having the most amazing prompting skills.

The biggest bottleneck is your habits. How much have you embedded AI tools and new different workflows and ways of working in your day-to-day work?

I’ll give you an example, I love making this metaphor. It’s like going to the gym. So, let’s say that you have the best equipment. That’s the equivalent of having the best tools. And let’s say you also have amazing skills. You have a squat with perfect technique and you know exactly how to do a really good bench press. And that’s the equivalent to having really good prompting skills.

But then let’s say you never go to the gym. Guess what? Your muscles ain’t going to grow. You’re not going to lose the weight that you want to lose. That’s not going to happen. I would rather see somebody with okay tool selection and with okay prompting skills, but, really, somebody who’s invested a lot in rethinking the way that you work and is curious and is constantly trying new things out than having somebody who has read all those scientific publications about best prompting techniques and has bought all the AI tools, but then has not adapted the way that you work to work with these tools.

That’s the most important thing today in this context. I wouldn’t have said that three years ago, but that’s where we are today. You need to really change the way that you work and embed them in your workflow. And that requires a little bit of effort.

Pete Mockaitis
I hear you. It does require a little bit of effort. And I would also say some discernment, because I think that my impression is, and you can tell me if this is accurate or not from your research-based perspective, it almost feels like a lot of companies, CEOs, products, just kind of want to shove AI into something because investors want it, the stock market seems to like it, and maybe some people are impressed.

But I’m almost at the point now, when I see a tool say, “Oh, now we have AI,” I’m like, “Oh, geez. Is it any good or is it just going to disappoint me again like all the rest, you know?” And so, that’s my take is that, yes, we should take a look at our habits and get into the groove of using AI tools where they’re genuinely helpful and useful and handy. And that requires a little bit of change management on our own parts.

But my hunch is there are also times where you say, “No, AI has actually no place in this little piece whatsoever, and so we’re going to deliberately choose to not stick it here but instead put it over there.”

Gianluca Mauro
You’re spot on. And there was actually research about this that I found really interesting. It was done by Stanford with a couple of other people, and then the Harvard Business Review wrote an article about this that went quite viral. The title of the article is “AI-Generated ‘Workslop’ Is Destroying Productivity.”

Pete Mockaitis
That’ll get some clicks.

Gianluca Mauro
That’s going to get some clicks. Exactly. And so, the main outcome of this research was that if you ask people, “Hey, what do you think about your colleagues who use AI?” You’re going to find that colleagues who use AI are often perceived as less creative, less capable, less reliable, less trustworthy and less intelligent. And that is not great.

Pete Mockaitis
Yeah, fair.

Gianluca Mauro
You do not want to be perceived as less intelligent, trustworthy, reliable, capable and creative. So, the interesting thing in this case was I honestly don’t think, and that’s also what the researchers found, that that’s a problem of AI per se. The problem is that a lot of people are using AI just in the wrong way. What does that mean in practice? Well, AI workslop is basically when you are trying to use AI as an amplifier of your laziness, basically.

Pete Mockaitis
Ooh, tweet that!

Gianluca Mauro
Yeah, and I want to give a practical example, okay? Let’s say that you ask me, “Hey, Gianluca, I want to know how I might use AI in my podcast,” okay? And let’s say that I am just so lazy, I don’t want to think about what are your challenges. I don’t want to think about what are your objectives. I don’t want to think about your audience. I just go on ChatGPT and I ask, “Hey, how might a podcast producer or host use AI?” I get a research. I copy it. I send it to you. What happened?

I got a generic piece of, like a bunch of text basically, on a PDF. I gave it to you and it took me no time to produce that. It took me, like, 30 seconds to get like a bunch of text that sort of makes sense. But I’m going to waste your time reading something that is so generic that you could have found on one Google Search. So that is something that damages you because you just wasted your time reading the report that is generic and has wasted my time as well, because now you’re going to ask me questions I need to go and fix it and you’re going to think less of me, etc.

Now let’s see what I should have done if I wanted to use AI to make it way, way better, way more interesting. I would have started asking you questions, “Hey, Pete, what are the top challenges that you have? What are your objectives for next year? What do you think could be the thing that helps you the most? Do you want to be more productive? Do you want to repurpose your content more effectively? Do you want to be able to research your guests better? Like, just tell me, tell me what’s going on.”

You provide me some context. Context is a keyword that is super important in today’s AI era. You give me some context. Then with this context, I go on ChatGPT, and I say, “Hey, I interviewed Pete. These are his top challenges. What do you think might be a relevant use of AI?” Now start getting something interesting. I start getting something that is more relevant.

And then I might say, “Okay, cool, ChatGPT. Now go and find top case studies of similar podcasts to How to be Awesome at Your Job that have done something similar. Now find some tools. Now tell me what could be potential risks.” The output, then, that I send you is going to be much higher quality and it’s going to actually give you value.

