Makes NO Sense! Why Do Developers Want Better Models?!

Transcript is auto-generated and may contain errors.

Hey folks, we're going to go to the Experience Dev subreddit with my beeping car. Um, this guy or gal, I don't know, uh, was asking why developers are cheering on when the models that get released are doing better and better on tasks, saying they don't understand why someone is, you know, happy about a language model that's able to do a change in fewer minutes than last time when otherwise it would have taken them multiple weeks to do. And so they wrote a pretty lengthy post. Um, and they uh are pretty bothered by I guess like models getting better and people leaning into this. Uh I think some of the stance maybe fair to say a majority of their stance is that uh they're quite concerned that it's just going to you know remove development jobs. So like why are developers okay with this? And I thought you know we talk through this this is I don't think this is like a new thing that people are concerned about.

Um, I don't know. Like over over a year ago, we were talking about this kind of thing on this channel. Um, I don't know. I still have the same perspective on it, so I'm kind of happy to chat through it. And the end of this person's post, um, you know, I think basically they're they're not they're basically saying, I'm I don't understand why anyone is like encouraging this and like I welcome I welcome thoughts on this, but like as long as they're good thoughts, like basically as long as you agree with me, then it's okay. Uh, which I thought was kind of funny because I think if we're going to talk about this thing, we have to talk about, you know, multiple angles of it or else it's not really a a productive conversation. Like, do you want an echo chamber or do you want um do you want to discuss different perspectives on it?

And I get the irony me saying that, right? Like, I'm making this video. Um, I am my own echo chamber. Uh, but I need to kind of challenge myself to talk about it from different angles. And so I we'll try to do that. So I'll start with taking this person's perspective, right? Like I get it. Um you know, if we're able to Man, I have this stupid cable is like just reaching my shaker bottle and I can just hear tap tap tap tap tap. Um this person's perspective around like having models become more efficient to be able to do more of the work that we're asking it to do. Like yeah, I would say on the surface I I absolutely think that makes uh requiring some people to do certain types of work like a complete thing of the past, right? Whether that's entire types of work or or not like maybe that's more of a you spend significantly less time on it.

like uh we could someone who might have had to be like way more uh skilled or trained in an area um maybe is not required because someone else could do that type of work while balancing it with some other tasks kind of thing. Um I think we'll see a lot more of that personally. And so again, yeah, the side effect uh could very well be that um particular types of work are are replaced, right? You don't you don't need people or as many people doing that type of thing, whatever those things are. Um because AI is just able to do it, right? So like I don't know. I've I've given co-pilot things to do that are full features and it's been able to do it sort of one shot. I've done the same thing, you know, and it's delivered something completely wrong. But for sure, like I' I've absolutely had it do work for me that would have taken me significantly longer before.

Um, and I didn't have to hold its hand and it did it exactly how I wanted. um not only from like a code perspective kind of what I would notice like as a software engineer um but also in terms of like the feature that was delivered it was like this is exactly the functionality I want. So code that was good uh test coverage for regressions functionality that behaved exactly as I want and it was effectively a prompt inside of a I don't know like an environment that has other guidance and things in place. So was able to to do that great um very well and saved me lots of time. So with that said, someone who uh is less skilled than me, someone who has less experience than me uh in particular in the area that I was having this feature built with a similar prompt which was probably pretty lightweight and said like build me a feature that looks like X could have achieved the same thing.

Right? I've had um experiences where I have say an example project in terms of like maybe the UI or um even like the architecture right saying like I have just to give you a really simple example having a uh a web app so client server right and um it's also you know dual deployed uh or dual release is like a an electron app so it can be you know run to seem like it's a desktop application and having like an example app like that that I've built where I'm like, you know, I like this pattern. This feels pretty good and had co-pilot effectively like one shot um like cloning that to start a new application and then start filling it out, right? So, like was that prompt that I gave it, you know, ridiculously complicated and like lots of thought put into it? Like, no.

