Cómo uso AI como Educador Técnico

Quiero comenzar esta charla leyendo un par de comentarios que recibí recientemente. Algunos dicen, por supuesto que te encantaría porque no eres un desarrollador junior. Otro dijo que toda tu publicación parece muy fuera de lugar para cualquier persona que esté entrando en la industria. Y otro dice, sí, solo dices eso porque ya eres un desarrollador senior. Todos se están convirtiendo en codificadores de vibraciones. Estas cosas están reemplazando a los desarrolladores que realmente están haciendo su mejor esfuerzo. Estos son comentarios reales de desarrolladores reales sobre mí. Entonces, ¿qué los causó? Bueno, esto. Vamos a leerlo juntos. Dije que la IA arruinó la codificación, es la opinión más candente en tecnología en este momento. La IA tomó lo que odiaba y dejó lo que amo. Y esta es la mayor diversión que he tenido en la codificación en los últimos siete años. Esa es la publicación. Y a los desarrolladores senior en X y LinkedIn les encantó, pero a la gente en YouTube no.

Fui y leí todos estos comentarios, y mi instinto fue simplemente defenderme. Pero en su lugar, me senté con ellos, y tenían algo de razón. No sobre que la IA sea mala, sino sobre esto. No tenía realmente una respuesta para ellos. Solo tenía una opinión. Tenía una década de experiencia como alguien que ya había construido la base. Pero no tenía algo que pudiera poner en sus manos y decir, aquí, así es como realmente usas la IA como junior, desarrollas tus habilidades, entras en la industria, y aún entiendes lo que estás entregando al final del día. Solo tenía una opinión sin un marco y sin pruebas. Solo mi experiencia.

Así que me quedé en silencio, y fui y construí la respuesta. Y estoy aquí hoy para mostrártela. Este video fue publicado hace unos seis años. Era una aplicación MIRN de pila completa. Aproximadamente 1.6 millones de vistas y 135,000 horas de tiempo de visualización, con una duración promedio de visualización de unos 5 minutos. Eso es normal. Eso es bastante bueno para contenido de desarrolladores. Fue uno de los mejores videos en ese momento. Pero luego, mira esto. Este es el video de desarrollo agéntico que publiqué hace 30 días. Medio millón de vistas, más de medio millón de horas de tiempo de visualización, con una duración promedio de visualización de una hora y 8 minutos. En esta economía con la atención de TikTok, eso es mucho para contenido de desarrolladores. Eso significa que los desarrolladores se sentaron y realmente vieron el video por más de una hora. Y obtuvo estos números en 30 días. 600,000 horas son aproximadamente 70 años de atención humana en un video en un mes.

Así que no te muestro estos números para presumir. Te los muestro para hacer un hecho de que algo cambió entre esos dos videos. Y ese va a ser el punto de esta charla. Para averiguar qué exactamente cambió. Así que hola a todos. Soy Adrian, el fundador de JavaScript Mastery.

Uno de los canales más grandes en YouTube con más de un millón de suscriptores. Al menos en el nicho de JavaScript. Y también el creador de JSMastery.com, un canal de educación para desarrolladores. Durante el último año y medio, he estado obsesionado con una pregunta. No es cómo usar la IA para enviar más rápido, como muchos de nosotros hemos estado preguntando. Sino más bien, es esta. Si la IA escribe el código, entonces ¿qué diablos se supone que debo enseñar? Esa es la pregunta que me mantuvo despierto por la noche. Y responderla es lo que produjo esos números que viste.

Así que un poco de historia. He estado haciendo estos videos durante unos diez años. Tutoriales basados en proyectos de pila completa. Donde trato de enseñar a los desarrolladores cómo construir lo que parece ser un producto final. Y enseñarles el proceso detrás de ello. Codificamos todos estos videos a mano. Eso significa que los guié a través del proceso de cómo realmente construir. Y en el camino, usamos muchas herramientas de desarrollo diferentes que puedes ver en ambos lados justo aquí.

Y resolví cada error a mano. Guiando a las personas a través de ello. Eso significa alrededor de tres a cuatro semanas de tiempo de desarrollador para cada uno de estos. Una o dos semanas de grabación. Y finalmente, un par de días para finalizarlo todo y combinarlo en un video final, que dura alrededor de cuatro a cinco horas de contenido final. Estaba orgulloso. Abarcaba un éxito real durante los últimos años. Y recibí cientos de comentarios, incluso decenas de miles de comentarios de personas diciendo que les gustó. Y que les ayudó a mejorar su carrera. Lo cual es genial. Pero luego, la IA sucedió.

