From AI Hype to AI Harness Engineering – Building AI That People Can Actually Trust with Alan Buscaglia [MVP] from Gentleman Programming

Yeah, hello everybody and welcome back to another episode of the M365 podcast. I'm your host, Moccopitas and today we are diving into one of the most misunderstood areas of artificial intelligence. Everyone is talking about AI, everyone is building AI, but very few people are talking about how to engineer AI systems that are reliable, manageable and actually useful. Today's guest is someone who lives exactly that intersection. And Bruce Gargidy is a difficult last name, no? The name is Alan Buscaglia. It's difficult. Oh, yeah. Yeah, Alan Buscaglia is an app lead at Proudar, Microsoft MVP, Google developer expert in Angular, EduKTRA, open source contributor and creator of massive Spanish-speaking tech community, Gentleman programming, reaching well over 200,000 developers across YouTube, Twitch, Instagram, Discord, GitHub. Yeah, I'm a real man, I'm everywhere. Yeah, it's a great focus, something incredible exciting, loop engineering and AI hours engineering that is the pin of creating AI systems that constantly deliver high-quality outcomes instead of unpredictable magic tricks. We are also going to discuss community building, developer education, integration leadership, Microsoft AI ecosystem and what is the future of software engineering looks like in AI first. So welcome Alan to the M365 show. Yeah, thank you, man for the meditation. It's a great thing, I received your message and said, "Okay, I had to go this way." So perfect, thank you for that. Yeah, I think there are a lot of people who know you about for people meeting you the first time. Who is Alan? Oh, Alan is a strange, I'm a worker, but more than anything, I'm a teacher. I consider myself a teacher. I love teaching, I love sharing knowledge, I think knowledge has to be free and has to be everywhere. So that's why I started my YouTube channel that is gentleman programming, you can also find me well as you say, everywhere, right? Apart from that, I'm also, apart from content greater, I'm also an open source developer. I create in AI tools like gentle AI or NGram that are trying to simplify people's lives. Not only people are developers, but I have people, for example, there are COs or they are working on data science or even finance. And they use my tools and they say, "Hey, this is not anything, perfect." Yeah, and you did become both a Microsoft MVP and a Google developer expert. Exactly, exactly, a foot in world walls. Yeah, that's really amazing. So I think it's the, yeah, three big, I think Amazon, Google and Microsoft, you have three big, so you have two or three of them, so you have to collect. Yeah, Amazon, you know what to do now. But when you look back in your career, what, yeah, decision change, everything. Oh, a lot of stuff. But the principle thing, and no worries about this, I'm, you're going to find that I'm super open about everything, okay, and through and through transparent. And okay, in this case, the thing that make a click in my case was my father's death, okay, it was like something that really made a click on me. It was one of those, you know, people that are just, I don't lazy. And when that happened, everything came true, it had to be like the man in the house, you know, and, and they care of things. And, yeah, that then really made it change on me. I started having, I don't know, grabbing a university more serious, studies, a third of my career, also everything, everything. And when I start working, I do have a mentor and have a tutor, right, for everything that is a program related and such. So I try to be that one, for people, but that's why I'm creating content. I try to be that mentor that I didn't have. Yeah, your, your contradicts, experiences, it's really interesting. You have the gentleman programming and I have tried to validate it on, on some tools, but it's, all says it's one of the largest Spanish-speaking developer community. How did all this begin? Well, in pandemic, you know, in pandemic, you could have two things, or you have a kid, or you start creating content. So in this case, they're creating content, the kid came after. So I don't know, as I was trying to get myself, you know, something, trying to do something. I always love creating content. And I use to, for example, to create content for my, my companies that I work at the time. I also started communities in those companies. So I say, okay, let's merge those things up and boom, the anti-mob programming came through. That was the main thing. Yeah. And have you, I think on YouTube, you have more than 120,000 followers. Oh, yeah. I think it's a straight word. Have you ever expected to reach this, or was this a plan? No, not at all. Not at all. I still remember when I had my first thousand, that it was almost crying. It was crazy, because I was thinking, okay, first, why? Why people are watching me. What I had to offer to these people and having people tell me, you know, I had a job because of you, or you have in this and the other. I came through this difficult challenge because of when you're teaching, because I don't only teach programming and now AI. I also teach software skills, sorry, soft skills. I teach, I know, I, one of those guys, they have like a phrase for everything, you know? So I know, I like giving advice. So for example, I used to have a space in my streams because I also do a stream on Fridays and I had a space, I was one on one where you can just get into this, go with me, talk about something, you know, ask me something and I was going to try to help you. And the number of people that came in, just to say thank you, you know, for those kind of stuff, thank you, was amazing. And I, I still don't get the numbers I want to like live from this. I don't even want to leave my coding job because I love my job. I love working, you know, because when you get in the, in my case, right, I'm able to open