The AI Doom Fatigue Cycle

Transcript is auto-generated and may contain errors.

Hey folks, we are headed to CrossFit. Got a topic from Reddit, but it's not what I usually do, I guess. Um, I went looking for a topic on Reddit to talk about this morning. And I am scrolling through experience devs and basically everything that I could see today, like maybe except for like two posts were something to do with like like AI replacement. um like people being replaced by AI to various degrees and like this isn't a it's not a new topic obviously like talked about this kind of stuff on this channel before. I'm sure everyone is fatigued from this kind of stuff. I know I am. And uh I guess it's kind of why you wanted to talk a little bit more about it. Why not, right? Let's pile on. But um I think this is the first time and I'm I'm I'm not saying it is the only time that it's happened, but this is the first time for me where I'm like it's like everything is that.

Um usually there's other things mixed in, but like this is I'm feeling like another level. Um, so you know, scanning through a couple of the threads, uh, one of them was talking about someone is, uh, you know, Seale person was talking about replacing the whole dev team because some other they're following the footsteps of some other company and they have like a 100 person dev and QA team and they're like, we'll just replace them all with AI agents and they're like convinced that this is uh, This is going to be good. Um, there's other people that were kind of talking about like has anyone seen something that was, you know, 90% plus built by AI agents without human intervention? Um, so like these kind of topics in experience devs, right? Um, someone was saying, you know, I'm I'm doing like really low-level programming. They were talking about like managing file descriptors and and memory and things like that.

Uh, I didn't even read the post, just kind of the the preview blurb as I'm scrolling through and um and they're saying like, you know, is this hurting me in the age of AI? Like every it's just everything is everything is this fear. Um, and don't get me wrong, I'm not trying to to come across like people people shouldn't be thinking this way or something like this, but it's uh it's getting overwhelming now. I've been trying to be very transparent with like well in general I try to be but especially with how I'm using AI I'm trying to like when I make these videos um or on my main channel as I'm using it I'm trying to show like this is how I end up using it and I'm productive or not. Um and I have not found anything so far that's like remotely close to not needing people.

I was actually telling one of my co-founders for Brand Ghost on the weekend. I said, and I'm not I'm not making this up. I said I I legitimately think that I may like need to find like blood pressure medication at some point because I get so frustrated working with AI agents that it's starting to feel like, you know, building is not it's not enjoyable. And it's not that it's not enjoyable because I'm not the one writing all the code. Like I know some people have been nervous about that. For me, I'm okay with that. I don't need to be the one coding it. Um, I like being part of the creative process. If something else or someone else wants to go hammer out all that code, like that's cool because that for me is something that like time cons like I as a human, I am time constrained from doing that kind of stuff.

There's times where I really want to refactor something and I'm like, but the value in doing that is just not worth the time right now. I have other things that are more important to do and I have to keep catching myself, right? Uh or like it'd be cool to rewrite this because like now that I've learned a little bit more like see some other patterns emerging, I'd like to redo that. But like it's I have to catch myself. Is it actually going to be any more valuable than doing that? Or am I just doing it to stay busy because I like to program? And it's almost always the latter for me. But when you have access to AI agents to be able to do some of that stuff for you, in theory, it should just be, well, get the prompt together, set them off, and and wait, right?

Like that's that's what we're after, isn't it? That's where all these companies are able to do this, building products end to end, right? That's what's happening. I've been programming for 20 plus years. And I'm not saying that I'm the best programmer in the world by any means, but I should be able to go prompt AI to go build things as effectively, if not more effectively, than some of these people that have never even seen software before. So why is that the case that it's so frustrating? I spent the whole weekend, the whole weekend when I was programming, I was working with AI agents and this was by far the most frustrating weekend I have ever had. And that's because the different agents I was working with, not only did I have to teach every single one of them about patterns and practices that I've already done a million times before, but the agent system I was using, the memory is broken.

