Everything is AI Slop! Or is it?

Welcome back to another episode of the Not Brothers Podcast, where today we're talking about AI slop. Everybody's got it. And a lot of people accuse everything of being it. so the question is, what do you do with it? And how do you identify it? What impacts does it have? and how do you avoid it, maybe? we'll see, we'll see how much we can get there. But, you know, we were talking kind of I guess pre-recording about a few different areas where this Kind of rears its head. Where do want to start? I guess

Mm.

maybe a definition of what AI slop is.

Fair place.

I think everybody's definition is different. Some people would class AI slop as anything produced by AI. There's a whole group of people who think that nothing should be produced by AI. There are a number of projects in the Linux community, especially right now, that are adopting the stance that nothing can be LI LLM assisted, inclusive of documentation and other things. I have my own opinions about that. I think it's a terrible idea. but it's their choice. They can run their projects however the hell they want. So that's the extreme where I would class things that I would say are AI slop are when they're they lack refinement, right? It's not you could look at something and say, like, this is not understandable, it's overly verbose, it repeats itself, it it says a lot of it has a lot of words to say a lot of nothing, is usually what it comes across in like documents and things like that, or or potentially, you know, in some sort of code that is just like extremely verbose. has a whole lot of code to really accomplish no real objective here. That's what I would class as, you know, kind of AI slop and where I would expect the pilot to identify somehow, some way, and course correct for those things.

I I'll put a a human element on top of what you just said, because I agree with everything you said. The other piece of it is maybe piggybacking on what you called the pilot, and you've said this phrase many, many times. If you produced it via AI, it's still your work. So if it's not something that you'll be willing to put your name on and be proud of it, it's probably slop. In which case you should refine it until it's not slop. And and you feel good about it and that you would on the if you were on the receiving end, you would feel like it accomplished the objectives of of what the ask was. and you know, you've also said this many times, and we'll get into this a little bit more as as we get into some examples of of some danger areas with with with what AI can do. But AI does give you superpowers in a lot of ways. It'll it allows you to do things that you couldn't do on your own before and often took a team of you know, sometimes four to six people to do what an what an LLM can do for you if you use it the right way and if you put the right guardrails on it. but that's that's the old Spider-Man conundrum of with great power comes great responsibility. And I think when we're talking about AI slop, that's the piece that gets missed where people believe that they have far more capabilities than they actually possess. And so when they get into areas that, and I've been there, They get into areas that they're like, I actually don't know what this thing is doing. It's telling me it's the right thing. Maybe not.

Like the time we tried to set up a Python server and kept rebooting our OpenClaw server and I was like, What the fuck is going on here? And it was like, Mark's trying to set up a Python server. And I was like, Mark's fucking what? Okay.

It told Sheldon told me that was the right thing to do.

Like, hey, what are you doing? You're like, Well I need it to not I need to save my stuff. Like use a fucking JSON file, SQLite database. Something other than rolling and exposing an entire Python server. But like that's a good example of, you know, you know, where you start to delve into that AI slop territory.

Well, and and if you don't have some of those parameters already built into skills and whatnot within your AI models, they will go completely rogue and go do things that don't fit the parameters of what your organization would expect. That's a perfect for instance, right? So in early day in early days of OpenClaw with for us, we didn't really have those parameters in place. So when we get went to an LLM we're like, Hey, I need you to produce this thing and you know, it needs to have some kind of database to it to save stuff. And then the the You know, in this particular case, the way that Sheldon decided to do it was like, well, I think a Python server would probably work. And you're like, that's not best practices for what we do at all. So how about we set up skills that if you do need to do that, we're gonna do SQLite and we're gonna do JSON and we're gonna do it in a way that is lightweight as opposed to this other way. so it was a good learning opportunity, but I think still most organizations have not had the 10 months of evolution that we've had with putting guardrails on on what is otherwise a wild wild west.

Well I think there's it's a complex subject too, right? Like There's a there's a couple quotes floating around from Toby at Shopify where he's quoted in one a while back saying that at Shopify, anybody asking to hire a new person needs to illustrate why AI can't do that job, which was controversial at the time. I think still is controversial amongst some people. But I think a good test to say like Do we really need a headcount to fill this, or are we just using that as a crutch because that's how we've always done it? And at the same time, there's another quote from him more recently that talks about how AI slop grenades, as he's called it, are creating a real bottleneck and challenge within the organization. And, you know, your kind of anti-AI folks take that to say, like, you know, it's proof that AI, you know, can't do the job. And And isn't as helpful as you think. And it's really like to me, that's not a very complex, like those two can coexist and do coexist. Because the whole idea is like use AI to provide yourself superpowers and take use it responsibly and use it well. It's really not that fucking hard.