Uh, so my point is that someone again with less experience in that space who wanted an app that looked like mine could have done what I did, which was like, "Hey, co-pilot, see this code? I want to use something like this to build another app. Please go do it." Right? like the the prompt not a lot more complicated than that and it was able to set things up so saved a ton of time right again required skill level and everything lower um and so if we lean into what this person's saying like does that mean that the more and more that I'm able to do stuff like this the the less and less you know it's required that we have people that understand software engineering right the the models are getting so much better now, right? They continue to get better. By the way, I don't know enough about this to be able to say like, "Oh, by the way, there's a plateau coming like tomorrow." Like, I don't know.

Um, they seem to keep getting better. Great. I'm going to keep using the better and better models because because if they're doing better, I want to use the better tool. Um, but to this person's point, does that push us closer and closer to to like, wait a second, I'm not needed anymore? Um, and so like I think I think it's an interesting like perspective to take and I think a lot of people have that perspective. So, I get why this person's kind of frustrated, but I also think that it's like kind of silly to like uh I don't know to like fake invite people to share perspective and then and then be like as as long as it's, you know, the same as mine, it's okay. Um, when I think about this though, I'm not I'm still not fully convinced that that just eradicates people from from the job, right?

I think there's lots of examples of companies trying to do this, right? Companies got to make money, right? So, like, how do we make money? We got to have lower costs that we have revenue. How do we lower the cost? we drop the, you know, the expensive cost which is people in particular, the developers, right? Let's get rid of the devs. We have AI. You know, we're just prompting the AI now, baby. We got the we got the software getting produced. Like, what could go wrong? Um, there's lots of examples of things going very very wrong. And um it's I don't know like maybe there is a point in the future uh the models continue to get so good the tooling around the models get so good that you just genuinely require less technical people to do technical things.

Like I see that as a direction, but also like what's a good way I'm trying to think of like a good analogy to this like um it's not a perfect analogy, but just to give you an example like I don't know car mechanic, right? So I'm driving a Mercedes AMG GT right now. There are basically dedicated master mechanics for uh for AMGs and Mercedes, right? There's people that are like literally have like a master tech certificate and so like they would know a tremendous amount about this car, other Mercedes. Um, and so like if they needed to, they could go figure a ton of out about this car. Significantly more than like more able than I would be to go dig into this. Now, also, as someone who is, you know, uh, an engineer by trade, like I'm very curious. So, if I needed to, could I figure things out?

Sure. But, like, they're way way better, way more trained, way more skilled. They could do it faster. They'd be more effective than me. And for me to try and even catch up on a particular thing would take me a lot of time, right? So like there is this efficiency thing. Now you might argue well if the models that we have are just so good at looking you know they're trained they can go through data anyway they can find the answers they need. They can answer things perhaps. Right? So I think that you know we can start closing the gap on that for sure. But the point is like unless you have that role completely replaced, like entirely replaced by someone who doesn't even need to interact as a human, then I don't know like I I think that you're still going to need a human for it.

And it's the same thing with like usability of software, right? You can have something go build your feature and then what like how how are you going to have someone sign off if that's actually doing what you expect, right? You need people that are sort of like the expert in the situation to be able to like navigate that properly. Then where I actually wanted to go with that example was more like I'm able to drive this car, right? Like I can do that. Now, if something was weird going on with this car while I was driving it, I could still probably drive it and figure it out and be pretty safe and get to safety, but the master level mechanic would probably know almost instantly what's going on, be a lot better equipped. So, it's kind of like, you know, c could someone who doesn't have the training and the skill set and whatever navigate things?

Yes. Are you in a better position if you have the skills and the training? Um, and I want to layer one more thing onto this which is like the thought process. If you have those things, you will be better equipped. So now let me do another parallel here back with software. So I got to I got to pass some people. We're in the fast lane going below the speed limit. That's not that's not okay. One sec. There we go. Much better. Um, we don't we don't stand for that. Um, it's like unsafe because then people are forced to pass you when you're going below the speed limit in the fast lane. Okay. So if you have two software developers, right, and they both have AI tools to help them, I would I would make an argument that quite I feel quite confident saying that if someone was just better skilled in software engineering, right?