As in, you're supposed to be able to get it to store memories about behaviors and um it's not persisting. So to give you an example, the biggest issue I had before with this uh this is using claude by the way um and a wrapper around it called clawed flow to have like multiple agents running. Um, the biggest issue I had before was, you know, it's I'm working I'm on a Windows machine working in PowerShell and using WSL so I can have uh sort of like this Linux interface and the biggest problem was that I would watch it try to go build my C code for like 20 different attempts. Oh, that doesn't work. That doesn't work. I can't find MS build. Oh, uh, let me try these 10 other ways to call MS build. Like files still not found. And I'm like, dude, you're in WSL.

Here's the file path. Translate this for Linux. That's it. And it then it would work, right? But the problem was the agent system I was using still cloud flow between different sessions. It's not persisting the memory. So I would wrap up doing something and then go to start the next sort of project within the same scope, same product, different project, like a different feature I'm building, and then I would have to retach it all of the same things again. And people that are watching or listening might say, "Oh, put it in your claw MD file." Uh-huh. It's there. It's been there the whole time. It's not actually sufficient. So this weekend because the memory system was broken, not only did I have to teach it between different agents, but even the same agent and it wasn't even MS build this time. It was like, you know, I have these packages might not make any sense to people.

If you're not a .NET developer, that's totally fine. But I use a testing framework called XUnit and I use a mocking framework that's calledQ and it's pervasive in my codebase. Like in my one codebase there's 1,400 or so tests. That's all that's used. That's it. and it decides when it's going to write some test files. Oh, I'm just going to go randomly pick n substitute which is another mocking framework and just add it and all of my instructions, all of my documentation everywhere say this is what we use and go look at other examples before continuing. Absolutely does not. So now it integrates and substitute. Okay, no thanks. Next thing is there's a package called Fluent Assertions which I think has moved to like a a paid model and honestly it's not really offering me any value like basically zero.

I don't know what all the hype was about personally but anyway that's in my codebase and I'm like I've been trying to strip it out and it keeps adding it back in and then I give it explicit instructions. Don't add it. Cool. Like next test it's fine. Five minutes later, it's readding end substitute and fluent assertions. I'm like, I'm gonna lose my mind here. Um, honestly, like I have never been so frustrated when trying to write code. And I when I say write code, I mean build software cuz I was not the one writing it. like I could I could genuinely feel um like physically frustrated, right? And for me, building software has always been like sure there's going to be things I get stuck on and like I'm mentally exhausted because I've been trying to problem solve and then eventually like I have a breakthrough and it's great.

This was like I I physically felt unwell because I felt very angry. And it's weird, right? Because I was talking to someone about this recently. If I rewind like one year, I never would have thought that I would have be able to generate code at the scale that we're seeing. So, it's it's truly it's really cool that we're progressing that way. Like, I'm I am excited about that. But in the current state of things, if this Like I I'm hoping that my future self looks back on sort of this era when we're just getting into AI programming and laughs because it's like I can't believe you had to put up with that. Like I'm I'm very much hoping for that kind of scenario because right now trying to tell LLMs to not make the most trivial mistakes, to not repeat the dumbest patterns on repeat, to not constantly invent new things.

And when I say invent new things, I just mean introduce new patterns into the code. Um it's it's nonsensical. And I know because this happens on every single platform. Someone listening or watching will go, "Oh, just prompt better." Uh-huh. I get it. But like just code better, right? Like there's the other argument. Just stop being such a useless AI agent and then maybe I wouldn't have to prompt as as uh specific, right? It's crazy. The other thing that I've already mentioned is that I have some very specific prompts and it just does not listen. So, I know that that's an argument that people like to use, but like it doesn't hold in every situation. It's uh yeah, a noticeably like an upsetting experience for me this weekend.