Así que nunca fui un escéptico. Mantenerse a la vanguardia del desarrollador, echando un vistazo a lo que están haciendo estas herramientas, es literalmente mi trabajo. Los desarrolladores y las empresas realmente no tienen tiempo para investigar cada nuevo modelo o cada nueva herramienta que llega al mercado. Eso es lo que hago. A tiempo completo, para poder compartirlo de vuelta.

Así que antes de que el desarrollo agéntico y la codificación de vibraciones fueran siquiera una cosa, cuando la IA era sinónimo solo de la ventana de chat GPT, y cuando los agentes de codificación apenas estaban naciendo, yo estaba lentamente probándolo todo. Y esto ha estado sucediendo durante unos tres años. La exploración comenzó cuando visité la GitHub Universe Conference en San Francisco y vi cómo presentaban Copilot por primera vez.

Así que esta es la primera vez que se lanza un agente de IA. Y luego, durante los siguientes años, hasta este momento, construí con ellos, los probé y rompí cosas. Fue impresionante todo este tiempo. Pero en este punto, era un poco escéptico. Como, todavía no está cambiando cómo trabajo. Claro, puedo hacer algo un poco más rápido, pero sigue siendo solo un asistente.

And then, I personally started noticing a change around December last year or January this year when Cloud Opus 4 Syncs was released. Overnight, every serious developer and founder I knew stopped arguing about whether it's possible to build with agents and just started building. I'm pretty sure that a lot of you have started noticing that shift around the same time. And I see hands of people who started noticing a change about the start of this year with more people being more open to agentic development. And they started shipping real code. So a lot of things changed at that time. AI obviously didn't replace me, but it replaced how I learn, how I build, and eventually how I teach.

So the question was, how do I do that on a channel that watched me type for seven years? Well, I first had to bring a change in myself. So I started using AI on my own projects. First, I started with small things like a date formatter, validation schema, dummy data, you know, the small, easy stuff. But then, I tried it on a search page and tested it. It worked. And then, an API route, which also worked. And eventually, I started developing full features with it. And it kind of compressed three weeks of works into about three days. Pretty consistently.

But on YouTube, I was still teaching things the old way. Writing every single thing from scratch, pretending that AI didn't exist. It felt a bit disgenuous, because that's not how I or my team develop applications anymore. But that's kind of what I was doing so far. I told myself that developers needed fundamentals. So I purposefully didn't show them AI, as that would skip the parts that build the real understanding. But that was just a lie I was telling myself. Here's what was actually going on. I was scared.

Because for seven years, my entire job, the thing that I'd built a business around, was teaching people how to write code. Line by line. Step by step. And if I was now doing the writing, then... If AI was doing the writing, sorry, what was I then supposed to teach? Well, if I wasn't teaching coding, then what the hell was I teaching? That question kept me up, because if the answer was nothing, then a decade of work was over.

We're now living through the most rapid change in the history of software development. What it means to be a developer has obviously shifted overnight. And humans aren't built for that kind of speed. It's okay if it's disorienting. It's disorienting for me, and I do this for a living. But eventually, I figured out that the answer wasn't nothing. The answer was more. There's now more to teach. Not less. Not the typing. But the thinking behind the typing. You just have to be willing to throw out the format that got you here and build a new one.

So, that's what I went and did. Six-file methodology was the thing that was used in that senior engineering video that developers watched. We covered the brief architecture, the UI, the data standards. We essentially taught the agent how to behave like an engineer. Essentially, we're forcing the engineering and the thinking up front. So, before the AI builds the UI, it understands the UI. Before it implements the logic, it understands the logic. And before it ships, it understands the code standards of your project or the ones that your company uses.

So, no longer it was a guessing game. It was executing against a spec that a developer wrote. And that was the breakthrough that the senior engineer video proved. But six files weren't the final answer. Every new video we did, we kept iterating on the process. We added what the original six files were missing. The edge cases, the recurring patterns that happened. And we kept adding more.