More experience, more understanding, uh better ability to debug things, more tech stack understanding, better concepts. If you have one person that is better at doing this versus the other, you give them AI tools and ask them to build software, I would I would argue you would have probably better quality software from the person who has more experience. And again, like does that does that go away when you have better and better models? I don't know. I don't think it goes away. I think that you can narrow the gap, but I think that, you know, in I I'm hardressed to think of any situation where if you have everything directly comparable and then you have someone who just has more experience and skill compared to the other that you somehow don't get a benefit. I don't know. Like again, I think you can narrow the gap, but I think you'll always have more benefit.

I think it's a matter of like how much that's worth. Now my sort of direction here is that I think the thought process is is ultimately going to be the thing that differentiates when you keep narrowing this gap. I think that this the thought process is really what matters. Now what do I mean by that? Well um when we think about like software architecture just as an example, right? Like there's there's a lot of different software architectures and it doesn't mean that one is right, one is wrong. There's only like, you know, there's the software architecture. I talk on this channel a lot about how like I love using plugins and build plug-in architectures. And I will fully admit that I don't think plug-in architectures are the best universal architecture. I just like building them and it sort of is aligned with like how I think about software.

So I like using them when I can in the projects I have. These wipers sound like farts if you're new here, by the way. Um, they're not even old wipers. Anyway, um, I think that even with architecture, there's like the thought process around like, well, why are we leaning into this architecture? Is there um, so like it's not the architecture itself, it's sort of like the rationale and the thought process that goes into it. you start with one because you have information today. Um, you know, do you have the thought process, the rationale to understand like why you may need to migrate? Do you have the foresight? Like the thought process that goes into this versus just like I know what the architecture is and what it does and when to use it. It's like it's more than just like knowing facts about things and more about like how you navigate through.

I don't really know how to explain it better than that. Um, similarly, right, if I was like, let's go back to the example I gave about cloning the app. um the thought process around how I went into cloning this app was like I think the reason it worked well for me and I'm not saying that it couldn't work well if you do this other ways but I've talked a lot on this channel around like how what I'm doing on the side for um my agentic software development is like learning the patterns trying to put them in instruction files putting guard rails in place I think that I was able to clone this app as effectively as I did and still have it like be a new app.

It's not like it literally copy pasted the code and then like was like okay like I copied the folder like just renamed some things like it it took the design of it effectively because I I think a lot of it had to do with the fact that I had guard rails in place. It was able to take the app with the guardrails with the instruction files cuz I gave it a little bit of direction on that. Oh my goodness wipers. Um, and so when it started trying to use that as a basis, it it had all of the guard rails around it to keep it on track, right? I didn't need to give it all of the prompts that built the original app because it had the guard rails to guide it. So again, like, how did I get to that? It's not because I'm the best in the world and I think that, you know, I did something so novel, but it's just because I captured my thought process, right?

Like that that's sort of what was differentiating in this case was like I captured my thought process. It's more it's not obviously perfect, but for how I wanted to build things because I've been trying to move in this direction, I captured the thought process. I could give the agent guidance, whatever. So I think that long term the thought process that we have will help differentiate us and like that's where the value is. It's like it's not the mechanical things that we need to go do. We've been we know this as software engineers. We've been automating for a long time. Like your your value is not in mechanically what you can crank out. It's it's in how you think. And so that's how I want to wrap up this video is like I still think that there is a ton of room um for moving in that direction and adding value.

Uh even as the more mechanical parts get replaced, even as parts of the design get easier, like you get more effective at being able to do the things that you're good at and where your value is. So no, um I'm still not of the perspective that we just all get replaced. And I think by the time we do all get replaced, it's like basically all jobs. And at that point, it's a different type of crisis. It's not, oh, software engineering is is gone. It's like I don't know. Like what in general is left to do, right? Do we get there? I don't know. Maybe. But I'm going to keep building cuz I can and it's fun. So, hope that helps. See you in the next one.