Don't Just Vibe Code, Learn While You Code

[00:00:00]

Matt: We've all heard of vibe coding, but the next step? I mean, let's just actually back up a little bit. What if you don't wanna vibe code at all? What if you're an aspiring junior developer or you're a junior developer looking to level up your skill, and now this AI coding thing has come and kind of jumped on everything.

They've-- It's jostled the entire kind of development world, the software engineering world, and now you're like, "Well, hang on. Do I learn these hard skills? Do I not? Do I spend time learning the hard skills so I can prompt better? Like, what, what do I do?" But then you have something else kinda nagging at you, I'm sure, where you're like, these...

Some of these AI workflows are not exactly simple. It's not as simple as opening the phone, telling, uh, you know, sending a quick prompt, and that's it. Some of it requires some setup, some finesse. You wanna make sure there's version control involved. There's a whole bunch of other things you can be doing, code review bots and all kinds of other things. And then you realize, uh-oh, I have a whole bunch to learn. So in this episode, I wanna talk, and it's gonna be more editorial-based. Uh, we're gonna [00:01:00] be ta- I'm gonna be kinda chatting with Mike, as I'm sure we're gonna have separate methods on how to do this. But we wanna have a conversation on how to learn while you're vibe coding or doing, uh, like you're actually trying to learn AI coding, sort of quote unquote, properly in a software engineering setting So how do I continue to further myself while I'm using this automated tool, more or less, to boil it down? So if this sounds interesting to you and you wanna support the show, you can go and check us out on that Patreon, leave a review or rating on your podcast app. Join us in our Discord server or share this with your friends. And if you wanna share, save up to off a Scrimba Pro plan, you can do so and use their interactive media player code editor.

You can take some courses on there, learn how to code, and their me- interactive media player code editor allows you to pause the lesson. You can play with the code live, and then you can press play and just have the lesson pick up right where it was. go and check out that link in the RSS description as [00:02:00] well as on the show notes on htmlallthethings.com.

And the details on how the link works are also in those show notes on htmlallthethings.com. And so I, I said this is gonna be more editorial based. I wrote up... I'm just pulling it up in front of me here. I, I wrote up, like, quite a detailed, um, show note, point form still so you can kinda like skim along and, and that type of thing if you're interested. But I'm not gonna be covering absolutely everything unless the conversation naturally takes us there. Uh, that, that's in the show notes. So my point is, if you wanna go check it out on our website, you know, please do so. But, I wanna start with where this idea kinda came from. So I, I, we're, Mike and I are working on a new, a new website, and I was mostly doing agentic coding.

I was using the, uh, the project as more or less an excuse to sorta get into agentic coding 'cause I've been using AI for a while, of course, but I just haven't really, you know, made workflows and things like that. I've been making little plug-ins for Webflow and the sort of little things like that, mostly with ChatGPT. And so I was like, okay, like I'm starting to do this agentic coding stuff, and, [00:03:00] you know, we're, it's starting to generate code for me. And then I realized, like, okay, a lot of the stuff that I'm generating, the website that we're making right now is quite simple from a technical standpoint, so I understand it's generated.

I understand all of it. But I was thinking to myself, you know, what if I didn't have context as to what it was generating? Like, if I was a complete beginner, I've never touched code before, I've never touched any sort of web development, and I go to it and I say, "Hey, man, like, I really need, uh, a website," and it gives me an HTML file, I wouldn't even know to call it an HTML file.

I don't know what that means, right? And that's just starting to scratch the surface. Like, we know HTML as sort of the skeleton of a website, but we don't have any CSS for styling. We don't have any JS for interactions. We don't have any back-end code running forms or doing other kind of fancy things, more web app like things.

We don't have any of that. so I thought, okay, I have this AI thing, and it's pumping out these massive amounts of code, these droves of code, and- [00:04:00] What if I don't have that context? What's gonna happen? Like, I'm not overwhelmed, but I could see someone else becoming overwhelmed, whether I've never written one line of code, and it just generated me an HTML file with like 700 lines.

What is this? What if it generated all three, HTML, CSS, JS? What if it did that? What if it, what if it took off and started doing Node.js stuff? What if it started doing that? What if it pulled in React, and it just starts going? then I thought to myself, I thought, "Okay, well, a person that's in the be- in their kind of beginner stage would have to hit the brakes, and they would have to say, 'Okay, I'm gonna spend some time learning some hard skills.'" But Mike, you pointed out in a previous episode that that feels like wasting time, 'cause you're like hitting the brakes, you hit the W3 schools, you hit the MDN, you hit some forums, maybe some Stack Overflow, things like that. You're hitting those things, those old school things, and then you're, you know, you're learning these hard skills, and then you're not learning any AI skills at that point.

And then, and then now, now you're like, "Okay, I kinda generally understand this, but I'm not even gonna really use it. I'm not even gonna [00:05:00] put it into practice. I'm not even gonna gain that knowledge in like a solid way where I've practiced it for a long time and kinda honed my skill to the point where I'm gonna remember it, then I'm gonna go on and I'm gonna tell this AI bot what to do." So my thought is, and this is what I do, I think of, I, I think of it this way. We're using AI to code things because it is efficient. I know that's... Uh, some people are gonna say that it isn't, whatever, but I find it quite efficient. So we use it to, to, to code things. And there's sometimes it'll get stuck, and I have to go in there and fix it or go in there and kinda give it a, give it a whack, and then it'll start, you know, kind of, kinda like the old CRT TV where you just give it, like, a bit of a knock and then it'll keep working for you kinda thing.

You kinda, kinda gotta do that sometimes. my thought is, is we're using this super kind of advanced piece of technology to make all this code. Why not have it be the teacher? And so that's, that's what I do personally, is if I'm reading through and let's just say I... Like, I'm a, you know, very versed in [00:06:00] CSS, but let's just say I'm not, and I'm stuck on something that would be simple to most of the, most of us, like most people who have been in, you know, web development for, let's say, a few months at, at minimum. know of CSS and you know of the CSS display property. But someone's brand new. They go like, "What's this display grid thing? What's this display flex? Like, what is this?" their first at display might just be display none, so they'd be like, "Wait, I thought this was the on/off toggle to, like, show and hide.