Today, that system is nine files. But it could be something else in the future. Nine files already was a lot of overhead for every project. So, we built smaller, sharper tools on top of it. In form of agent skills. That we can invoke whenever they make sense to use. One skill was for architecting a new feature. One for reviewing the code AI just generated. One for teaching a specific topic in a way that a viewer can absorb. Each skill, in a small, focused process, you invoke the one you need for the moment you're in.

And that's what we used in the most recent videos. That methodology became modular and something that we constantly improved. The latest experiment, the one we're still running, is teaching AI to write its own prompts. So, instead of me crafting the prompt, the perfect one prompt, by hand, the agent collaborates with me to write the spec. It asks the questions that a senior engineer would ask. And then it writes the brief. And finally, once we have the spec, it builds against it. So, that's the system that I use to build every project on the channel. It's all open-source, and I'll share the link at the end of the talk. But that's not really the point. Prompting harder isn't a discipline. A real engineering process is.

So, this is also where most companies that I talk to are stuck right now. They've given their teams access to AI tools, but they haven't yet given them a way to actually use it at scale. These tools aren't the bottleneck. The process around the tools is.

So, the next part that I want to talk about kind of dives a bit outside of just developer niche. And it actually has to do about a conversation that I've had with my wife. While I was venting to her about all of this, she told me about something called Bloom's Taxonomy. It's a module, it's a model that educators have been using for 70 years to think about how people actually learn. And as she's a psychologist, it made sense to have that conversation to see how people can actually absorb and learn the new stuff.

So, there are six levels to this. At the bottom, we have the Remember layer. Memorize the syntax. Memorize the API. Memorize the framework. Above that, there is Understand. Know what the code is doing. Then there's Apply. Use it to build something. After it, there's Analyze. Break it down. See whether it works. Then, Evaluate. Judge it. Critique it. Compare it to different alternatives. And at the very top, there is Create. Build something that didn't exist before.

So, here's what hit me when she showed it to me. For seven years, I've been teaching developers at the bottom two levels of this pyramid. Remember and Understand. Watch me type. Watch me memorize. And memorize the syntax. Maybe understand what each line does. That was not just me. That was the entire industry of developer education. All of us.

And here's the thing. AI handles the bottom two layers now. Remember and Understand are kind of sold. The AI remembers every API. And it understands every line of code. Which means that the entire industry of teaching now has to move up the pyramid. Apply, Analyze, Evaluate, and Create. That's the thing that's actually worth teaching. And once I understood that, three things changed in how I teach.

Reason one was I stopped teaching syntax and I started teaching thinking. My job was to show people the correct way to write code. Here's the auth flow, the API structure, and follow these steps. That was useful when following these steps was the key and the skill that they needed to learn. But now, I teach context engineering. How to describe a problem that people are having before asking anyone else to solve it. How to write a spec file that captures what you actually want. How to think through a system before you touch your editor. Since that skill moved up the pyramid and developers who understood that became noticeably better. Not because they prompt faster but because they think clearer. People don't want to watch me type.

They want to watch me think. So when I was writing manually, the room was watching my hands and now they're watching my decisions. Why this architecture? Why I rejected that approach? Why I pushed back on the A output and asked for something different? That's what made the watch time go up. Not shorter content. Not simpler content. People, developers, will actually sit with you for an hour and ten minutes on a four-hour video if you're genuinely thinking out loud in front of them. They won't sit through an hour of someone typing.

So before AI, slow developers and fast developers looked similar. Everyone was bottlenecked by typing. But now it's the ones who truly understand the process got faster and better at the same time. It's all about operating at the top of that pyramid. So this is what those YouTube commenters were actually afraid of. They weren't afraid of AI. They were afraid of being judged on the output that they didn't fully own. And honestly, they were right to be afraid because that's exactly what's happening to most developers.

Here's the thing I want every junior or mid developer in this room to hear. Seniority isn't the key. Open-mindedness is. I've watched 20-year-old veterans or 20-year-of-experience veterans refuse to change and stay stuck at the bottom. At the same time, I've watched developers six months into their career operating at the top. The advantage isn't your experience. It's a nice bonus, right? But your willingness to update, especially in today's world, is. Every conference talk will tell you that AI is changing everything. I think they're wrong. Not because nothing changed, but because what changed isn't what they're talking about. AI didn't change what developers do. The job has always been the same. Think clearly, build something useful, and understand what you've shipped. AI got really good at that middle one. It just exposed how many of us were skipping the thinking and the understanding part.