But notice how the difference is not the tools. It’s not that I used a different tool that is not ChatGPT, or is that I had some special prompting skills. It’s just that I’ve been mindful. I’ve been mindful of what might be interesting, what might be relevant for Pete, and how might I use ChatGPT to basically boost my productivity and make my suggestions for Pete even more and more relevant and useful.

You see the difference. It’s not about the tool. It’s not about how good am I in prompting. I didn’t talk about doing anything particularly fancy here, okay? It’s not fancy prompting technique, there’s no coding involved, it’s just a different mindset. I tried to use AI to amplify what I would have done if I didn’t have AI. And that really works.

Pete Mockaitis
Yes, it really does. And what you’re reminding me, and you’re talking about amplifying your laziness. I’m thinking about there was a fabulous interview on The Copywriter Club Podcast, which I listen to, even though I’m not a professional copywriter, but we’re doing copywriting all the time. And there was a famed copywriter on there. We’ll look him up and put him in the show notes.

And he said, “When I’m using AI to assist me with copywriting, I don’t say, ‘Write me a sales letter.’” It’s like, “What I do is…” well, first of all, he’s using the custom APIs of an AI tool as opposed to any off-the-shelf chatbot. And then he’s saying, “Okay, I’m going to write this part of a sales letter, given all of these instructions that I have previously written for what I’m into, as well as several examples, as well as what the product is and how it’s helpful to a certain user base on these needs and want and preferences and desires and pain points. And then, so voila.”

And so, there are numerous multi hundred-word prompts associated with doing a thing. And then he was like, “Okay, this is a pretty good draft. And from that I can tweak.” And so, we’re not amplifying laziness. In fact, a tremendous amount of thought has gone into what we’re doing here. And then, because he’s done it many, many times, and he also said AI does not account for taste.

Gianluca Mauro
Yeah, exactly.

Pete Mockaitis
And then from there, you can get it. And that was a real lightbulb for me, which I’m connecting now with your amplifying laziness comment. It’s like, yeah, if you just say, “Hey, do this thing,” you’re going to be disappointed. But if you put a ton of thought into it, it can kind of get you to a draft substantially faster.

Gianluca Mauro
Absolutely. And I’ll tell you what, you can amplify laziness, but you can also amplify your expertise. You can also amplify your perfectionism, if you’re a perfectionist like me. And I will give everybody a very simple thing that they can try right now. So, I’ll give you a simple prompt structure that you can use. And it’s very simple, okay? Just four lines.

So, start with some context. Context is basically the who, the why, and the what. So, you might say, “Hey, I am a podcast host. I need to…” whatever, “…prepare for a new interview with this person. My objective is to make sure that I ask the most interesting questions about this person.” Okay, that’s context. What are we talking about? I’m assuming I’m putting myself in your shoes, by the way.

Okay, so that’s the context. Then you say, you ask AI, “I will give you, for instance, a list of questions I prepared.” Something you’ve done, okay? Something that, you know, maybe 50% effort, something that is almost there. And then it will say, “You will tell me…” that’s what you’re telling the AI, “You will tell me three things I’ve done well and three things I could improve.”

“For each improvement opportunity, provide suggestions on how I could implement them. Make your feedback concise and reference specific parts of the text I gave you.” And then you just paste in your work at the end. I use this all the time.

And it’s such a simple way of using it, right? It just takes something you’ve done, and you just say, “Hey, this is my context.” Again, context is super important. It’s super important, because if you don’t put your who, why and what, then you’re get generic advice that might actually lead you in the wrong direction, right?

So, if you put the right context and if you ask this, so much value and, honestly, you can get to some pretty amazing return investment in like two minutes. Every skeptic I have, every skeptic I speak with, and, you know, I still meet quite a lot, I ask them to do this, and they always come out quite interested in the tool after that.

An example I can give you is I worked with lawyers. Gosh, lawyers are an interesting crowd, because obviously, they’re very critical for really valid reasons.

And I always tell them, “Look, take a case that you have that you can share publicly, take a response that you have written or something that you’ve produced, and just ask for three things that you’ve done well and three things that you could potentially improve and how.” And, usually, they get one thought, and they’re like, “Huh, I haven’t thought about it. Interesting.”

Then they might decide not to use it. That’s up to them. But having a really expert second opinion with a one-minute effort and for free, honestly, “Where do I sign?” It’s amazing, isn’t it?

Pete Mockaitis
Yes. And I’m thinking about, when you said expert opinion, it’s funny, when I heard that, I reacted a little bit because I’m thinking about Sam Altman talking about, you know, doing his very Sam Altman storytelling thing that he’s good at. Talking about the release of GPT-5, it’s like, “You know, before it was like you’re talking to a high schooler. And now it’s like you’re talking to a PhD in any area.”