Like, what is, what is going on here?" And the AI's gonna be using display all over the place among a bunch of other properties and other lines of code. My thing is literally have it stop and tell it, "Teach me this." Now, there's a, there's, there's two pauses there. The first pause is that many people will, as I've already mentioned in this episode, be like, "Oh, I have to, like...

If I'm gonna pause, I have to go almost learn the old way. I have to learn the old way. I gotta go to Google. I gotta mess around in Google. I have to, like, do this old way." It's like, well, no, you can ask the AI. So that's, that's f- that... First and foremost, that's good. The second [00:07:00] thing is gonna be context problem, where there's gonna be like, hey, like, I'm actually trying to produce production-level code.

Even if I'm producing production-level code for myself in a learning way, where, like, let's say I'm trying to make my own portfolio website as my first project as I'm learning, and I want this to eventually go to production, so I don't want my code base, like, all filled with all this garbage. I don't want all this, like, extra commenting, over-commenting, uh, extra files, uh, it...

Like, the AI doing things more inefficiently to show me a concept I'm specifically asking for, things like that. And I just tell it what I want. I'll literally say, "I don't know what that is. Do not touch the production code we just, that we just made. Make a separate," usually it's a separate directory or a separate folder. "Take what we just did- Put that in there and over-comment it for me 'cause I don't understand what's going on. Sometimes I'll say, "Over-comment it and explain it to me." Or other times I'll say like, "Take this and like, you know, make another file and make it very clearly labeled as like test or something all [00:08:00] in caps, like in the name of the file so I know what to delete later, and I want you to take this and explain it to me." And I use this advanced system, this super advanced system, and I don't have to go back to the old way of learning. I can use the super advanced system to teach me while I'm still learning AI code. And if you use AI enough, it starts to, you know, gain a memory about you and things, especially if you're signed into the same account, and it will remember sometimes, and it, you know, it's not deterministic, so it sometimes will, sometimes won't. But it will remember like, oh, Matt struggles with the display property, the understanding of a CSS display property. noticed it'll come in and it'll like give me a little explanation, just like a little, just a little more context next time we use the display property. And I don't even ask it for that.

It just does it

Mikhail: It, it's important to understand why you're doing this. I think that's the other part of, of, of this whole thing. I think it gets lost a lot with maybe quote unquote "vibe coders" [00:09:00] because a lot of times vibe coders' goal is to just create something real quick to test it out or to just, you know, get their idea live, whatever.

And that goal doesn't-- is not conducive to slow learning. So when they hear something like, "Hey, slow down, learn what you're actually doing," they start to get almost defensive and being like, "I don't need that. That's bullshit," whatever. And to some degree I get it. Like, okay, if your goal is to literally get, like, this little application out there that does, you know, A, B, and C and you just wanna see if it works, okay, hats off to you.

Maybe you don't need to learn what a for loop is. Fine Who this is for is the people that are looking for jobs, because the reality of today's job market is this. I've talked to a few people that are looking for jobs right now, and what people are looking for is someone that has understanding of code that can then drive AI agents to [00:10:00] build more robust systems.

To do that, they will actually give you small tests. Now, apparently LeetCode isn't as popular, but it's still being used. But a lot of that, a lot of the time it's more realistic tests like, "Hey, use like the array property to merge a couple arrays together and then, like, make sure that they're all, um, uh, there's no like repeating digits.

Put them in these two arrays." That's kind of LeetCode-y, but not super LeetCode. It's just like a basic like, "Hey, do you understand what, you know, how to use the .set property? Do you understand what arrays are in JavaScript? Can you do that without having to look it up?" So for those people, that's what you're doing this for, is to get that basic understanding of data structures, of algorithms, of syntax, so that you can have that foot in the door to then be able to understand the deeper concepts and use agentic coding and, and, you know, go all the way to, from zero to getting a job to being able to build robust systems.

If you put yourself [00:11:00] in that perspective, right, then the necessity to understand how to write those b-base level co- base level code without prompting will make it so that when you're prompting to learn it, you can kind of give it more structure, right? So you're like, "Okay, hey, I'm trying to get a job at X company.

They have coding tests. I need to learn this code like so I can actually write it myself, so I don't need you to write it for me." You can tell that to the AI model. You could tell it some examples of what, you know, you could usually find online, some example problems that, you know, different companies, uh, put out.

And you can give it, give you a lesson plan to follow, structured, structured material to follow, and then you can go through and actually start learning. And what-- Another thing that I'll say is that I think this is a good, like, cutoff for the industry [00:12:00] because a lot of people are going to take the easy route.

I-- Not even a lot. I think most people are going to take the easy route to just stick with agentic coding, and they're gonna start to deteriorate their skills of manual coding. And when they get back into the workforce or back into interview loop, they're gonna have to build that back up if they wanna get in, or they're just gonna have to bounce out.

So I think there's going to be a shortage of people that can utilize these tools to actually learn something and actually take advantage of, like, their own brains, and that will be kind of the, um, gate to keep out the people that are too lazy to, to not do that. Personally, I'm kind of on the lazy side.

Like I'll, I've, I've said that multiple times on the podcast, and this is a really difficult thing for me. Like especially when I'm just in my job, I'll, you know, I'll utilize agentic coding to its extreme essentially at this point because I know that I can review the code quickly and make sure that everything is okay.

But that's going to hurt me if [00:13:00] I ever have to start looking for a job because I am specifically out of that practice of writing my own code. Whereas you, if you are able to grind through the actual learning, the, you know, the, the manual coding, the memorizing of things, that's going to put you ahead in some in, in some instances, and in quite a few instances right now because the job market still is going to be asking you these manual coding questions without AI involvement as part of the interview process.