Mm-hmm.

the challenge is when people get lazy, in my opinion, where you you get a little complacent with it and you start just throwing it at things and saying like, You know, go do all of this stuff and then you don't have a proper review cycles or anything. And review doesn't necessarily mean to me anymore, reviewing every line of code. You know, depending on what it is, depending on the what the risks are. You know, we have some automated review processes that exist within every project that we touch. so in some cases it may be like, Hey, fix this issue because it's a defined issue, dispatch an adversarial agent to critique and refine that issue. Once you're happy, push it. Wait for the review agents and review processes that we have to sort of bless and approve it. Once it meets those criteria, you can deploy it. And I've done, you know, over time, you know, you kind of hand pick just like you would in like a factory, right? You just you're not testing every unit, but you go test every now and then and see, you know, did anything sneak through that I would have changed. And I haven't found one yet that I would have done differently, kind of following that path, because we have a really well defined pathway that it's going down. It's a defined piece of work, it's a defined outcome that you're looking for. Where it becomes difficult is when you don't have a defined outcome. And you're like, make thing work. And you just kind of throw that, blindly trust it, and then blindly throw it through on the other side. And I think that's some of the stuff that comes through and people accuse of being AI slop. Probably rightfully so. I also think that like personally AI slop is a term that I try not to use. I have used it. but I try not to use it because I think that it's an easy cop out. I've seen people, you know, accuse everything they don't like as AI slop, basically. That's that's what it becomes. Anything I dislike or I don't think was right becomes AI slop. And I think one of the most hurtful things you can do, and also just a way to lose credibility, is accuse somebody of something that they actually spent their brain power and time on of being AI slop. And it happens. It will happen. It it does happen. so I think there's better ways to give those critiques and and and things. It's actually funny when you know, David at Rails World went up on stage and and told everybody about AI and that they should be using AI more and that AI agents are awesome. And you know, kind of explained how he's using them today. And, you know, there was a lot of there were big reactions to that from a lot of people. I I guess is what I'll say. some good, some bad, some kind of nervousness in the middle. and Jake and I were, you know, we were kind of batting it around while we were at the at the place and talking to different people. And at one point, you know, We sort of realize that there's this constant discourse around AI. And it's like, well, AI generates buggy code or AI generates, you know, documents with errors in them, or AI does, you know, whatever. And that's what we equate to slop. And we're like Yeah, do you remember the b before times when we were all handcrafting bug free code? We were handcrafting perfectly tuned documents that had no issues. It didn't fucking happen.

I had someone that recently came to me and said, Well, if we're just gonna skip the process, meaning because we used to have workflows that had, you know, whatever, eight steps. And now because we have other tools and parameters and checks in place, like you were talking about, you'd and never instead of being eight steps, it's it's six or it's five or whatever. If we're gonna skip steps, like what what what what's the what's the point of this? Like why are we skipping all these steps? It's like, well, we didn't just choose to skip steps. We Just because we've had AI, we've always skipped steps for various reasons and always attempted to justify it because of one reason or another. So it's not a unique AI specific thing that we're just does like we have we feel like we can just jump ahead and skip good process. That's always been a thing. And it's probably even more important not to skip good process in the age of AI because of all the things that we've talked about, right? So Part of s part of being a subject matter expert in our field of digital marketing and advertising, or in in technology, meaning websites and applications and whatnot, is setting up what is good workflows and process, what is good adversarial conversation. Right. So in in then the before times it was a human. Yeah.