And so, I was like, “Hmm, this is really not my experience at all, good sir.” But I think it’s expert in the sense that it’s been around the block. It’s like, “Yo, I’ve read the whole internet, okay? So, in that sense, I’m expert.” And I’m thinking about, there’s this book called Obvious Adams. It’s all about thinking, “Well, what would be the most obvious thing?” Or, Tim Ferriss says a question, “What would this look like if it were simple?”

That’s often my experience is it says the thing that’s not crazy, innovative, and brilliantly never before seen, but it’s like, “Huh, I probably should have thought of that, but I didn’t, and you did. And because you’ve surfaced that, we’re moving this forward, and that’s helpful. Thank you.”

Gianluca Mauro
Yeah, absolutely. So, I think one thing that is really not intuitive is that AI sometimes feels extremely smart and sometimes feels extremely dumb. And it’s really hard to predict, whether for my specific task is going to be, you know, the former or the latter, like, “Is this a 10 out of 10 question or is it going to be a one out of 10 question?”

There was this famous viral thing, viral experiment that came out, which is if you asked AI to count how many Rs are in the word strawberry, it would just say two, and there are three, right? I think a five-year-old can do that, probably, you know, but AI can’t do that. But, hey, it can write a pretty good legal letter for, you know. It’s just like so weird. It’s like it can do math, it can write code, but then it can’t count Rs in the word strawberry. Like, what is this?

And I think we just need to understand that it’s called artificial intelligence, but it’s not intelligent in the same way that humans are. It’s a different kind of intelligence. It processes data in a different way. It’s really hard to just give people a sort of like cookie cutter, very simple rule of thumb to understand when you’re in a good space to ask questions to AI and when not.

You just need to develop a little bit of sort of a gut feeling for, “Hey, this is something where I might get something good, and this is something where I might not get something good,” but there are guidelines. And the guidelines are, there was this research done by Harvard Business School, and they basically came up with a very simple classification of skills that AI has, so to say, AI capabilities. They call them within the frontier skills.

And these are four, very simple. Copywriting. AI is amazing at taking text and just turn that into other text. Now, a professional copywriter might argue whether that’s good copywriting or not. That’s a different conversation, but it’s amazing at just manipulating text, writing poetry or, think about this. It can write poetry and a legal document. I can’t do either, okay? So, it can do all these things. So that’s the first one, copywriting.

Second one is persuasiveness. So, it can write pretty good arguments if you ask it to, which is interesting. The third one is they call it analytical thinking. And it’s quite interesting if you give it a complex problem, and if you ask it to analyze it, it can give you recommendations or different ways to look at it.

And that’s the example that I gave you before, right? If you give it something that you have produced, legal letter, interview questions, whatever, and you say, “Tell me three things I have done well, three things I could improve based on the context,” it does it really well. So, this sort of like analysis, analytical capabilities.

And the fourth one is creativity. Now, people argue whether that’s real creativity or not. I don’t want to get into that philosophical conversation, but from a pragmatic point of view, it is quite creative, honestly. I had this thing a few days ago where I had a framework that I came up with to support companies in finding use cases for AI. And I was like, “How do I call this thing?

And I just gave it to GPT-5 Thinking, and I said, “Just please come up with an acronym.” And I would have never come up with any of them. It was super interesting and creative and it worked quite well. So, these are four things where you can feel quite confident. So analytical thinking, copywriting, persuasiveness, and creativity.

Now they also found where AI does not perform well at all. And that’s when you’re asking it to give you a recommendation, analyzing a bunch of different conflicting pieces of evidence. Let me give you an example. What they did is they took a few researchers, sorry, a few consultants from Boswell Consulting Group.

They took these consultants and they asked them to analyze a bunch of evidence of different strategies that a business might decide to go for to launch a new product, okay? Three different strategies. There’s a PDF with a bunch of interviews. There’s an Excel sheet with a bunch of numbers. All of these things, you need to look at this evidence and ask AI to help you in identifying the right strategy.

What they found is consultants perform better if they did not use AI to come up with the right strategy. Why did that happen? Well, because when you have conflicting evidence, conflicting pieces of information, in this case, imagine data said, I’m just coming up with stuff now, data said that sales were going up. But in an interview, somebody’s sales are going down. There was this conflicting piece of evidence.

AI was basically just like going with one. It was really hard for the AI to understand what was true and what was not. Whereas, for humans, it just made more sense to, for instance, look at Excel sheet, but ignore the interview because they thought maybe this person doesn’t know, doesn’t have the most updated data, for instance, that’s an example. So, AI was just like misled by the data that you provided.

Unfortunately, that’s how a lot of people use AI. A lot of people use AI today this way. Get a bunch of PDFs, a bunch of data, a bunch of emails, a bunch of stuff, throw it in, and they just ask for a quick answer to the problems. AI doesn’t work that well when you provide an insane amount of information and just ask, “Hey, tell me what I should do.” You should go step by step. You should use it to, again, amplify your thinking.