They'll probably ask you some AI coding questions as well, but usually that will come with the territory of like learning A, B, and C. As you, as you go through all the fundamentals, you'll learn how to properly prompt as well. Um, so being aware of why you're doing this I think is super important because it's really difficult to, like, force yourself to learn this way when something easier is available for you.

Matt: Well, the one thing that really stinks, and I'm sure you'll agree with this, Mike, is that, like, this agentic coding stuff is supposed to be doing the coding for us. [00:14:00] And the, the issue is, and, and I don't really even know if I have a solution for this particular issue. Maybe this is like a, a, you know, a further discussion point, is if, if I'm, if I'm coding away...

So like I'll, I'll, I'll give an example. So like one of the, one of the things is that, that I'll be doing is I don't... When I get, when I get the AI to generate a bunch of code for me, even if there's some stuff I'm kind of iffy on, I'm not fully understanding, I don't dig into absolutely every single concept and every method and every function and everything. What I do is I follow my curiosity, and I like will look at that and be like, "What is that?" You know, "I don't, I don't know what that, I don't know what that's doing." And then I'll ask those questions and, you know, ask it to over comment in a side file or whatever, right? Like, come up with a lesson plan. Show me some links. 'Cause eventually as you get into something, you don't just rely on the AI. You'll eventually be like, "Oh, I actually do wanna read some Stack Overflows and some MDNs to like further learn this and get into it," if you're like, you know, really interested in whatever concept it is. But like for example, if you were a newbie and you looked at like a for [00:15:00] loop, you have no idea what a for loop is.

Like, it's not something that we think about in real life. And so you'd be like... Like, like for example, if you were to check every car, every door on your car to make sure it's locked manually, like a program uses a loop to do that. Our brains just go, "Oh, door one, door two, door three, door four, I'm out." program will use a, a loop.

So like a for loop will throw you for a loop, let's say, um, and you won't understand. So like my advice there is follow your curiosity. If you keep seeing these for loops or these for things, you don't know what they are, and you're like, "What is all this?" Then ask the AI for assistance and, and go through the process and, and follow, follow your curiosity. And, and like that's a good way to learn it. But we do have an issue with agentic coding even when you do this manual method With this manual learning because you'll learn the for loop today, maybe you'll learn it from today till tomorrow. Like, it'll take you two days, you know, a part-time learner. Take you two days, you understand for loops.

You're like, "All [00:16:00] right, sick." And then you keep, you keep coding away. You're not actually putting it into practice anymore. Like, you're not... Like, I've-- I know what a for loop is 'cause I've written a million for loops. know, you've done the same thing. And I, and I, I... It... That is one of those conflicts where, like, I think you're 100% right, Mike, where, you know, especially if you're going to a place that is gonna ask manual coding and then also a- AI coding, uh, what do you call it?

A- AI coding, like, questions. Like, you ask manual and AI coding questions, automated questions. If you're, if, if you're... you know the manual code, that's great. The problem is, is, like, AI almost lends itself to us l- us, lazy or not, our manual coding ability 'cause

Mikhail: Mm-hmm.

Matt: be learning, like, that curiosity.

You pull that thread. What's that for loop thing? You pull that thread, but then once the thread's pulled, I mean, what's the chances of you actually, like, long-term hanging on to the knowledge? What I'm, what I'm saying by pulling the threads and stuff is to me it's [00:17:00] enhancing our ability my ability to prompt better here's the thing. If you don't know what a server is, you don't know what a database is, you have no idea what HTML is, you don't know what a host is, whatever, mean, you are giving some pretty broad prompts, and chances are you don't even understand, you don't even know that they're a very broad and generic prompt. can come in and I could say, "I specifically want this in a MySQL database.

I specifically want this done with vanilla HTML, CSS, JS wherever possible, and if we need back-end code, I specifically want PHP at this version." And I know what all that means. Know what all that means. I know what database is. I understand all that. I understand the networking, right, of, of how the servers work, and so I am-- my prompts are informed, and I'm aware of everything around it, right?

Like, I'm aware of the server infrastructure and everything around it. And so to me, I think that, like, learning the way that I'm talking will actually help you that way. And I... And it, it will help you in the way that [00:18:00] you're saying as well, but the problem is, is, like, how do you keep that stuff with agentic coding?

Like, that's gonna be... I don't know if you have a piece of advice for that, but that's gonna be... 'Cause I'm gonna forget stuff. If I have to... Whenever I ask the AI, I get a little more context and my prompts get a little better, but my knowledge goes from I, I asked it, like, I just recently asked it, so, like, I really understand something.

Then I don't use it, so then in a couple weeks it might slightly, slightly give me more context and my, and my prompts are better for it, but I couldn't tell you what that AI, what that IP address is for. I couldn't tell you exactly what that property can be set to anymore

Mikhail: Yeah. Well, my, my method is very short-term thinking. And I, I, I really do think we have to think a little bit short-term right now because we don't know what everything will lead to.

Matt: Sure

Mikhail: so that's why I said like, hey, think of the sh- you know, what your goal is. Your goal is, if your goal is to get a job, then it sounds like you're gonna need to learn some manual coding, so you need [00:19:00] to get through that, grind through that.

But you're absolutely right in the sense of like, hey, once you get that job, most jobs right now, at least from my circle, are all gonna be agentic coding. Not only are they agentic coding, but you're being forced to agentic code and you're being told specifically to not write any code by your superiors.

So in that situation, to actually retain that information that you've learned is very, very difficult. Now, there's some people out there that have really good memories, and this won't impact as much. For me, I don't. I have a really shitty memory, and if I don't use something very often, I will lose it. So I very n- like specifically and like purposely will Only learn things that I neces- that I need at that time.

And that's how I've always done it. That's how I've always learned to code is like, hey, I'm working on a, you know, [00:20:00] GIFs right now. Like, I'm working on an app that creates GIFs, and you have to, like, you know, I have to figure out what that is. So I'll deep dive really, really far into how GIFs work, how to take apart a GIF, you know, into different images, how to put it back together to create GIFs.