source, that's amazing. I open source, sorry security company. I'm still learning and I'm, I'm hands in the, in the challenge, right? And if I get out of there here, what do you have to teach people? I don't know, theories? No, I like teaching practice. So it's a little bit of everything, right? It's, I don't know, it's amazing, it's amazing. What people gives me, the satisfaction of that kind of feedback is just, it pays for everything. Yeah, I think then is one, one, the, what the people say and the feedback from the people was one thing that motivates you to continue screening the educational content. Yeah, you did. But I think at the start, it's, I feel the same. I have, I don't know, on, 100 follow on YouTube and do every day. It's really hard. How did you motivate at the start? Yeah, I don't know. I don't really know. I think it's just one of those things that, you know, gets to you. I don't know. I used to help a lot of people in the university, you know, and even school. You know, those guys that always give a hand when there's an exam, you know, I was that guy. It was the one that tried to explain things in the last minute. But the issue is that the people that I teach, they pass the exam and not me. That's a crazy body. But I know it's something that is just like a call. I don't know. I really love doing this. That's the excretion. And I think you do so much in all this channel and you have a drop. You do a speaking. You come to podcast. Yeah, I think you have also a family life and you do the content creating. How did you balance this is awesome? It's difficult. It's a lot of time because I also have a family. I also have a kid of two years. So I had to balance everything, you know, because as you say, for me, my job is my life. That's the thing. I work eight to nine hours in my, let's say, official job, let's say, but then I have my integration. I have the open source and have everything. For example, now I'm a vacations, but I didn't sleep at all. I have been doing open source in like crazy. And I love it. I love it. It's one of those things. I know people like watching movie when they're vacation. So going somewhere, I love working. I mean, I love having my own time, you know, I don't have any pressure on my things. So yeah, it's like that. We have my phone with me. I connect remotely to my computer. I give an instruction and continue my life and just manage things. Yeah. That's awesome. You have a kid, will you also say it should also become a thought leader or I think in this generation, it's more in flencer to word? It can be. It can be his crazy. And that's why you have to be, right? He's a little bit crazy. Yeah. And what have you learned from leading such a large online community? Well, that you never know enough. Doesn't think you never know enough. And the imposter syndrome is there always, always. And when you're tired, when you think I did this and this is wrong and people are going to hate it. And from nowhere, you have a message saying, Hey, this is so cool. Thank you for this. It's a black me and this and the other. That's amazing. I mean, the feedback doesn't have to come from you. It has to come from people that respect you or, I don't know, even people that also respect you, even they have a critical, bad feedback, you also have to see in between lines, right? What can I get from this message? And, I don't know, it's crazy. It's crazy. And again, about the toxic feedback is going to happen. It's going to come. There's always someone going to say something about you and that's the thing, you have to care. You have to care. You have to be yourself. And that's it. Yeah. How do I help you through handle all this community and stuff that you use AI or are you critical? I do everything. I mean, when I started, I wasn't a thing. Even automations were difficult, right? At the start, it was just me doing everything, but it's still me. It's still me, but it's different. Because when I started, again, it was just me. I was like getting this truck into motion, right, trying to start the car, then join. And now it started and it's going. So if I'm not there, nothing happens. It starts seeing people that come through, right? And I say, there's a question. Previously, I had to be the person responding to questions. Now the own community is responding, it's helping, it's trying to give you a hand and that's crazy. That's really crazy. So that's where I say, I'm not alone anymore, right? Because my community is there helping giving a hand. But I use AI. I love AI. I'm going to do something that AI is not going to replace your thinking, but it's going to help you give it, that's it better to other people, right? To share it better. For example, my post on social media is on this. I literally spent hours and hours giving instructions to the eye on how I talk. So I just say, okay, this is the message that I want to give. Give me this post and it has to, you know, with all the distractions such as for great CEO that has to be organic, it has my own way of thinking of this. And boom, I have my post and when I read it, it's exactly as I was going to do it. So I had to change anything. Maybe one or two things, but it has me a lot. It really has me a lot. I never, but never live AI to pull the trigger. That's me always. I always part of the loop. I'm the human in the loop. And what excites you most about AI today? Well, there are a lot of things. I mean, we are in strange times, right? Because we always have like this generational lip when you say, okay, now we have the CLI, the compilers. Every single thing has been a generational lip. Right now we're in the middle of it. We don't know what is the next big thing because we are in the next big thing. Right now AI is evolving, is changing stuff so much that we don't know the end of it. We don't know it. So in the first thing, I'm excited for you. And in my case, the way of coping with all these changes is being part of the change. Being at the top of