So, a better way would be, put this data in and say, “Hey, can you summarize the key takeaways from each one of these documents?” You take them and then you say, “Okay, what might be a good strategy? What might be good arguments for strategy one? And what might be good arguments for strategy two and strategy three?”

You see how you’re using it as a co-pilot. And that’s a really good branding from Microsoft, by the way. You’re using it as something that assists you in thinking rather than a, “Hey, I’m going to throw all my data. Just go ahead and do my thing. I’m lazy. I’m just going to copy your output and give it to my boss, you know.”

Pete Mockaitis
Yeah, I like that a lot. And, in a way, it really makes sense that it is that way because it just says, “Hey, I just know what words mean and what words tend to come after and next to other words. I don’t actually know that some dude’s opinion is of less importance and should be given less weight, gravitas, than a summary sales data reflective of millions of transactions.

Gianluca Mauro
Yeah, and it’s sycophantic as well, so it’s trying to please you. Imagine like, you know, you go to a doctor and say, “Hey, I have some headache,” and the doctor tells you, ‘Get this. Get this pill and just go.” Well, that’s not a good doctor. You should ask a little bit more questions and trying to understand.

What AI, and this is improving by the way, but historically, has been trained and, you know, it’s used to just get an answer. And so, if you provide maybe conflicting piece of information, as we said in the case study before, it’s just going to try to give you an answer rather than pushing back. And I go back to what I was saying before. This means that the tool is powerful, but it all comes down to the mindset that you have when you use it.

Do you want to have quick answers and you just want to get as fast as possible to a bunch of texts you can send to your boss or you can publish on LinkedIn? It’s probably going to just boost your laziness and just not get anything high quality. But if instead you use it as an amplifier for your curiosity, for your expertise, for your capabilities, well, now we’re talking. Now you can really get to some amazing outputs.

Pete Mockaitis
Yes, I like that a lot, the amplifier. And it’s interesting, is sometimes, I think when you look at the prompt that you’re sharing, it really does kind of garbage in, garbage out, and it’s the opposite, you know, magnificence in, magnificence out. So, I could say, “Hey, give me some information about sleep apnea.” And so, it can say, “Oh, well, this is a common affliction, blah, blah, blah.”

But then what I’ve said is, “Show me the results of several human randomized control trials that utilize novel interventions for the treatment of sleep apnea, i.e., not a CPAP machine. And give me a summary of the quantified impacts associated with the apnea hypopnea index reduction associated with each.” Now that, and sure enough, that has led me to some interesting places. And I found this thing called inspiratory muscular training. You breathe against resistance. And what do you know, that really helps.

Gianluca Mauro
Interesting.

Pete Mockaitis
And I’m not using a CPAP machine. So, thank you AI for putting me in some good directions. But I think it shows that, “Are we amplifying laziness or are we amplifying a targeted, ferocious curiosity?” Like, “No, find me precisely this, and then we can play ball.”

Gianluca Mauro
Absolutely. You’re spot on. It’s perfect. But, to me, the interesting thing about this whole concept is that there’s quite a lot of responsibility on the user.

It’s basically telling people, “Hey, if you don’t get the right output, it might not be because of the tool. It might be because of the way that you are using the tool,” which from one point of view, I think is empowering because it’s basically telling me, “Hey, amazing, I have some agency over the output that I get.”

But from the other point of view, I think some people might find it a little bit stressful, “Now I need to learn about A, B, C, D, all these different things so that I can actually use this machine well.” Well, yes, but at the same time, honestly, as I was saying before, it’s about changing habits. It’s not that hard. You don’t need to get a PhD in Math to understand how to use one of these tools.

And so, what I recommend to people who might feel a little bit maybe overwhelmed, or maybe afraid that you’re using it wrong, I always tell people, “Hey, find your little safe space to experiment. Take a hobby that you have. Maybe you’re interested in, I don’t know, Formula One.”

That’s one of the latest things that I’ve been nerding about. And just go and try to do your researches and prompts and test things about Formula One that’s maybe not related to your job so you feel safe. There’s no fear of putting sensitive information into these tools, and just try to get a sense of how the tool might be helpful and useful for you in a setting where you’re free to experiment. And then you can take all these learnings and apply them to your job.

Pete Mockaitis
Okay. Well, so you are somewhat famous for your CIDI framework – context, instructions, detail, and input. And it sounded like you were giving us exactly that in the context of, “Hey, give me some feedback on a thing.” And so, can you give us a little bit of detail for how we might think about applying this in all kinds of different ways?

Gianluca Mauro
Absolutely. So, the CIDI framework is a framework that I came up with, I think, a couple of years ago, maybe. And my objective was to find a simple recipe to get people to think about their prompts in the same way that I think about my prompts.

And so, it’s quite simple. It starts with C stands for context. Tell me who you are. Why are you doing this task and what do you need to do? Just try to make AI get into the zone of, “What are we talking about?” Think about this, AI might act like a lawyer, might act like a doctor, might act like anything, right? So, you need to zone in.