I'll go as far as I need to go to complete the task, and I'll lear- usually learn quite a bit. But as soon as I'm done with that, unfortunately my brain can't retain that information unless I'm continue to work with it. So I'll purposely, like, be okay with forgetting it. And that's how I've applied myself to AI coding as well.

Like, I, I understand that I'm losing my skills in coding naturally, but I cannot retain my skills while still efficiently being an AI, you know, agentic coder. I can't do that. I don't have, I don't have enough time in my day. Like, I have a family, I have other interests, you know what I mean? Like, I can't spend every second of my day either agentic coding or then learning what I'm doing.

So I think you purposely have to [00:21:00] be, have to acknowledge that this is what's happening, and I-- you're thinking short term here, but it... We don't know what's gonna happen in the future. Like, agentic coding is evolving at such a pace that honestly, there's a very real possibility that the coding skills, like the generic coding skills that we've all kind of learned as we've, you know, made a million for loops, will fall by the wayside, and the prompting skills will still be more relevant than the coding skills.

It's easier said than done from my perspective, because like Matt said, we've done the for loop a million times. So I have that foundational knowledge that I can use to prompt a, a, a, you know, prompt the models The problem comes in is if you're just learning now, how do you balance it to still retain the foundational knowledge and build your agentic knowledge at the same time?

And the, kind of the unfortunate answer is like, yeah, you might have to grind [00:22:00] it out really, really hard. Like, you might, you can't, you might not be able to be like me and be like, "Oh, I have these other interests and I have a family." You might have to be like obsessed with getting into this industry to the point where you're splitting your time 50/50 between learning agentic coding and prompting and retaining all this really important, you know, foundational coding knowledge information on a day-to-day basis.

So maybe you spend one day a week not using AI to code so that you can retain that information. All of this is very much easier said than done, like I said, and I, I really do think this is gonna gatekeep a lot of people out of the industry that it wouldn't have before because the people-- Because we just don't have enough time in the day and we just don't have enough, like, ability to, to retain this amount of information.

And maybe that's on purpo- like, not on purpose, but maybe that's just the eventuality of the industry where only the people with ridiculous memories survive that are able to do [00:23:00] this. That might be the unfortunate reality. Uh, I'm not sure. Or it's the people that can grind it out and, and, you know, figure out how much they need to learn from the foundational side to be able to prompt just enough to be able to then, you know, evolve with the, the, the way that agentic coding is evolving.

'Cause who knows what in a year from now you'll actually need to say. Like, maybe right now you need to kind of be a little bit more directional with like, "Hey, please use MySQL for this. Use this. Uh, what is a database?" You know, "How, how does a server work?" I, I, I still think that that foundational knowledge is super important if you're wanna, if you're working on production systems today, like extremely important.

You could not do agentic coding without it. But in a year or two from now, I don't know. I-- This is why I really do think that thinking short term and figuring out what the next step is, is the play right now. If you, if you need to get a job, you need to learn manual [00:24:00] coding. If you're just doing some sort of little vibe coding project, maybe that's not the case.

Maybe it's okay to go down the route of just like, okay, learn the basics, learn the vocabulary of, of coding, but lear- just, just get the, get the project out there because you're not, you're not relying on it for work.

Matt: I'm really glad you brought this up because the way that I, uh, made this episode, like, with the show notes and how I kind of prepped the episode is more of the short-term thinking, where I'm like, basically like, as a high-level framework, it's, you know, code along with the AI, do your thing, find something that you're like, "What is that?"

The for loop or what have you. What is that? Ask it for help. Ask it to be y- your teacher. Ask it to be your tutor. If you do well with something like flashcards, ask it to make flashcards or ask it to make fill-in-the-blank questions or whatever. then I never even thought of the angle of keeping that information in your head for a long time, and it's a, it is a, it is a very, [00:25:00] like, uh, like, sincerely a very good point. 'Cause I, I think that my method of doing this is more, more so thinking of I wanna, I wanna use AI to code now and into the future, and I'm gonna have better prompts by learning these small things. 'Cause once I learn what a for loop is, I may forget some things about it, but I'll generally remember what a loop is.

I kinda get it. So then when I scan through, I'll be like, "Oh yeah, that's that loop thing," right? Um, even for you, you were talking about messing around with GIFs. Like, you g- you know that you can deconstruct a GIF. Like, you know that. I didn't know you could do that, right, until you were

Mikhail: Mm-hmm.

Matt: ago when you were messing around with that.

And so, like, even I... So even I know you can do that, and so I might literally put that in my prompt and be like, "Can we deconstruct this in some way?" Right? Like, even that is a better prompt. But it's an extremely good point that if you are trying to go for something like a job interview that requires both the manual, like, true manual coding and then, like, true AI coding, that you might have to [00:26:00] do, do the grind.

You might have to... E- even if it's the AI helping teach you the manual coding part, you really... It very m- very much might, like, very well might be, like, a huge grind. And what, what's, what really sucks about all this is that we're, we're kind of, like, entering an era right now where I was al- I... And I, this, this show's been going on for years, and in the past I've been Uh, say I- I've, I've been critical on how the, uh, and how many technologies and things that, you know, we had to or should or whatever learn, right?

I was always critical of that 'cause I was like, you know, telling a junior dev, "Learn the HTML, CSS, and JavaScript, then go over to React. Okay, that's not good enough. Now go over and use Next.js." Okay, but now the job that you're looking at uses Svelte, so now you gotta go learn Svelte. Oh, wait, there's this thing called SvelteKit.

Now go learn SvelteKit. There was Webpack for a while. You know, there's all these other things. And then have people that work for what I would call a traditional agency that don't really bother with any of this, and they're still running things on just [00:27:00] WordPress. So it's like, okay, so now what's that about?