And I think adapting to it. Like the you know, one of the things that I think that is probably where you're going is we're seeing a ton of adaptation necessary. Our processes that have existed to date as an agency are highly tuned around people doing a lot of stuff. and I've even had conversations internally with our team, it makes them super uncomfortable in some ways where, you know, someone's like, where's the sprint plan for this? Like, we're not doing sprint planning. Like, okay, are we doing, you know, we we used the shape up methodology for a while. And you're like, so, like, what's that look like? Well, not fucking doing that anymore either. Because they just don't fit those models that existed don't work and don't fit for how the work flows anymore. And the question is like, how well how are we organizing or executing that work? And it's super big, like, I don't know vibes right now because we really don't know. Every project is a little bit unique. And we're kind of experimenting with each one to figure out, you know, how do you in an agentic world, how do you measure results, measure success, plan the order of operations for work? Because you could theoretically just plan your whole project and just throw the whole goddamn thing at the agent and be like, go do and keep churning until you're finished. And maybe it'll work, maybe it won't. Probably won't. You'll probably have a bunch of stuff to clean up. So You know, now you need to slice it into some sort of units, but we don't know what those units look like and how you plan those units. You know, the funniest thing today is like, you know, you use any agent, they'll still all still do the same thing, all the LLMs. If you map out a set of work and say we need to do this work, it'll it'll be like, Okay, this is you know, sprint one, sprint two, sprint three, this is nine weeks worth of work. And you're like, Listen, buddy. This is your next two fucking hours. Like, I don't know how to explain this to you. This is not nine weeks. This is you're gonna go do this in the in about two hours, you're gonna be like, Okay, we're done. so that compression of time is really confusing right now because we still need time to make the products good, to QA to test them, to work out any bugs and kinks that are introduced, to handle user experience flaws, all of which are still things that we we always needed. Like that's not new for AI. in some cases it's reduced because of AI. but

But some of that, some of that adver I'm gonna keep using the word adversarial feedback, right? It's basically the word, it's QA, but in in in an agentic world you call it adversarial q feedback, right? but those workflows and and those ways of doing things have always existed in professional services like ours. You've built systems and tools and feedback, you know, and and and frameworks and documentation and all these different things to make sure that. some of those things that are the most important part of projects like this, like a website or whatever, to make sure that those things are being followed. Is this satisfying the business requirements? Is this satisfying the user experience that we intended to to set out to deliver? Is this

Previously that would be like peer reviews, that would be where things like pair programming comes in, that would be, you know, where you submit a pull request and then, you know, your boss reviews that pull request or whoever is responsible for that project reviews it and gives feedback. Right? That's how we used to do all of that. Now we're able to sort of spawn agents and some of them are just completely fucking stupid, right? Like I I've had plenty of them. I've only found one product so far that does automatic code review that I don't just want to absolutely yeet into the sun. but they which is macroscope, by the way. We've used other ones, we've used CodeRabbit, we've used all the other ones. Some of them are just like so tuned, like I'm pretty sure that their prompting is like find an issue. Because it

Ha ha ha.

You just constantly are going back and forth with them. You're like, you know, it finds issues and you fix the issues, and it's like, okay, now here's these other issues. And the issues it'll find will be like, all right, if it's the second Tuesday on a solar eclipse and somebody did this while you're deploying the application, a record could be dropped that would get restored automatically in the next tick. And you're like, what the actual fuck?

W one of the one of the things you're you're picking on here is one of the things I think for our industry, we're going continue to have challenges and face, right? So because AI give give a perception of having superpowers, especially early on in projects, where you can you can get to a visual deliverable of something that looks pretty decent incredibly fast in terms of whether it's whether you're designing something from brant from scratch or you're creating a a a full-scale website. But there's a massive difference between going through and understanding user experience principles, going through and understanding things like code review, going through and understanding things like speed, and and going through and understanding things like server setup and server administration, all these different pieces and parts that aren't, they aren't really set up in a in a way, and I don't think they will be for quite some time, where an agent is going to be Good enough to do all of those things for an average user because there are they they're gonna select things that is fast and easy for the agent to produce and perform those activities, which may not actually be best for the intent of whatever the outcome is. And unless you know we should we should do this in in this framework, and we should do this in this framework, and we should, we should host it in this type of environment because of these these reasons, those pieces don't come together in the end of that project. And so the the upfront part of a project looks great. And then the wheels fall off the bus as you get through like into the project 'cause you're like, shoot, we didn't think about this or this or this. And and now you're you're left with something that visually might look okay, but it's a bowl of spaghetti.