The second part is instructions. When I say instructions, it’s important that you’re very clear, and you’re talking to a thing, not to a human so you can be very direct. And I typically give my instructions this way, “I will tell you this, you will do that.” “I will give you an email I wrote, you will give me feedback on it.” That’s the instructions part.

The third part is details. Details are, basically, I look at it this way, it’s very simple, “Explain what good means for you. What does a good output look like for you?” And that’s an interesting question. I feel like it’s almost meditative. It’s almost like therapy. You need to ask yourself, “What do I want? What do I really want? How does a good podcast script look like? How does a good LinkedIn post look like?” And just describe it in plain words.

Pete Mockaitis
Or, I guess this could also be examples, like, “And here are three instances that I consider good.”

Gianluca Mauro
You got it. That’s the part of the prompt where you might want to put, for instance, something you’ve done in the past, and say, “Hey, look, this is something that I consider to be really good,” or, “This is something that represents my tone of voice, and I want you to try to replicate that.” That’s the details part.

And the last part, the input, is when you put actually what you need to produce. So, for instance, if you need to have feedback on a legal document, you put it all the way at the end. The reason why I structure it this way – context, instructions, details, and input – is that it’s very easy to reuse.

So, if I write a really good prompt that explains exactly who I am in the context and what I need to do, exactly what I want out of it in the instructions, in the details, and then the input is, let’s say, this legal case that I need to analyze, the next time I have a new legal case to analyze, I just need to replace the last part of my prompt. The first part of the prompt, the context, instructions, and details are the same.

So, it makes you, number one, think about all the important things in a prompt and leaves really little room for error, because you need to think about all of them – context, instructions, details, and input. But it also makes your work scale a little bit. Because some people, and I get it, get stressed, “I need to write a good prompt. How long is it going to take me to explain who I am, what I need to do, yadda yadda, yadda. What does good look like?” I understand it can be a little bit of a pain if you want to write a really long and cohesive and complex prompt.

But if you write it this way, then it’s very simple to reuse. And that’s your copywriting podcast guy. That’s a perfect example. I think he was prompting, using, maybe without knowing, but he was using that sort of structure, it sounds like, because he had some things that you could probably copy and paste again.

Pete Mockaitis
Certainly. I think it is John Morrow is his name, and we’re going to include that in the show notes. Indeed, the context is, “Hey, we’re a sales letter for this product for, you know, which serves this user with these needs, wants, concerns, who use language like this.” And then the instructions are, “Write the headline of a sales letter.”

The details are, “Here are some other headlines that we think are fantastic, as well as the general guidelines of copywriting that we find to be effective in this industry.” And that might be hundreds or thousands of words, and that’s acceptable, right?

Gianluca Mauro
Yeah, I mean, always try not to go too crazy because then it becomes too much context, right? I think AI could maybe process it, but I always say, you know, if you try to put too much stuff and you’re not fully sure about what you actually put in, then you might have added something that is actually misleading. So, try to keep it in check. Don’t put too much if you don’t know what you’re putting in. But, yes, conceptually makes sense.

Pete Mockaitis
And to that point with the context, I mean, I believe, you know, we’re like a hundred thousand plus tokens. So, is it your professional opinion that, okay, you might have a hundred thousand tokens, but don’t use 50,000 words? Or, what’s your take?

Gianluca Mauro
Yeah, so, I mean, for people who don’t know what tokens are, tokens are basically AI breaks down your messages into parts, parts or tokens. So, if I say, for instance, “Hello!” it might be two tokens, “Hello” and the exclamation mark. And there’s something the AI models have called a context window, which is basically how many tokens they can take a look at, at once. And I think GPT-5, the latest model from OpenAI, is at 400,000 tokens. Some models are up to one million tokens.

So, while you can add a million PDFs and resources into a prompt, then you risk getting into a situation in which you’re adding, if you’re adding not high-quality context, you’re just misleading your model. An example that I always make is the following. You go to a doctor and you say, “Hey, as I said before, I have a headache. What do I do?” That’s way too little context.

But if you say, “Hey, I have a headache. Let me tell you my medical history. When I was two years old, I once fell and hit the knee on the floor, and it was really painful. Then when I was three years old, I once ate spoiled milk. When I was four years…” that’s too much. You’re just confusing your doctor, right? So, you want to try to select some context that might be relevant.

Because, again, you never know if you’re just putting something that is just misleading or it’s just not very relevant to your question. Don’t stress too much about not putting enough but also don’t go crazy.

Pete Mockaitis
Well, I guess what I’m saying is, what I want your professional judgment on is, if I throw in the full transcript of a meeting, or a book, you know, is that likely to help or hurt me or under what context?