Well, you're coding... You can code in a variety of languages, but, like, generally speaking, it's like you're coding in PHP. You're, you know, you're messing around with it in there. And, and yes, I know there's exceptions to that, to that rule. I'm talking more of, like, traditional WordPress development is, like, largely PHP, some HTML, some JS, some CSS, right?

To, like, maybe make some stuff. But, uh, like, it... Like, I was always kind of critical of the industry for adding layer upon layer upon layer, and it's almost like AI came in and was like, "Don't worry about the layers." But then when you really look at it, it's like, I'm a layer too. Like, AI is a layer as well. And so it's, it's a, it's a, it's a layer that can teach you, thankfully. Like, you don't have to go back to that old school method. You don't have to go back to, like, Google what this thing is. Like, what's this loop thing, and w- what's this display thing? What's this? Like, you don't have to, like, go and do all that.

But you do, unfortunately, need to, uh... You do unfortunately need to, uh, like, choose what your goal is, like you said. You need to [00:28:00] choose whether, like, do I... Am I just interested in this for my future prompts, or am I interested in this to a point where need to have some learning material? And I mean, admittedly, uh, like, we, we all do this in school.

Like, if you're in, if you're in post-secondary for a long time, like, I don't remember, you know, how to factor and everything anymore. Like, in math, like, we used to, you know, factor things and do Laplace and do derivatives and do all that. I couldn't do that, like, off the top of my head anymore. Mesh loop for solving circuits and a whole bunch of other stuff, and solving transistor circuits and things. I couldn't do that anymore. Uh, but I know it's almost like my prompting, at least in my brain, like, take AI out of the equation, it's like by just being aware

Mikhail: Mm-hmm.

Matt: solving circuits and that there's a thing called mesh loop, I'd be like, "Oh, okay, maybe I can use mesh loop here." Then if I start doing that research, it's like, oh, mesh loop not the method, 'cause I remember there was multiple methods.

There's other, other methods other than mesh loop and things. And so it's like just, just having that knowledge is, is, is helpful. [00:29:00] Um, but no, it's a, a very good, it's a very good point y- you made, Mike. It's a... It's important

Mikhail: And, and I think you make a good point too that maybe I undervalue that kind of base knowledge that you gain from doing something deep and then go-- even, even leaving it. Yeah, you do probably lose maybe 80% of that, or I will lose 80% of that, but that 20% might be able to s- like, you know, bring you back pretty quickly, especially with agentic coding, where if you gain...

Like, if you get the base knowledge of whatever you go deep into and you take, you know, 80% of the, uh, of it away, 20% might be enough for you to continue down that path later on, like three years down the line, like with the GIF example. I thought that resonated really well. So there is some-- or not some, there is quite a bit of value in just understanding the basics and going a little bit deeper and unders- and learning it.

Even if your intention is to not use it later on, you will retain some of that knowledge to then [00:30:00] potentially bring it back if you need it, you know, six months, a year, 10 years down the line. So th- I think there is some value in that. You-- That's a good point.

Matt: The, the other thing I think of, and like, you know, correct me if I'm wrong here, but I think that we're, we're, we're kind of... I, I wasn't even thinking of having like a, like a, a, a sectioned off episode. But I kinda think that through this conversation we've, we've sorta like roughly determined that there's roughly three kind of like camps.

And I, I wanna be clear, this is rough. So if you're the listener out there, you're like, "There," you know, "There's more," or, "You're wrong," or whatever, this is just sort of the course of this conversation. But the thing I'm thinking of is it's like the first camp is like a vibe coder, which is only want the end product.

They're not there to learn what a for loop is. They don't care what, what a database is. They don't understand what like client or server is. They don't understand any of that. It's just, "I'm trying to make a fishing app," or, "I'm trying to make a website for myself," or whatever, and that's all they care about. And then you have the person that's like the aspiring junior dev that maybe is spending their time learning [00:31:00] AI coding because they think that that's the most valuable or like it, like they, they've They don't wanna, like, go and do the old method of, like, Googling and everything. So then they're getting a little more involved, and they're just sort of, like, improving their prompting and improving their understanding generally what's-- of what's, you know, generally going on. And then you have, like, your other group, which is your goal-oriented group that would be the one that requires both the manual and the sort of automated, the AI coding as well, where those people would need, like, almost study material. Like, they would... like let's say the... Go back to the display example just 'cause it's such a, such a simple example. It's like, if they were like, "I, I just can't... Like, I always forget what this display property is, and I have an interview tomorrow, and I-- and it's about-- and it's-- and I'm gonna be asked CSS questions," it's like you can ask the AI 'cause you know now, like you've already pulled the thread while you were AI coding and you've learned, like, what the CSS is and what display is, and you're like, "I just can't remember [00:32:00] what some of the property values can be and what it, like, fully does," you can ask it for literally, like, a sheet to study from.

Like, "Hey, like, get, get this out." So then you can study like a traditional test like you would in school, and you get some of that study material, uh, from the AI. Like, you can ask it. I guess the, like, the, the whole point here is that, is that you, you, you don't necessarily have to avoid AI to do this, which is nice. Back in the day we went from, like, Webpack and then we had React and all these other things that I've already mentioned, w- back in those da- back in the day with all that, thing that sucked is, is everything was learned either trial by fire, whether that be a boss putting you trial by fire or you doing it yourself, or you're just like, "I'm learning React.

I've never touched code before. Boom, here we go." Or you could do, like, sort of a, a, a slow lesson and you could, like, kinda go through lessons, and there's some different differences in between. You could do hybrid and things like that. it was all sort of the old method where there was a lot of Googling, a lot of whatever. Now this new layer that we've added is actually a layer that can assist [00:33:00] us in learning, only learning itself, y- the AI can help you understand itself, but it can also help you understand the old, like the manual coding, the old way of coding that some people, uh, some people are calling it

Mikhail: I, I think there's another part to this as well where, again, looking at it a little bit short term, what does-- what do people wanna see? Y-y-you-- the ability for you to learn, right? Like how, how do they... If, if for example, an interview situation came up and part of the interview was, "Hey, here's a task. Do whatever you can to complete this task."