the wave, trying to be there, right? To be, let's say, to have some control on what's happening. There is really difficult, really difficult. For example, my AI tools are the result of that. I am on top of what's happening, trying to, for example, I am trying to, I am trying to do a lot of harness engineering and look engineering. Okay, what I'm doing is trying to give some control to this amazing thing that is an agent and trying to share that knowledge in a way that you installed it. And you can use all those hours that are spent for it, you know, and super easily. So it is a part of that. I'm excited a little time. I'm tired. We can say, super tired because it's a lot, it's a lot of changes, but I know it's exciting. I really like it. Yeah, I see you talk a lot about the AI harness engineering topic. When you should explain it in simple world for people that are not familiar with it, what is it exactly? Okay. Let's say it this way, every single time you start a code or codex or codec or policy AI or whatever tool you want, you don't know it, but it's not just like you ask a question on something happens. If you ask exactly the same question in everything, one of these tools, every single one of them is going to respond differently because every single one has a harness engineering behind it. That's why some people say, I like codex. I like a pilot. I like code code. Why? Because when I work with them, they work this way. And for me, it's better. Okay. The thing is, when you cope with what a company wants or what the company thinks or is exactly what you want or maybe it's not exactly what you want. It's different. And you defer, right? They wait for thinking. That's why you can create your own harnesses. And that's why, for example, Gentile IS is a collection of harnesses and ways of thinking and reasoning. So a harness is exactly that. When you ask an ancient question, it doesn't go directly into the response. It's, let's say, directed is guided into how to give you that response. For example, if you're working in development, right? The flow will be, I implement something, I committed something, I create a pull request and boom, I merged it if it passes. That flow is a harness. Exactly that is a harness, right? You say, first, whenever you have implemented, you have to see how the code is made, what are the conventions of the project? Then how is the template of the comments that you have to use to the comment? What is the template of the PR that you have to use to create the PR? Okay, all the continuous integrations tools are, sorry, actions are green. Are they working perfect? Now you can merge it, okay? Those are harnesses. There has been here for a lot of years and now we're expanding them into agents, into AI. Yeah, I have the last week. I read a funny study from how in which language you ask the same questions in Cloud Code, it gives different answers and it's try to be more like the stereo chip, like if I ask a German, it's more straight. And I don't know, other language. Yeah, that's really funny. So, did you think that the AI harness is something, why a lot of AI projects fail? Yeah, yeah, it's really what you saw. What most I think mine can fail at any time, because something that you do for one model, for one, it's not only the model, the F4, okay, for that model, may change the response, right, the output. So every single time that something happens, in my case, I have my great community, there's always time to test and stuff, right? So it's like, okay, this works amazingly on, let's say, a GBD 5.5, but now that 5.6 is here, it's not working correctly. Okay, let's see why? No, because it thinks too much, okay, so I had to change the harness. Maybe the harness is not needed anymore because the model is trained in a certain way, and it just works, right? So those are the things, it's difficult, it's really difficult, really difficult. And when you create a CLI, or I say a CLI, because I know, I think it's a clock code, it's a CLI to execute the agent, right, with all the prompting and this such. But I don't know, I think that most tools fail because they are, yeah, they try to over-engineer the solution, the harness. And when you over-engineer the harness, a lot of things happen first, when you harness something, you give context, and when you give context, you are getting the context from the agent, okay? So you're reducing the amount of context window that the agent can use, and even you're using motor guns for that. So when you over-engineer, these kind of things happen. And let's say it, yeah, I am a gentleman, okay? So in Latin America, we had to say it, having to spend a lot of money in forced descriptions on the AI is difficult, right? It's really difficult. But that's why a lot of people use, for example, say open code, and they use one subscription of the code code X, compiled, whatever, do have it as a main base, and then they use open-source models for everything else because they chip, right? So that's the thing. When you over-engineer, you're also affecting these people that trying to get into AI, but don't have the resources to spend a lot of money to have it, let's say, cut, right? So it's super difficult, really difficult. So I think a lot of companies and people start with the small language models, actually, and start building their own or try it. So, yeah, I think that's, yeah, yeah, the claw tokens are so damn expensive. Yeah. Mother, how do you make AI outputs reliable? What's it? That's the thing. It's difficult, really difficult. What's more, while we're talking, while we're talking, I have one, two, three, four, six stations doing things for me, right? And what they are doing is testing, they're crazy, all the harnesses that are implementing. The thing is the following, you have to have a complete set of end-to-end tests is like 100% and needed, and apart from that, you have to have a certain way of benchmarking it, okay? You have to bench it. Are you spending more