I think that I mean there's a principle that exists that I can't remember exactly what it's called, but it's effectively that you know, users equate the project the progress of a project with the progress of the UI. And that's always been the case and it still is. So when people see that like you can progress the UI and what I visually see in my very narrow scope is really good, You know, they think they're moving really quickly. What you may not realize is the underlying is really, really bad, or that the mobile experience is also poor. And now you're you're ostracizing all of your mobile users, or your site speed score is terrible, or that it doesn't work without JavaScript, or that it's you know, all these sort of things. Beyond that, there's another aspect that like I I saw a post a couple of days ago about it was like, you know. your how your LLM can get you it can cost you a hundred thousand dollars before your your SaaS product launches. And it was effectively a lot of things that we talk with our clients about of like, do you need GDPR compliance? Do you need CCPA compliance? Do you need COPA compliance? Do you need WCAG compliance, like all of these sorts of of of compliance and things that nobody really thinks about or knows about, that you absolutely could. Like if you just build a website, import all your email lists, like and then just start blasting some stuff, like you've violated CAN-SPAM, you probably violated GDPR if you operate in the EU, you've violated CCPA, if you don't have age verification, you violate COPA. Like there's so much That you could fuck up just right there without ever knowing. And the agent doesn't care because agents aren't tuned to deal with that. And that's where, you know, from a professional services standpoint, where, you know, someone like us, we're like, well, we know those things because we have to know those things. Like I don't know all of those acronyms because I enjoy knowing acronyms. I know all those acronyms because they're required for us to, you know, do business and make sure that we're protecting the the the people that we partner with.

Mm-hmm. Yeah, it's it's kind of it's a bit of strange times, right? Because you've clients in our world have always had access to similar tools, right? They've had access to things like Photoshop and Figma and pick your favorite code engine and whatever. The difference now, I think, is it feels like you need, in some cases, you you could need less of those professional services aids. The

Uh-huh.

the problem With that is that the time and energy still has to come from somewhere. So if that comes from you and you're you're technically proficient and you're really you're you're sort of a a jack of all trades, you know what good design looks like, you know what, what to look for in in a code review cycle. You can write the equivalent of skills to help spin up, you know, adversarial agents and whatnot to be able to get you through these things. Well, you might be able to do a lot of the stuff on your own. Like that's completely true. The vast majority of people, and I'll I'll give it a number, I'm gonna say 95% plus of people do not have that level of skill set to do those things. Not even anywhere close to end to end, let alone one portion of the process. And so I I feel like this is a little bit of it's a little bit of danger mode that I have, and it's a it's a yellow flag for for clients that are thinking of in-housing a lot of these sorts of things. Because while it can be done, doesn't mean it should be done. Because in the end, You may actually end up to your point, Ryan, either with legal compliance issues or with a project that doesn't progress because there's no accountability for the progression of the project. it's easy to hold a an agency like us accountable, right? 'Cause like we we don't get paid if we don't have the deliverables. Your FTE internally can have as many failed projects as he wants and still get paid.

I think we see this we see this like in housing and outsourcing of of agent or a agent of agency work consistently. So I think this'll you know, it'll probably create yet another swing where that happens. I think that

Yeah. The pendulum flings.

You know, you can look at it from our perspective and say, like, maybe it's a maybe it's a risk to the agency world, or maybe it's a boon to the agency world. I'm not sure. and I think if used correctly, it can actually be a really great collaboration tool, but it has to be looked at that way, right? If you have a client who is non technical and they have an idea that they're trying to get out of their head and express to the team or that they want to explore, by all means. I'll give you the fucking tools and let you just kind of play in the play-doh and help us understand what you know, what's this idea that you've got in your head of how this works. What you produce might look great, might be complete dog shit underneath, and we'll fix that part. But it at least helps convey the if there's just one thing AI does really well right now is it helps business and non technical people convey ideas in a richer way than they've ever been able to before. because before it was all just words and maybe a sketch. But now, you know, I can just give you Claude code and say, go bang around for a while. Go tell Claude what you want to produce for a while. It doesn't need to actually work. Just tell him what you want it to look like, tell them what to change until you're happy with it. And that can sometimes allow you to work through the problem as well and find that, you know, the initial idea that you had was actually like kind of a bad idea. It doesn't really work. So I need to maybe rethink that a little bit and wind up somewhere that's actually better and then you know, now you send that over to the team and now instead of trying to read your fucking mind or trying to explain to you why this is a terrible idea, we've already just moved past all of that and we're on to