That, that apparently is a new, like, technical interview. Out-- after like a manual coding test, a lot of times you'll be moved on to this, like, task-based interview. In that task-based interview, a lot of times you'll be thrown something that you just don't know,

Matt: Угу

Mikhail: to see how you ramp up at it. How do you learn that [00:34:00] thing?

Now, from my understanding, this isn't gospel, so don't take this as gospel, but from my understanding, what people wanna see is, hey, there's this little niche thing that you need to learn. They want to see how you prompt AI to learn that thing and then implement it, right? If you go in that specific scenario, and let's say you're kind of anti-AI, and you go and you look up MDN for a couple things, and you go on Stack Overflow for a couple things.

If the spec- if the interview mentions agentic coding or you, you know that the company is using agentic coding heavily, I would highly recommend learning that if you can before, like go on Glassdoor, you know, search Reddit, et cetera, et cetera. If you can find out any information, if they want agentic coding, then I would recommend almost avoiding the traditional sites for learning and showing how you use AI to ramp up in something in that specific scenario.

And learning how you do that is [00:35:00] going to help you immensely in those situations, because someone that has obviously high experience with learning with AI is going to stand out in that specific scenario versus someone that, you know, uses traditional methods still, um, that are potentially slower. I'm not, I'm not saying that those methods are wrong, by the way.

Like going to MDN isn't a bad thing in my eyes. But where we are right now with the industry being what it is, I would lean towards utilizing the AI more versus less in those situations and learning how to do that now versus later, because that's what hiring managers are looking for. That's what technical c- like technical leads are looking for right now.

How do you prompt the AI? How do you leverage this new thing that we've been given to code faster, right? To, to build something that you don't know. Saying you don't know something is- isn't an interview killer anymore at all. It's the whole [00:36:00] process of you getting, getting to a solution, a good solution, using the skills you already have.

That's what people wanna see.

Matt: It's a good point because many people, if they're stumped, like th- that's the, that's why that question, like sometimes it's like, uh, if you don't understand something, sometimes it's like, "Oh, it's over," you know, whatever.

Mikhail: Mm-hmm.

Matt: s- with, when the internet was kind of first starting, some kind of more modern tech interviews, 'cause I've experienced this in the IT field, is they would ask you how would you figure out th- how would you figure that question out? that was kind of using the old method of like going to Google or like going here, going

Mikhail: Yep

Matt: at this, uh, looking at the manual if it was some sort of proprietary technology and things like that. Um, and, and you could also kind of have your phone-a-friend options where you would be like, uh, I, this is a security issue.

Like sometimes that's what they're looking for, especially in IT where they'd be like, "What, what do you do with this?" And it's like, well, I can't just talk to bo- talk to anyone about this. I have to go to my manager or an approved team member. And sometimes that's what they're looking for. They're like, "Good."

You would... You know, you're not just gonna go like on a forum and be like, "My IP address is [00:37:00] this and I have this on this port," and like telling everybody about your like internal infrastructure. but now it would be more of a question of how do you prompt it? And what's, what's interesting about that is in, in, in interviews of old, let's say, the Google question that I just kind of answered was like what you said, Mike.

It was either they didn't ask about like you Googling and how to figure something out, and they're just like, "Oh, you don't know this? You're out." I've had that a couple times where it's like

Mikhail: mhm。

Matt: know where a certain menu in Windows was exactly, and it's like, "You're done. Get out of here." And it's like, what?

Mikhail: And that's the, the,

Matt: one random

Mikhail: the...

Matt: is?

Mikhail: Exactly. And that might still be the case in some interviews. I wanna make that clear. It's not every interview that's gonna be so, like, kinda progressive. But yeah, keep going. Yeah

Matt: But what's interesting is like the progression there, like before AI it was like, "Oh, like how would you solve it?" And you kinda like bring up, you know, the, all that stuff, manuals, Google, and whatever. And now it's almost like that question is like a desired question, where they actually want you to get stuck because they want you to use this technology [00:38:00] because this technology helps you kinda like find your way through the maze of, of how to s- how to figure something out or how to solve something.

And so it, it, it's interesting that before it was seen as like progressive, and I guess it's still progressive in terms of like state-of-the-art tech, but it's almost like it's now desired. Whereas before it was sorta like, "Do you know where this menu in Windows is?" "No." "Okay, I guess we'll let you fall back on how would you find it?"

Now it's like, "Oh, you don't know this? Cool. How would you prompt an AI to do it?" It's almost like there's a bit of excitement there, and they actually want you, they actually want to see, uh, your workflow, like your own personal like kind of AI workflow.

Mikhail: I, I just thought of, like, a kind of a good question. Uh, a little bit of a sidetrack, but, like, I'm thinking, I'm thinking through, like, how do you see if someone is not only good at using AI but is also responsible using AI? Because I know a big problem with companies is something called shadow AI, where, like, they'll use AI and they'll upload, like, sensitive information to the AI, like company's [00:39:00] information to ChatGPT or something like that, or user information to ChatGPT, and that's kind of scary.

Um, I-- Like, asking a question like, "Hey, here's, uh, here's some customer data that you would have gotten." Like, obviously it's fake customer data, but, like, pretend that it's real customer data. We need to find some, like anal- like, analyze this data to find some sort of correlation between X and Y, let's say, and seeing how they would g- approach doing that.

So part of that problem would actually be removing the sensitive data before uploading it to, like, a ChatGPT, and that would be part of the test, right? Like, seeing if they can manually remove that sensitive data through code, potentially. Like, they could run, like, a function to remove it or something like that.

Like, they can create a function to remove it. And then using that data to then-- Using, like, the anonymized data to then actually go in and get some correlation and then revert it back, uh, if, like, revert it back so that you can actually [00:40:00] correlate it to the right customer. Like, that's ju- totally side, sidetrack, but as you were s- as you were talking about it, for some reason that just got into my mind, and I feel like that would be actually a really good, like, live interview question where, you know, they would share their screen and show you how they go through it.