tokens? No. Okay, but how much time did it need to reach the output? Less? Okay, that's perfect. Less tokens? Less time. But what happens if you start doing things like, something that I'm doing, okay, spoiler? I'm trying to use algorithms to reduce the amount of tokens that are used. So some of the computation is on the CPU instead of an affination. Okay, you're using less tokens, and maybe the algorithms are a little bit too much, and you spend more time. And that's the better per experience issue. And whenever you say, okay, use this, and it's going to use a lot of the less tokens, perfect. But if you have to, have, don't know, spend three hours of your life to change one line, wherein bad terms, right? So it's a balance, everything is a balance. You have to see how fast you respond, how much it costs, and again, how deterministic you can make that output. And that's one of the most difficult things also, because it depends, again, on the model of the effort that you use. It's super difficult, super difficult. So you do a lot of testing and spend a lot of token. Oh, I know. You know, one of your, both your cloud core are what you use. But yeah, but where do a God rails fit into this picture? Well, that's the thing. The Gar rails are, how can I explain it? Okay, I'm going to explain one, and then it's really good. In my case, something that I'm doing right now is something that is called commit, pre-posh, push, pre-release, gates. Okay, what are name, right? What I'm doing is the form. There's something really new. I haven't seen anywhere. That's why it's so difficult also. I'm trying to freeze changes and how the AI review it, review those stations. I give like a fingerprint of everything has been done for a current implementation. Then a habanation trying to reproduce or to test it, right? But manually, without tests, manually go execute this, do the other to see how it performs. And if the fingerprint passes the verification, then it goes into another step. So that's a Gar rail, right? That's a Gar rail. You try to stop the issue from going through issues or to overthinking or to over-engineering or to hallucinating. That's also super difficult. But again, as I say, this is new, this is super difficult because I'm spending time. It spends a lot of time doing that. Maybe it just changed the line. It really has to go habanation to test that you change the line. Maybe not. It depends. But it also depends in which line you change. If it's a line related to, I don't know, a billing, of course, they're going to test it. So again, it's difficult. It's difficult to categorize the change. It's super difficult. Can you a little bit explain how a feedback loop looks? When you're working and do you visit a human feedback loop, or is it also doing the feedback loops with the AI? In my case, I try to be over everything. I don't like to say, "Okay, do this. Perfect. Merge it." Now, that's not me. I try to be the human in the loop. I try to be the overseer, let's say, of everything. So in my case, what I'm doing, for example, I can give you an example of what I'm doing right now with all these sessions. So I have an issue in GitHub, open by one of the people from the community because they're testing my latest version and they come see an issue. That's perfect. I love them. As they are also using a gently eye, the issue is really reductive. It's really detailed. So I have the issue. Let's see what happens. Give me a detail of what is happening right now. It goes into the code, it reads everything and gives me a feedback. This is what's happening. Okay. Perfect. Give me some possibilities of some solutions. I'm going to also give you my take of how we can resolve this. Put everything into place and let's see what's the best offer. Give me advantages, disadvantages and the such. Boom. It gives me all the possibilities and it's like one. Perfect. From there, it depends. If it's a really big issue, I try to do something like SDD, expect during development. If not, I try to just go with it. Okay. Give me an app proposal and see the design, the solution. It starts working. And I always try to limit the scope of the work. So let's say that we have a big issue that has a lot of steps to resolve the issue. Perfect. Let's try to stop at every single step and see what happens. What's, how is going? What is the, at the output? It's performing as we want to also. And again, are you using TDD? That's during development because that's super important for me. Yes, I'm using perfect. All the end to end test are passing. All the end to end are related to the behaviors that the people in the issue are describing. That's a really good, got rid of also a really good harness. You see the, that's the greatest harness for me. So again, it's like that. I'm always part of it. And everything that I just told you is automated in GEDL. Every single part of it because that's my way of working. I try to create my own look that is this, this way of working. And because I always do the same thing, that's why I created this look, this flow, right? That aging has to follow to give me a solution. And when, when we think about people, what teams they will start with, elevate AI quality, what tips can you give them? Okay. First, don't trust AI too much. That's the thing, right? Because you're going to see these people like, for example, the greater of open cloth, you're going to see that he has 20 monitors, right? With, I don't know, 100 Asians running everywhere. And they'll say, they, I watch a picture of him. So these are the amount of Asians that I could run on my Mac mini. I had like 27 nations running. First, that's amazing. And I trust me, he has a lot of gates and position to stop all the Asian. He has, he has spent so many hours configuring all the setup, you know, all his loops and even for this to work. That's the first thing. Second, again, you have to test things on your own. You have to check things out. You have to also talk with your