Gianluca Mauro
Oh, I do this all the time, by the way. Like, if I take the whole transcript of a meeting, and I need to write a sales proposal, transcript of the meeting, the whole thing, because it’s all relevant. It’s my meeting with a customer, it’s all relevant. And I take that, I take an old proposal that I wrote, and say, “Adapt this proposal to the context of this meeting.” That’s perfect. But think about what I added in. I added only relevant material for my task.

Pete Mockaitis
Well, that’s helpful. You mentioned it’s bad at analyzing conflicting information. And I’m thinking, sometimes it also seems bad at giving me precise pinpoint pieces. For example, if I say, “Give me verbatim quotations of something,” it seems to really struggle.

And maybe it’s just trying to not violate a copyright or something, but I feel like the more I want it to be super precise, specific, narrow, exact, data point or quotation, it seems to struggle.

Gianluca Mauro
Yes, and you will be correct in saying that. And, I mean, there’s a technical reason why this is happening. It’s just in the way that these tokens are processed. It’s basically making a big average of everything that it has read up to that point. But conceptually, you can look at it this way. This is not a truth machine, okay? This is not like a search engine.

A search engine takes a bunch of data, the entire internet, and it just points you to the right point. It tells you, “Hey, this is the link that it’s the most relevant to your query.” This is not that. A good way that I look at AI is it’s a compressor of knowledge. It took all the knowledge of the internet, compressed it into a thing. And then when you ask questions, it can decompress it and give it back to you.

So, what this means is that sometimes you lose some information in that, say, decompression. And I mean, I think this is a metaphor that is, really, maybe it makes sense just to people who are into audio and this sort of stuff because you have this thing. You’re losing quality as you compress it. It’s the entire internet, but you can just like quit it like this in a second. So, you lose a little bit of quality. And so sometimes you have these errors. But I have to say it’s improving really fast.

Pete Mockaitis
Okay. Well, is there anything else that it’s bad at?

Gianluca Mauro
I think a good way to look at this is thinking that it’s an amplifier, okay? So, it’s bad at telling you, “Hey, what you’re doing, it’s not ideal.”

Pete Mockaitis
Yeah, “You’re asking the wrong question,” you know?

Gianluca Mauro
Exactly. It’s not going to do that, which is like, I think, the most important skill today is not being able to find answers. I think it’s being able to ask the right questions and being able to look at answers and be like, “Oh, this isn’t, actually, this doesn’t make any sense.” I’ll give you an example. I think this is quite funny.

I asked AI to give me feedback, as all of you know that I do that quite often, on a PowerPoint presentation that I created. But instead of, like, uploading the slides, I just took all the texts and I put that in. And in the feedback, it told me, “Hey, you’re not using enough visuals.” And I was like, “Of course, you’re telling me this. I didn’t give you the visuals. I just gave you the text.” It makes no sense, right?

But looking, you know, critical thinking, and this is a very simple example, but I had to take that piece of feedback. And even though the best top AI model in the world told me that I need to add more visuals, I discarded that feedback because I was like, “I know that you’re just lying to me. You’re just coming up with random stuff.” So that’s one important thing.

And, by the way, I think something that really scares me is a lot of people are using AI for almost as a therapist to get support in relationships with their loved ones. And, again, remember AI is sycophantic and it’s going to try to please you, so you’re always right. It’s really rarely going to tell you, “Hey, Gianluca, you know, your…”

Pete Mockaitis
“Your behavior is toxic and causing problems. Look in the mirror and fix it.”

Gianluca Mauro
Exactly. It’s always the other person’s fault. Yeah, I had this friend of mine who came to me, and was like, “Hey, I have an issue. Every time now I have an argument with my partner, she goes in a room and then comes back and has a perfect, like, perfectly phrased argument to explain to me why I’m wrong. And I know that’s coming from ChatGPT.”

So, she’s just getting in a room, and saying, “My boyfriend did X, Y, Z, you know. How can I just try to win this argument?” which, again, I think there is some value in that if you use it correctly. Again, I feel like I said it a few times, but I really want to make sure the audience comes back with this. Think about if you instead use it this way.

Go back to your room and say, “Hey, I had this argument. What might be the other person feeling that I’m not thinking about? What might be some blind spots that I might be having? What are things that I’m not considering when I’m accusing, I don’t know, whatever my partner of, X, and Z?” Now you’re actually going to use it as an empathy machine rather than as a, I don’t know, ego booster kind of thing, right?

Pete Mockaitis
Yeah, well said. Are we amplifying the, “I’m right, they’re wrong,” make the case, or are we amplifying, “I’m trying to be understanding and compassionate”? And it will seek to please you. And so, yes, if we amplify the wrong thing, we’re just getting farther down a bad path.

Gianluca Mauro
Correct. And isn’t this cool? Like, the idea that I have so much agency and power over the outcomes of my use of AI, depending on how I use it, depending on the questions I ask and what hat I decide to wear on this day. “I want to wear the hats of the empathetic person who tries to understand what this person might be feeling.” I can have vastly different outcomes. I found this really empowering.