Or maybe even a take-home. I don't know. You would have to, like-- The take-home would be a problem because you'd have to, like, see, like, there'd have to be a way to see the train of steps, like all the steps that they would have taken.

Matt: Screen recording, like Loom or something, right?

Mikhail: yeah, maybe something like that. Yeah.

Matt: What's

Mikhail: I don't know.

Matt: is that we're learning that there's com- there's complexities in, in prompting.

Mikhail: Absolutely

Matt: in the day, I mean, I'm sure there's tons of episodes of us back in the day talking about how to Google things,

Mikhail: Угу

Matt: Google for certain technical issues, how to, like, kinda grow that search, if you will,

Mikhail: Mhm.

Matt: like, start with, a very generic question, and you start kinda drilling down, drilling down, drilling down, and then you eventually figure it out. Um, and solve the Curse [00:41:00] of Oak Island. But the, uh,

Mikhail: Had to slip that in

Matt: fun, fun, fun fact, Mike, the, like, the pre, the pre-show, there's, like, a pre-show, like, when, when there's a new season. They have a special which is, like, usually a couple episodes called The Curse of Oak Island Drilling Down. So there you go.

Mikhail: There we go. Okay. Good to- good fact. Like it

Matt: But no, like, what we're learning now is, like, there's, like, a similar thing with the AI prompting, but obviously, like, prompts can be more power- like, can be very powerful depending on what they're, they're controlling. And so we're learning that this is like what you're saying, like, it's almost like you're, you're prompt engineering with, like, with a caveat, like, with a security caveat, um, with... there's, there can be so many. Like, with a security caveat, with a, in terms of, like, programming, you might say, "I specifically need Python. Uh, I need this to run in the then, like, the GDPR and other privacy laws get, get included and stuff like that. Um, can the person anonymize details? Like, [00:42:00] when we tell stories, I'll sometimes say, "I'm anonymizing details," and I'll change people's names, and I'll change how the project is, but you're still getting the gist of the, the gist of the story. The re- if I had to tell that story to an AI, I wouldn't wanna say, like, the actual names of the people. I would still want to anonymize it, and so, like, even that's a skill where the person has to have at least a general understanding of what's going on. Like, if they, if they keep using MySQL and they don't even know what that is, they don't know what a database is, they can't anonymize what MySQL is.

They're like, "Uh, you know? If for some reason they had to anonymize it, they, they don't even know to say database

Mikhail: Mm-hmm.

Matt: like, "What is MySQL? Okay, so we're gonna say database so you don't know I'm using M- you know, MySQL." It's just silly at that point

Mikhail: That's the thing, like you, you can, you can easily kind of just having a couple of those test questions, uh, during an interview can probably weed out like 99% of candidates honestly at this point. And it makes it really hard. Like it makes it very hard because y- again, you need to know both the agentic coding, prompting, a- and all of the foundational information.

So [00:43:00] the, yeah, the industry is getting more and more competitive. The, the people that are kind of diving in and embracing the AI coding but still foundationally like retaining that information that they learned from their existing, like from their previous work, are probably the people that are farthest ahead right now.

Um, and it's, I'm gonna bet not a high percentage of engineers, just 'cause there's a lot of engineers. Like these comp- th- these companies sometimes have thousands of thousands of engineers. So you have to think of it like if you're listening to this podcast, you're probably already in like that top 5% of people that are trying to learn, that are trying to expand their knowledge.

Most people will go to work, do their work, go home, and forget about programming. And that's, you know, no fault to there. Like that's, that's a normal work experience. Like that's how we were taught work was when we were growing up is like you, you go do your work and go home. I think that reality is starting to become more difficult with coding specifically, [00:44:00] and that's unfortunate because like there's just so much you have to keep up with now and like this AI throws an entire massive wrench.

It w- like even before AI, we were talking on this podcast a million times about how web development is moving at a pa- at, at a ridiculous pace and just like keeping up with that was kind of crazy. Now I don't think at that point it was so much of a necessity as more of like a, you know, an interest thing where like I just liked keeping up with it.

But now with AI it's becoming almost a necessity where like again, if you're falling behind on that route and it's detectable by management, they're looking for pe- for cuts. Like every c- freaking company is looking for cuts. So like now you're competing against people that are coding outside of their time and are lear- learning about all this AI stuff outside of their work time, and now it's, it's becoming this, I don't wanna say toxic, but like a much more difficult environment to just, you know, clock in, clock out [00:45:00] of.

Um, which is unfortunate, but knowing that it's that is probably better than just, you know, putting your blinders on and doing your nine to five and hoping for the best, I, in my opinion at least.

Matt: I mean, it sucks because like some people are just not, aren't gonna

Mikhail: Yeah

Matt: do, do it. Some people are not gonna be able to do it.

Mikhail: Угу

Matt: a side effect. I think this, this is like a whole other topic, but it's like, it's almost a side effect of like, it's like the, the smartphone freed us from the desk, so we don't have to work at the desk anymore. But then it's like, yeah, but still come into work and then work when you go home too then, right? So it's like, well, it didn't really free me from the desk. It, it just allowed me to work at all times

Mikhail: It giveth and it taketh away

Matt: like it, well yeah, like, it kind of taketh away more, uh, in,

Mikhail: Yeah.

Matt: cases. It can certainly be abused, let's say that.

It can certainly be abused. It can be a blessing, but it can also be abused. Like, it's not the, the, the greatest [00:46:00] solution, which is another whole... That might even be like a web news or something down the line. But But, uh, the one thing I do wanna mention out there, if you're still listening to this episode, is, uh, I wanna encourage you to, to comment on this episode, not for, like, the, the traditional kind of algorithm thing. Um, I wanna know how you are learning alongside your AI coding. Are you not doing it? Are you just saying, "I'm just AI coding, I don't care about learning any of the manual stuff"? Are you asking it to create separate projects for you? Do you have, like, a separate project where you're like, "Hey, on my last project I got stuck on for loops.