peers. Okay. How do we resolve this issue? This is the way that the team has to work to resolve this issue. This is a team convention. Perfect. I'm asking this because that's exactly what the AI has to use. Okay. So when you are working on a team and the team, even the team doesn't know how to work, the AI is going to make the famous AI slop. Because maybe it's not slop is the best code in the world. But the issue is that it's not the best code for your current context. And that is the problem. How the AI is going to understand how your team works or how your project behaves if you don't give it to him. That's why you have to cap the correct set of skills, right? And every single time that you are working on something and you see that the AI is making these mistakes, why don't you just create a skill for it not doing it again? Whenever this happens, this is the way you have to behave. Perfect. And share with your team. Let the team know how to behave. So it's always been the overseer of everything being there. Awesome. But I think a few years ago, the front engineering was the goal of mind. I don't know. Very big company. It's for $5,000,000 Euro for doing this. So I think now we are in a new era with agents systems and so on. Did you think the new gold mine becomes the, I don't know, I call it, AI Audita? Yeah. That's the thing, right? I mean, if I ask everyone right now, but every single person in the world, I see, I as a developer, and I tell them, tell me the problem you just use. And more than sure, the prompt is 20 words. So did I. right now the prompt engineering for me and this is going to be a divada for me prompting engineering is that for real for me is the I don't have to engineer a prompt. Whenever for example if I have to review an issue I suggest explained i just say read this issue. Good. That's it I don't have to give him a super-proposal with every single step because that's what the skill is for that's why you use a loop you create your you engineer your loop or your harnesses because you don't have to spend time doing a prompt. All the hour has to be inside how the agent has to behave so when you give it a prompt even if it is super simple it's going to be heavy like you want right? For example what was the issue before that the skills or the agents MD and the such word not the best or they were not available right? So when you had to prompt something you have to give all the specifications and all your needs and all the behavior that we were talking about previously and the such but right now it doesn't care. It doesn't really care. You can do whatever you want as you have put some time previously right? You think about what you thought about how to use it and from there it's just going to do whatever you want. You're not going to spend time doing prompts you're going to spend time thinking that that's what we want right now and that is another issue and when you tell me what do you feel what do you think about the eye right now and I say okay I'm excited but I'll also tire is because right now it's what I always wanted. I was always someone with a lot of thoughts a lot of ideas a lot of I want to do this but I don't have time right? And now is I have all the power in the world to do it. I can do whatever I want but the thing is that I doing a lot of things at the same time and that's tiresome really tiresome. Even when you're working for example not only my things right but when I'm working on my company I'm a problem. I'm working at 20 things at the same time okay I'm creating these reports at the same time going to review this PR at the same time and going to do everything right and that's it takes a toll right in your brain and how things are working there is difficult so whenever you see that people say okay I'm going to use AI so I have more time for myself that's not the output you have more time to do more things that's the thing there was an explanation of the CEO of Gloubant that is a really big company what more it's starting in Argentina in my city more de plada and exploring okay previously you have a client okay and you have this amount of work and now for doing that you have less amount of time so what do you do with the other you know quantity of time you do more jobs that's the thing you create more work and it's really difficult and that's is the market right now even if you create a company that is a software factory you have to produce a lot more to reach exactly the same benefit as before because other companies are doing the same thing are you competing on how I can have a bigger output with less cost but the thing is that there's a human cost in here right so it is difficult it is a really interesting challenge we could say yeah thank you well I called a iA harness engineer become a common as deaf ops I think it's true yeah it's true what's more what's more get hell it has a release a new function where you can just use natural language to set up everything for your project for the from the continuous integration the the working actions everything you can just use natural language for that and behind it you have an age doing everything so yeah I think I think it's gonna happen like that yeah and for people they will look more deep dive in an a harness engineering are there frameworks learning material or tools for them okay that's difficult why because right now I'm trying to be 100% honest 100% honest let's say that I create a video right now okay and explain harness engineering and loop engineering and everything in one week is gonna be updated in one week and for example do you remember Ralph loop the theme of wrong flip that is where it was named after a character of the symptoms well yeah exactly you it was difficult to remember that right but it's not it has not happened a long time ago for real but it's such a thing that it's just the time pass right there are new things so whenever you say no because you have to use Ralph loop no one use it anymore no one it was