I understand it might be a bit scary, because it’s like, “It’s all of me?” Yes, I get it. But, again, if you have that approach of being curious and just trying different things out, I find that super empowering, honestly.

Pete Mockaitis
Cool. Thank you. Well, now I’d love to get your take in terms of, boy, there’s a lot of different chat bots and AI tools, if you want to do ChatGPT or Gemini or Claude or Grok. Is there a way, and this is going to change every few months but, you know, for now, is there a way you think about for certain use cases, “I prefer this tool over the others”?

Gianluca Mauro
Yes, but in a way that might be unusual for the audience to think about. So, I think about it this way. I think in my AI, I call it my AI tool stack, like all the tools that I use, I think about three main categories. The first one is generic AI tools. These are ChatGPT, Gemini, Copilot, Claude, Grok, basically these five. And, for me, it doesn’t matter too much which one you’re using. All have each respective strengths and weaknesses, but they’re all quite similar at the end of the day.

I personally use ChatGPT. That’s the tool that I started on. That’s just the tool that works the best for my use cases. But, again, I don’t have a strong argument for people to say, “You must use ChatGPT.” Use whatever you want. But the interesting thing is when I look outside of these generic AI tools and I start looking at specialized AI tools, and these are tools that are specifically built for one use case.

An example, I use a note taking tool called Granola, which I really love. I have a lot of meetings in my life, and Granola is specialized in note taking during meetings. Absolutely beautiful. And I’m not affiliated with Granola at all, so I can tell you there are other tools that do that as well. Otter is one. It’s pretty good. There are a few ones.

But, again, for me, who, I take probably too many meetings. Having a specialized tool for note taking during meetings is super valuable. But there are specialized AI tools for lawyers. There’s a tool called Harvey. There’s a tool in Europe called Legora that’s amazing. And these are specialized for lawyers. They give you a bit more features that you might be interested in. They’re a bit more accurate. They have maybe all the laws of a country already loaded in. You know, they’re more helpful.

I have a startup called Epiphany and I built a specialized tool for instructional design. It helps people who create training, create better training faster. And it’s really interesting when you start looking at those specialized AI tools, because, again, you might find something that, for you, specifically for you, can have a lot of value.

I know, for instance, there are tools for podcasters. Like Riverside has some pretty interesting tools to, like, repurpose content. You might find a lot. Curious to know if there’s anything that has really changed the way that you work in that space.

Pete Mockaitis
You know, it’s funny, we do a little bit here and there, but we’re actually still transcribed by humans, which could shock some people. So, we use it in specific, narrow targeted places, but, still, each episode is getting many, many human hours to put out the door.

Gianluca Mauro
And that’s perfect. Again, for me, what I advocate for is thoughtful AI use, not just like take it and put it everywhere. That doesn’t work for me. So, it makes total sense. But that’s the second sort of area that you might want to look at. So, pick one generic tool. That’s like saying Excel. You can use Excel if you do marketing, if you want to track your campaigns. You can use Excel if you’re in finance. You should use Excel if you’re in finance, but you understand where I’m going.

Same thing, ChatGPT or Gemini or Claude can be used by people in every single industry. But then look at those specialized AI tools that might help you even more.

And then the third area is those custom-built automations that you might want to build for yourself. That’s when we’re getting really nerdy, okay? But I love that. And the interesting thing is that the barrier for building your own custom automations has gone down so much. It’s crazy. There are all these AI no-code tools that allow you to plug different tools together so you can build an automation like, look, I’ll tell you one that we have in my company, in AI Academy.

Whenever somebody writes on our website, “Hey, I’m interested in a custom enterprise training.” There’s this custom automation that researches the company and just gives to our salespeople on Slack a message, and says, “Hey, this person has reached out. This is who this person is. This is what the company does. This is what they want.” Research is done already. It’s like a sales assistant, basically.

We built it ourselves. It took us, I don’t know, we know how to do it so it took us a couple of hours maybe, maybe three, I don’t know, something like that, okay? Hours, not days. All right? But we have seen people starting from very limited technical skills, being able to build those custom automations for their business or for their freelance profession in just a few weeks.

Pete Mockaitis
That’s cool. That’s good. Well, talk about dorky, even though I am not at all a coder or a developer, my favorite YouTube channel is Fireship, which has all these jokes and stuff. And so, I’ve heard, I know a couple of the buzzwords associated with AI automation, like the MCP, the model context protocol, as well as the N8n.

Gianluca Mauro
Amazing.

Pete Mockaitis
And so, I know those are key words I might Google and research. But you tell us, if we are just starting to tiptoe down the, “Hey, I got a thing in my life I need automated. I think AI could probably do something about it,” what are the first steps to explore pulling that off?