Can you make a for loop lesson for me?" Uh, are, are you making cheat sheets? Uh, are you using it just to say, "Hey, I wanna learn more about for loops, like, where can I find that?" And it just gives you a bunch of links, and you can kinda... Kind of a hybrid, like, old school learning and new school learning as well, just kinda source to hopefully surface some of the best, the best sources.

I, I wanna know and, and if we get some comments, so if you comment on the Spotify version of this episode, um, or on the YouTube, [00:47:00] um, w- might... We're, we're looking at if we get some, you know, some good responses, we're looking at reading those out and, like, having a discussion in a segment in a future episode.

Because we're, we're seriously curious because, like I, like I said, like, I laid out this whole episode, took me a couple hours, then Mike comes in and throws a huge, like, wrench into the, the thing, but rightly so. And there's gonna be someone else listening that's like, "Hey, like, here's another wrench," and it's like...

Because this is such a new frontier

Mikhail: Yes

Matt: like, we need to have the, that conversation because it's a, it's a, it's a dang good conversation. Like, we need to, you know, kind of expand our, our horizon. So yeah, if you, a- a- again, if you comment, like we, there's a chance that we might read it out in, in a segment of a future episode. Um, so yeah, please, uh, please let us know how you are learning alongside your AI or thereof, for that matter So I wanna kinda conclude this with a question for you, Mike. I wanna, you know, one final kinda thing Do you think, in a utopian world, [00:48:00] do you think that you should be learning alongside using your AI? if you do see a th- something that you're confused by or interested in, you don't just let the AI run it, you actually ask it it teach you in various methods how to, like, do that thing or what the thing is

Mikhail: I, well, I would say for sure what the thing is, like if it's doing something that you've never seen before, like you literally just like says a word and you're like, "Oh, I've never heard that word before," or like that approach, I would say you at least need to dive deeper into it and understand what that is.

'Cause a lot of times it'll just be like a, a synonym for another thing or, you know, a correlation of something that you already know maybe. So just knowing that gives you a little bit more arsenal. But like learning exactly how to implement it yourself, I would probably hesitate to advise that for everything, [00:49:00] um, just because of time constraints and the reality of life and where AI is going

Matt: you don't have those problems. Like, do you think everyone should be doing both or is this more just like, ah, just almost vibe code it? Like, I guess where in that spectrum are you?

Mikhail: If you have a foundational-- Yeah, if you're, if you're just getting into engineering, I would say lean more towards that. Lean more towards learning what it's doing and actually following and following along and be able, being able to implement the foundational stuff that you're learning. Um, as you become more senior, like let's say you have your five, 10 years of experience, uh, and you have that foundational knowledge already, I would then start to go towards more of like, hey, just make sure you're understanding what's happening from a, you know, the, the, the sense of like, okay, this, whatever it's doing, make sure you understand that, not diving into learning how to do that.

Because the reality is, is that if you needed to learn how to do that after, you know, after you have that five, 10 years of [00:50:00] like manual coding experience, um, it's not that hard to go down that path again if you need to go down that path. So I, I wouldn't spend that time doing it because it's not a necessary requirement now with agentic coding, whereas if you're just starting to learn and you don't have that foundation to fall back on, I would lean-- In a utopian world, I would continue to try to learn how to code manually alongside the agentic stuff for as much as you possibly can, right?

Like, you know, utopian world, infinite time, you, you do them 50/50. Um, but yeah, in, in a real world sense, you're gonna need to invest some extra time, you know, into, into the manual coding aspect.

Matt: I don't know where I sit. I think I sit with you to answer that question. I think I sit with you in the utopian world sense. But, and, and I think, I think you might be in this camp too. I feel like the next two to three years will tell us which

Mikhail: Yeah

Matt: in, because if it just becomes a thing [00:51:00] where there's just no reason... Because there's already talk of like why even have React and stuff, because React is meant for like human teams to like sync up better and there's

Mikhail: うーん、うーん。 For

Matt: call these frameworks and libraries, they are human interfaces to the vanilla code.

And in fact, the

Mikhail: sure

Matt: can be considered a interface.

Mikhail: Absolutely.

Matt: on. You start

Mikhail: Mm-hmm

Matt: again. I think that if it starts going the, the way of like why do we have these human frameworks and things, then I would say, okay, like we're all in on this AI coding thing. But I think right now, utopian world 2026, end of the summer, I'd say I'm in the same camp as you, where like you would wanna learn as much as you can, you'd wanna learn how to prompt as much as you can.

So it's kind of like that 50/50 split

Mikhail: Agreed

Matt: But I think that's it. That's the episode. Hope you enjoyed it. A little more editorial, I intentionally did that. I wrote a whole bunch of stuff, uh, in the show notes. A lot of it we did [00:52:00] cover, but a lot of it is just sort of, like, specific parts for junior devs and things like that. If you're interested, they'll be on htmlallthethings.com.

And remember that if you want to support episodes like this, you can do so on patreon.com/htmlallthethings. And many thanks to our $3 tier patrons, Tim from The Web Hacker on Jason from Geek Life Radio via geekliferadio.com, Garrett Segall, Level Up Financial Planning via www.levelupfinancialplanning.com, Magnus from Yes Web via yesweb.se, Syntaxify from the HTML All The Things Discord server, and Stacy Mostiller from the website swingworthydesigns.com. And remember that we have a contributing author, and he, his name is Michael Larocca. He is also the author of Self-Taught: The X Generation blog at selftaughttxg.com. And if you want to learn how to code and take some courses to learn how to code in various languages and different methods and all kinds of things, you can check out Scrimba. Scrimba has a pro plan that you can get up to 20% off using our link. That [00:53:00] link is in the RSS description and in the show notes on htmlallthethings.com. And in those show notes is where you'll find how the link works. Again, up to 20% off that Scrimba pro plan. Feel free to leave a comment or review on the platform you're listening to this on, and we signing off