Um simple things too like you know finishing something like go write tests for this produces a bunch of code I go over to it the methods that it's testing don't even exist the code doesn't even compile like this felt like this weekend felt like the biggest step back in my AI usage that I've like felt to date and uh and it was frustrating. So, no, to answer all of the questions at once that I've seen on Reddit, I absolutely do not think we're at a point where AI is replacing people. I think that's the most the more I use it, it's the most ludicrous statement ever. We have so far to go before that's not needed, like before people aren't needed to focus as much. Um, and I still stand by it that I hope I genuinely hope that it gets to the point where I don't have to do all the in my software development that I never wanted to do anyway.

But um, if it can't even follow the patterns in the code base or can't follow simple instructions repeatedly, no. Like if I think about it like it was a person. Okay, I'm I consider myself a very patient person for let me quantify that for uh for newer software developers. Okay. And that's because I've spent a significant portion of my career trying to help more junior software developers. And as an engineering manager, that's something that like I take pride in is trying to be able to help more junior developers because they need the most help, right? It's for a lot of people their first, you know, first job in their career, you know, might be fresh out of school or from a boot camp or maybe they've never had to work in a team before. It's it's not easy. It's not just code, right? So, I consider myself pretty patient because I need to be because people take time to ramp up and learn.

That's okay. With that said, like I don't have the patience like it's exhausted by patience. AI, I mean, exhausted it completely because it's infuriating to have the same mistakes made on repeat with confidence. And I cannot imagine a human being. I'm trying to like picture if one of these agents that I was working with over the weekend was a person, I would probably be coming into work being like, I think we need to exit this person. Like, there's something wrong. Because if we have a conversation together and I say, hey, like, no, this is the thing we use. And you're like, oh, totally makes sense. Yep. and then immediately go to just the complete opposite of everything we talked about. I would probably have that conversation with you a couple of times to be like, "What's what's going on? We just talked about this before I start to question like, is something actually wrong?" And uh I am way past that with some of these uh agent setups to be honest.

So, um I know like I know it only gets better and this is kind of as I get closer to CrossFit here thing I want to close out on is like I I need to keep using the tools. I realize that we're still very very early in all of this. So, I get it. But like I genuinely feel that my health is being impacted. And I realize that sounds like, you know, a stretch or an exaggeration, but I I cannot express maybe clearly enough that for me writing code like I like to program outside of work. It's an enjoyable activity for me and uh it's because and it's not even because of the code itself like I like creating things and I I feel unrested and I feel unwell from this weekend because I have been so frustrated for so much time over the weekend

and that really bothers me cuz now like it's Monday I'm starting my work week and instead of feeling rested from things I usually like to enjoy, I do not feel rested, right? I I want to feel like I have a full tank of gas to start the work week because inevitably there's going to be frustrating things with work. This is just how life goes. But now my hobby time, the part that I'm supposed to enjoy is like physically uncomfortable. So I need to remind myself I would like to continue pushing through with the AI tools because I do think that they only get better. I I think that it would be unwise for me to give up on them now because of the bad experiences I'm having, but I think I need to continue to tune those experiences, learn from them, and then as the tools are evolving, kind of be right there.

So, I think that's mostly it. I'll wrap this one up by saying that I'm going to do a uh I plan to do it this weekend, but honestly like I couldn't bring myself to record any YouTube videos. Um I'm going to record on my main channel a video with um GitHub Spark and this is supposed to be I guess the equivalent to like a we'll use agents to build it end to end for you. Um, so I'm going to make a video. We're going to try it out. But what I wanted to do that was special about this video is not actually try it out ahead of time. Uh, so I have not used Spark at all yet. And I held off this weekend because I wasn't ready to record, but I genuinely want to try sitting down. It's going to be hell for my editor, I'm sure.

Um, but I'm going to sit down and record a video where I use Spark without any prior knowledge or use. Um, and see how it fares. And, uh, GitHub Copilot's been great, but I'm not hopeful that it's going to build something end to end that works, but we'll see. So, stay tuned for that. That'll be on the main channel, and I will see you next time. Take care.