Gianluca Mauro
Yeah, so I’ll just tell you basically how we help people go from, “I might want to automate something” to “I’ve built an automation.” And that’s just because we know that that’s a process that works. The first thing you have to do is understand what to automate, what not to automate. And it sounds very basic, but I guarantee you that’s where you decide if you’re going to be successful or not.

Most people want to automate too much. And then they start building spaceships that are never going to work, never going to give them the result they want. They’re going to get tired after some time when they try to build it and it doesn’t work, and then give up.

Instead I always tell people one day I’m make a T-shirt with this sentence, “Find the smallest possible thing that could possibly work.” The smallest possible automation that could give you some value. Start with that and then you can expand, all right?

The second part is don’t stress too much about the tools, you know, the N8n or Make.com or Crew AI, all these tools that are coming up, but try to write a pretty good prompt that should power your automation, okay? So, focus on the AI component, and find a way to test it well.

What do I mean by this? I’ll give you an example that I had. There was this one of our students, he was a doctor and he wanted to not just build an automation. He wanted to build a product to give to his colleagues so that they could easily write referral letters, okay? And so, he had to make sure that this thing worked really well because, again, doctors, you know, it’s a lot of responsibility.

So, what did he do? He just found a bunch of referral letters, or he wrote a few with ChatGPT, and he corrected them by hand. And that was his set of examples to test whether his prompts were actually producing something that was good enough.

Pete Mockaitis
Yeah, like, “Can the AI actually achieve the thing I’m hoping it to do? Let’s test that before I build out a whole thing and go oopsies.” Love it.

Gianluca Mauro
Correct. Another one of our students, I saw this last week, so I remember really well, did something to create LinkedIn posts. This guy works in marketing and risk is very low in that case. What’s going to happen if you publish a really bad LinkedIn post? I might get annoyed, but no one is going to die, right? But still what he did is he used his prompt to create a bunch of LinkedIn posts and then he wrote some and then he gave them to some of his friends and said, “Hey, which one do you like the most?”

And he tried that with a few different prompts, with a few different models. He tried with GPT-5, he tried with Claude, he tried with Gemini, and then he just found the best. And then he knew, he had the confidence that this automation was going to work because he had done the work of testing it and collecting the data.

After you’ve done this, step three is now, build your automation. And, you know, there are different tools that are pretty good at this.

Make.com is one that I really like. I use it a lot. Zapier is probably the easiest one to use. If you want to get started and don’t want to waste too much time learning how to use slightly more sophisticated tools, Zapier is a great place to start. N8N is probably the one that gives you the most flexibility on things that are the most capable. And I like that a lot as well.

But again, does it matter? Not really. At the end of the day, if you had a good idea about what to automate and there’s real value, then you can just swap tools and you’re going to be good. So, I suggest that that’s the last thing that you start thinking about.

Pete Mockaitis
That’s good. Well, shoutout to Zapier. We had Wade Foster. Or is it Zapier? I don’t know. Zapier or Zapier, we had Wade Foster on episode 466 back in 2019, and they are still going strong.

Gianluca Mauro
Amazing. Super strong.

Pete Mockaitis
Good, handy stuff there. All right. Well, Gianluca, tell me, anything else you want to make sure to mention before we hear about a couple of your favorite things?

Gianluca Mauro
No, I think we’re good.

Pete Mockaitis
All right. Well, now can you share a favorite quote, something you find inspiring?

Gianluca Mauro
I wish I could say who it comes from, but it unfortunately comes from a random guy on the internet. And the quote is, “The hardest part of getting what you want is figuring out what that is.”

Pete Mockaitis
I dig it. And can you share a favorite study or experimental or piece of research?

Gianluca Mauro
I will share the research that I was talking about before, the one from Harvard, where they looked at all the different capabilities of AI. And the name of the research is “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality.” It’s really cool, still very relevant from a couple of years ago, but, honestly, I still quote that, basically, in every workshop that I do because it’s really valuable.

Pete Mockaitis
And a favorite book?

Gianluca Mauro
Ruined by Design.

Pete Mockaitis
All right. And a favorite tool, something you use to be awesome at your job?

Gianluca Mauro
Another tool that I started to use recently is a tool called Wispr Flow. It’s quite interesting. It basically allows you to dictate and it just puts whatever you said into a box. But again, it uses AI to just change that a little bit so that it’s, first, it’s formatted already.

So, when you’re writing emails, you might want to just record and say what you want to say, and say, “Here, I want some bullet points,” and you’re going to see the bullet points. I’ve been using it for a couple of weeks and I might see that becoming a key part of my tool stack.

Pete Mockaitis
Okay. And if folks want to learn more or get in touch, where would you point them?

Gianluca Mauro
You can go on GialucaMauro.com. That’s G-I-A-N-L-U-C-A-M-A-U-R-O.com or on AI-Academy.com where you can see all of our trainings so you can get better at AI.

Pete Mockaitis
Okay. Well, this has been so fun. Thank you and good luck.

Gianluca Mauro
Thank you. Thank you for having me.