more codex it was trying to have what it has a functionality that is a slash go it's a rough loop it tries to get you in the solution iterating like crazy and expanding tokens like crazy okay and people are 50% using it not a lot of developers are using it so it's the same thing what do I teach about Ralph loop or the basics of everything that's why you have to see so in my case I have a book where I'm doing all the teachings this the amazing amazing book gentleman brum that's it if you search in google it's gonna have a appear is free open source and online okay you can even download the PDF from there and I have a lot of content on this subject apart from that I really really really a try to explain people to learn the basics don't go directly to AI try to learn about patterns about architecture for example clear architecture like some architecture this kind of stuff or and even even some let's say a XP development or a the scrum know because I hit scrum but you know this kind of of a soft skills right kind of titles because right now we are doing a scrum of Asians 100% we're doing a scrum of Asians we're trying to get Asians to work like humans and that is the orchestration that currently we have like for example a right now there's a concept that is called orchestrator minion right that having used it for month until they have the name that's the other thing now he has names everything is even named so in this aspect you have an orchestrator that is the one which you talk to it learns what you do you want and it orchestrates sub-Asians to work and do all the intrinsic ways let's say the details of the implementation they report to the orchestrator and the orchestrator talks to you okay this is what they have done okay so this is nothing nothing of the difference of what I know a construction company does right you have the orchestrator that is the person that manages the the workers and it's going to tell them okay here you have to create this wall here you have to graze this floor perfect you have to do this way this way this way perfect report to me when they're done and when everything is done it reports to me if the for example it's a the orchestrator finds that there's an issue if it's a blocker let's say I know I don't we don't have more concrete to continue working okay that's an issue okay that's an issue that I want but okay one of the workers did the wall but it did on the floor so that's an issue yes exactly I'll go fix it what do you want for me okay that's the thing so it's exactly the same thing it doesn't change it doesn't just it help people relate but we are trying to get that into Asians so if you learn about soft skills it's gonna be super beneficial for you I'm telling you so if you want to learn from this read x yeah the platform for the mask go and read x go and be there follow people from the industry there are 100% all the time talking about new things thinking's a waste of working and you're gonna learn a lot it's the best way of learning yeah you are the first guy in the show say x all other things and oh yeah yeah I prefer x I prefer x for example I really really like dax dax is the creator of open code okay you're gonna see here sorry see in the platform he's a bold guy with a beer and he is amazing whatever you read from him is amazing and I'm learning a lot from him a lot whenever he has a way of working let's say the other day has been a a really big fight over x of do you read your code is it super simple okay and the creator of gusty just put this phrase I read my code that's it there was a big discussion people bashing through each other fighting okay about that comment okay you have to read your code you don't have to read your code you have to read the output on the such so it's it's amazing to see a different ways of thinking about the same problem and what are they response into it right so it's great super great yeah awesome yeah I have to I was a long long time it was I was a long time not not on x because it's well sometimes I feel it was a little bit I only see out the mask every every pull I don't know how I see what you say that yeah so yeah but I hope they have been there better algorithm I have to check it out or or follow the wrong people so I have to check this very good all of my bias yeah and when when we we look at all all all all these amazing AI tools and I don't know so so many pop up and and the companies invest so heavy in in in in and yeah they make and video people not to get the cheapest guys in the world so how did you see the investment in AI and will you think there is one company that's yeah that's will win the race oh I don't know I don't know it's really difficult because the company I don't like is the one that is making the biggest let's say influence or yeah it has the advantage on companies on companies but it's losing developers the thing is okay I'm gonna say I'm saying it publicly I think for example this here on tropic and tropic is winning the company race the company race that companies you know they pay for tool for the people and the such and they're winning that aspect because it's doing what Google does Google it's everywhere right if you open Google Drive if you open Gmail what do you find Gemini you have Gemini everywhere and then you can't work without Gemini and the same thing with Copilot because if you used a Microsoft tools you're gonna see that you have Copilot everywhere and Chloe is doing the same thing now okay and tropic is doing the same thing you have clothe work clothe work space it is or co-work I don't remember they go work well where you can work sorry you can make clothe work your computer for whatever you need and you can share that with another person you can share it with a peer so for example you're in a marketing team and you are working on a way of delivering a response to a client as soon as possible with certain complexities and say you can share that workload as this is the way of having working