So what changed isn't the work. What changed is that the bottom of the pyramid, the typing, the memorizing, the syntax, is no longer where the job lives. The job moved up. The only thing left to teach, the only thing left to learn, the only thing that matters now is the top to create.

So the methodology, the skills, and everything I covered is all open source, and the link is right here. If you want to go deeper, we're also building a full agentic course. Again, it's all going to be on the same link.

To these three developers who left these comments three months ago, wherever you are, you were right. I didn't have an answer for you then, but I do now. Thank you. And one thing which is really interesting, and I think there's maybe other people who are feeling this in the room, and it kind of comes to the crux of what some people are concerned about with AI, which is that a part of what made them love programming was the complexity-related work, that spending time diving deep, figuring out a problem, learning a new skill, and sometimes maybe it feels a bit too easy. And they've said that seeing AI take that from them has made them feel a little bit empty without purpose and motivation, and it felt like it devalued their work.

And obviously, so many people have different experiences, but they're curious about how you came across, how you experienced maybe something similar or different. Did you feel this way at all? Or maybe you changed your mindset in order to not feel that way? Yeah, well, that video that I've done, that includes those six or nine context files with the specifications written before you actually start with the project, actually resonated with a lot of senior developers at Google and Apple that told me that they haven't seen someone mention that in a long time. And when they tell people that sometimes they spend two, three weeks, even a couple of months before they write a single line of code, that nobody trusts them. So obviously, that is what is happening in large corporations. But we don't see that, especially nowadays. People just dive right in and they start coding, as you said.

But that part beforehand, the thinking, the specifications, the edge cases, the limitations of what not to do on a specific project is very important. And I think now with the help of AI, it's just easier to do it. But a lot of people skip it just because how fast you can get into the second part, which is the building.

Absolutely. We talked about that critically, thinking about the code, spending time. And I know that at least for me, I didn't just take to learning engineering, software engineering easily. I had to really, really spend time in the trenches learning this. And so it can be very easy for me to look back and see folks who are building things I would have dreamed of being able to build back then without necessarily having put the same hours or having a deep understanding in.

What should I call it, right? Developers, builders, creatives, it's tough. I'm not sure how I would go through engineering degree and go through all of those lengths to learn it when there's so much to learn right now. I believe everybody who went through that before this whole AI thing started happening is lucky, because they still have the fundamentals, they understand what it means to have the fundamentals and then build on top of them.

I like to call them builders. That's a new category of these. What should I call it, right? Developers, builders, creatives, it's tough. I'm not sure how I would go through engineering degree and go through all of those lengths to learn it when there's so much to learn right now. I believe everybody who went through that before this whole AI thing started happening is lucky, because they still have the fundamentals, they understand what it means to have the fundamentals and then build on top of them.

For all these new people, it's going to be very tempting to think that these fundamentals don't matter, or that they can skip through all of these steps and just build something, build a demo, post it on X and get millions of views. But that's still not real engineering. I still believe you need the fundamentals, you need to understand what you're doing. Only then to be able to utilize the tools that are at your disposal, but it's a tough situation right now for new developers, or builders, as you call them, to be in. Maybe they're going to only get up to a specific point, and then we'll see. Maybe then they're going to go back to fundamentals and continue their journey. I think there's a lot of answers in the AI space, which is just, we've got to wait and see and really see where everything lands as well.

We've got another question, and I think there's a lot of people who are maybe senior, who have a team, or maybe they're trying to figure out how to maintain and nurture talent. Even me, I often get parents who have kids who were thinking about getting into engineering, asking me about this. Like, how do we evaluate juniors in these remaining skills that are now maybe not relevant, or some of the new skills are becoming relevant? Because it feels like hiring quality people is not becoming easier, and I totally resonate with this. Before, it was coding challenges, and now there is almost this unspoken distrust, like, are you doing this yourself, or did you just throw a prompt in? How do we evaluate and then also maybe nurture these junior developers as they're getting on this journey? Yeah. Let me try to answer a couple of levels of that question. First of all, email assignments of getting some code done are done. You can no longer do that. You can talk to people live and see what they know. The second thing is that it's tough right now for junior developers getting into the industry for a couple of reasons. First of all, the economy is not at its best state, which I think happens in circles. It's gonna get better, for sure. Second of all, companies have started utilizing agentic development, and it's easier to get one mid or senior developer who can do the work of 10 juniors. That's significantly easier. But there's gonna be a point in time when we'll need junior developers to continue working at these companies. I wanna retire some. Hopefully, yes. Exactly. So there's gonna be a time.