with this and you can share you can you can work in it it's great also for creating documentation I don't know those kind of stuff is amazing but on the developer side I don't know I think the most or real expensive real expensive for example I again I'm a Latin American so when I see that I'm you try to charge $20 for or 20 euros in my case for a tool for just two problems it's crazy okay the $20 subscription it doesn't do anything it's just two or three problems and you're done that's it and and for example GPD that I really like the models I really like also how cheap it is for me for me it's really cheap it has almost the same performance in if not better because if for me it follows better instructions and for the $20 you can do a lot you don't do a lot or even if you go into open search territory you don't yeah you're not gonna see that when or I don't know or a GILM are going to win a company race but the developer one at least in Latin America everyone is using GILM 5.2 or QAN or I don't know or Minimax or those kind of models right so again for me are two different options but company wise is always the same three is Google, anthropic and open AI they are the ones for the glory going through for the glory yeah I think in enterprise there's a Microsoft have had have this enterprise company so they had a really good starting point with the with the start with the baddest product I think from my perspective but yeah and but yeah also in the large large-length long large language models are interesting and Microsoft as I see they have also now one I never have to check it out it's called my route so on yeah have you just not yet not yet I have been reading about it I haven't read about well Microsoft also sorry not the big three the four okay the usual four Microsoft also but in this case I have not seen it I haven't seen it but there are a lot of companies that are trying to mix with let's say AI companies are mixing with let's say client companies when is it client for me is for example an IDE like cursor right you know Elon Musk are mixing now with cursor their own cursor I think and they are implementing these new models this flagship new models with GROC and they are reaching fabled let's say reasoning but at the fraction of the cost so again is the fraction of the cost that in any parts this is important for me so I think a lot of companies are going to do the same I'm going to release more models they're going to release you know drain to bowl but I think that right now the reason is not the issue but the resource consumption that I think that everyone has to be working on yeah I see a lot of people say or I I don't know the best images is this the best video is yeah yeah yeah according to this for I don't know writing emails is this I see a lot of people starting running on their own systems because they don't need for everything these these big models so I think that it's could be also be something that we are getting in this time a little bit back from from the cloud era back to the to the classical PC yeah yeah by the cost of yeah let's it's damn hard yeah let me do something really quick right now I'm going to have a tool that is from the greater open-cloth that is called codexpar everyone install it is amazing and the here I can see my weekly and my five hour limits how many resets do I have because open-cloth is giving lima resets like crazy and I can also see how much I will have to pay if I was in a subscription and that's crazy because right now with the amount of work that I have been doing with a $100 subscription but I have the $100 subscription I will spend more than $600 so yeah the motorway is crazy and for example with a quarter of the work 500 for an tropic and again I have been using it at all so yeah crazy yeah today I see a guy that had developed for the Mac Mini and in this player you can add it on that shows you the cost you're actually running so yeah yeah cool cool cool cool cool gadget but what did you think will will be the future of developer how do do do's look this will will I I don't know uh yeah uh make this just jump in the future I don't know less less important or more important or I think it's gonna be just like a keyboard for real but for me for example when you think about coding you're of course you think about code but how do you make that code with the keyboard right the keyboard is the the tool for an end AI is going to be exactly the same thing I have a coworker he doesn't use a keyboard he just speaks to it and it's funny because you are talking with him as you say okay yeah we have to fix this yeah let me try to search something and from nowhere he meets himself and you see him moving his hips and then he amudes as I yeah this is what's happening okay what what have you done right there no I was picking with claw I was okay thank you go for it you know it's crazy it's crazy people are leaving things that we have used for a lot of years for the new stuff and things are gonna happen but the main change has to be that people at least at least have to stop being coders they just have to start being programmers for me programming is not coding then coding is a part of programming yes but it's not the main thing the ideas the concepts the the issues and their solutions that are what programmers really aim for what to do for a certain challenge okay what are the best possible outcome that I can create with this requirements these are the things that are going to differentiate really good developers and you know people that are just there I yeah I work at coding yeah what I found really interesting I do every year yeah look at the LinkedIn graph the the the knowledge graph and you see how different jobs are changing what skills are companies searching for and you see especially in the development and architecture IT part it's it's that it's soft skills are are really come so important I think it starts with for for 10 years there was the obligatory teamwork