And I think right now we're in this very crazy, rapid growth stage that's gonna fall down very soon, and then juniors will start getting the opportunity again to work. I love that. And I also think what you said about cycles is so, so true, especially in tech, where we're very exposed to the finance cycles of the economy and so on and so forth. It really affects that as well.

You've built your whole career as an AI educator. I'm sure you've... I mean, you've spoke about this in the talk about how you've had to reevaluate what you are teaching, moving up in that pyramid. But a lot of people are using AI as the tool they learn from. Of course. Maybe they're not even watching videos. Of course. Is it a fair question to ask about, is teaching as a skill going to be something that's replaced with AI? And maybe I'm gonna add to this question and say, are there ways that people who are specializing in teaching education, as someone who works in developer evangelism, my job is to educate people on the products that we use. What does the future look like for us in the world of AI?

What does the future look like for us in the world of AI? AI is a phenomenal teaching tool. People use it to write markdown files, learn off of those. Now I've seen more people actually generated HTML files because they're more interactive where it can actually generate lessons for your level of knowledge to go through and learn. So it's an incredible, the best tool we've had for learning anything really ever. I think video still stands, has its place, right? People will still watch somebody who went through the process. AI is still not a human, and having somebody, you know, you interact with, either that's a teacher that you're, or a mentor you're learning from within a company, or somebody you're watching, I think that will always have a place. People need that human connection. But on top of that, the experience that that person has gone through in the way in what they're teaching. No, for sure. And I think you nailed something in the head. I think platforms will come and go, but really understanding that is going to be super important. This is circling back to those comments that you pulled up, and you talked about the fact that they were scared of not understanding what they shipped, right? And I think there is a lot of fear, maybe some of it misplaced, in AI in general. And some people are scared of just not being employed. They're scared of that. Is there anything you would say to someone who's a bit nervous? Yeah, they like the tools, maybe they even use the tools, but they also want a job. They want to continue to grow their careers and join in the amazing opportunity that tech has afforded a lot of us for a couple of years. Is there any thoughts or perspectives you have on that? Yeah, I think it's a multifaceted question. Everybody, I mean, everybody wants to build using those tools, and people who don't, well, we'll have to follow and see which way is going to lead, which one is going to take. But I think it depends, it's two different things. One thing is the industry, and the big tech, and the applications, and how many job places there are. And the other is what's happening with agentic AI. I think if the market goes up and the developers continue learning those tools, there's going to be places for them, for people who want to use AI, and for people who don't want to use AI. No, for sure. Thank you. I want to make this one the last one. Yeah. So you showed us that the pyramid, almost like the Maslow's pyramid of learning. Yeah. I'm going to call it now. And you had at the bottom, memorize and understand.

And you had at the bottom, memorize and understand. And if AI does the memory and understanding part, and we're seeing that happen more and more, especially if people who are just getting started, how do we maintain the ability to critique and review what AI is giving us? So for example, I really liked your talk right before this with Mark, where he went and said, you need to understand what's happening underneath the reactor. Because sometimes that's going to affect what you're doing, even though it's taken away that for you. And I can see in a world of AI and with AI agents, that issue gets multiplied times 1,000, times 100.

So are there any things we can do, we can encourage juniors to do, to make sure they maintain the ability to review and critique output that they're generating? Yeah. I think that that pyramid was never supposed to be taken with some steps skipped, even though that's how I mentioned it. All of these steps form the base of that pyramid, with memorize and understand being at the bottom, holding the pyramid in place. So the fact that AI is doing it, doesn't mean that we don't have to either, because those parts are always going to form the base, especially the understanding part.

We have to be able to understand our inputs that we're giving to AI, as well as the outputs that it's giving us back, and only then we'll be able to critique and review those outputs. So no, we're not freed from understanding and memorizing stuff. We still have to do it in order to be able to climb the pyramid. Thank you so much. Can we give it up for Adrian one more time? Thank you, everybody.