and then hard skill hard grads grads because now we have a lot of soft skill they come when he's searching for that's really really interesting yeah yeah you also teach soft skills how how can people learn or adopt soft skills in in in their work okay I I have one way of thinking that is super simple at least for me it makes the basics of what soft skills really are and it's being a good person being a good person yeah let's write to some rise some of the soft skills that companies want the ones I want to know have to teach perfect being a good person you want someone to that learns fast okay you have to learn how to ask right so you have to have that requirement again if you ask nicely being a good person then you have to understand your challenges and you have to share with the team again it's always trying to be a good person what do I have to do to have a great set of of soft skills think about what is the correct way you know aiming for for this target that is again having an outcome for example the main issue that is happening right now in all companies is communication communication is really difficult right now even in small companies having people from the team communicating with other people they are remotely working about a subject is really difficult yeah I just want going to say something that happened to me when I was a because I also teach companies how to use the AI correctly and there's a phrase that is stuck in my head that is no I don't ask my peers any kind of questions so how do you know what the issue is and even if you have the name of the person that made the issue why don't you ask it because I just use the AI for that I ask the AI really code and tell me what is personally but that's it I say no that's not what you have to do you got to ask the person you have to relate to him you have to you know share again if you stop sharing the AI elucinates it makes you elucinate you have to talk with your peers you have to talk with your team you have to talk with a proud that you have to be part of things you you can't just be a mediator right the person in the middle that's perfect we also have to be on front of it you have to lead what you want so again the soft skills are one of the most beneficial set of skills that you can have for real yeah so I have an every a session a quick fire round I asked some questions and you say short answer what what come in your mind what was the coolest project you ever built oh gentilai that's amazing yeah yeah it is what doesn't let me sleep for you in the evenings what is the one technology company says you get all the money and resources you you need what to build or what what what will you build I will try to make something well maybe this is a strange but I will continue doing what I want I mean doing what I'm doing okay with really are your own tools but I would like to have people working with me on it because right now it's really difficult doing open sourcing it's really difficult I don't know if you have seen already about open sourcing's death because everything is being made with AI and you have a lot of AI slot if you got a really good community it's not a slot it's ideas they are being shared that's what you have to see right but it's really difficult for me to say okay this person is great I trust this person I'm going to make him a maintainer it's really difficult because they also have to have time and they have to share your let's say your point of view of what you want right and let's be honest does a job 100% a job and you can expect being so let's say into the project as you okay I created it of course I go into spend a lot of the time on it because I love it this part of me my baby but I can expect the same for someone else so if I have all the money of the world I would have people join me to work on this brush it and try to you know increase it to do amounts and no one expects coffee tea or energy drinks during the development coffee coffee coffee 100% yes Mac Vanlos Alinux and Linux Rockos or Kubernetes? Grownies. YouTube or Twitch? YouTube. MVP or GDE? Difficult. In the middle because I like things from one and from the other things are these like between them is difficult is for real for this one is a point in the middle. Angola or React? Angola for now yeah. Barcelona or Silicon Valley? Barcelona. Box or podcast? Box. Yeah. One sentence that describes the future of AI? Let's see. Okay this is one that is with a lot of effort okay and hope and trust probably human in the loop if that happens I'm wow I'm I'm golden but it's not gonna happen people are going to go with it yeah yeah but that will be my reference for the year. What's the next for gentleman programming? I would love for it to be something great like the harm in your mother it will be legend. Very. Well yeah I always end up every episode with the same question if every IT professional in the world could remember just one lesson from today's conversation what should it be? Okay it will be that people you have to learn trusting yourself okay that you can do it that you are great that you have the knowledge to say no this is not what I want right that is the most difficult thing you're gonna see that the AI is going to shass I don't know it's going to to to Tommy Jaggand you with a lot of things a lot of options and it's like it's super in gelatin and your nothing is totally opposite you know your things you know the context you know what you have to do how to do it use AI as a tool okay it's not your boss it's a tool an assistant if you want. Yeah yeah yeah so and thank you for for joining me today this has been in critical conversation not just about AI but about engineering community building leadership and the building technology that helps people instead of simply chasing the latest trends so yeah thank you for staying with me it's was amazing thank you thank you so much Miko I really hope that everyone liked this episode and see you in the next one [BLANK_AUDIO]