Yeah, it's it's the old adage of, you know, in in our world, it was, I'll know it when I s when I see it from a client, right? So you're going through

Yeah, yeah.

design iterations and you're going through different comp explorations and big idea development, all this different stuff. And a client's like, I know I'll know it when I see it. It's like, well, now actually, before you even give us a brief, or as part of the brief, you can actually go and do some exploration and say, like, hey, I don't want to own all of this, but here's what here's kind of what I'm thinking. Like, this is, you know, a 30% baked idea of, you know, help me make it better. And that's actually to your point around is it boon or bust for agencies? I think most clients want that type of partnership with their agency. They want to be able to say, like, all right, I've got an idea. I don't want to own the idea, but I want you to help me progress

Wanna contribute?

it. Yeah. And but and I need I want you to respect my feedback along the way. And I think, I mean, we've even fallen victim to this, right? Because we are experts in our field. i you can have a tendency sometimes to discount and discredit what a client might provide you as that feedback and try to exclude them from the feedback loop because you quote unquote know better. And while that might be true for the end outcome and the end product, to to your point, in today's world, we've never had an opportunity for for clients or business people to express their ideas in a way that can be then taken and grown and extrapolated on and put layers of best practices on top of. And so it's I think it's this this change of how we work with clients going forward as opposed to fully outsourced idea from end to end of of a project. Maybe we enter an era where clients are much more involved in in some of that that production world. I don't know.

Well and I think that, you know, another thing that we've found is it can be it can open up opportunities for us to explore multiple options where previously that'd be cost preventative, right? If it's like, hey, let's explore like three different header options that are really intricate and have, you know, mouse movement things tied, like we can do some concepts of those, and still keep, you know, within a a define a well defined budget. Whereas before it was like you kind of gotta know what you need. And so that leads to kind of taking the safe road on everything. And you know, not to say that what comes out is bad, but certainly not as good as it was. So, you know, kind of circling back to the topic of of generating AI slop, I think in some cases, like that would be something that I would class as like potential AI slop, right? If I just kind of plop together three ideas of like, hey, I I want to create like a space themed header that has some movement tied to my mouse when you know mousing over it. And I need these titles in here. Like create three different variations of that that are distinctly different. Right. Like I'm just kind of slopping some shit together. But what I am doing is I'm getting an idea. I get three f different ideas and can kind of go, this is actually what I'm I had an idea in my head, this is actually the closest to what it feels like. So let's continue down this path and refine this path. Versus before, you kind of just have like you would either not do that at all, which is the most likely scenario, or you would just have to drill down one path and not be able to explore and see like, is there how does this feel? How does this look? Does this actually f look as good as I think it does in my head? 'cause we've had plenty of things that we've spent And we've we've paid a team for a week to work on and we're like, this is good work, but not for us. And

Yep.

you shelf it. And then like that's painful, right? That's when you take multiple FTEs worth of work and just shelf the work, that's painful. And it's a little easier to do that with, you know, AI generated items.

Yeah. You're you're a hundred percent right. I think what from a client perspective, what clients are wanting, they're they're not wanting things to get cheaper from all the survey data that I've seen in our world. What they want is more value because they're they recognize and understand that AI is gonna be part of the solution set. And so to your point, if you can add some different value along the way, so instead of producing, you know, one concept of of a header in your example, it's two or it's three, and it wasn't promised. It was just exploration that we went down. We believe we should go this direction because, but we want to show you that we actually thought about other things too along the way. And I think that's a good way to to to gain trust and and make sure that clients really feel like they're part of the process. The part that that gets tricky about it, and I going back to things that we don't quite know yet, is that that's a whole that's a completely different process than we've ever been able to do before. It doesn't fit nice and neatly into our eight step website solution project anymore.

Well, taking a step further, there's also the question of what is value, right? As we've described and talked about here, there's there's a lot of stuff that we've invested o in over the years, you know, even previously, like pre AI. But, you know, we have things that we've invested in like website frameworks and for CMSs and s now with agents, skills and adversarial workflows and review tooling and all of these things that make What we could do much faster and more efficient and have a higher degree of quality with what comes out the other end than what somebody potentially with the exact same tools could do, because they don't have they haven't spent the you know hundreds of thousands of dollars worth of tokens and hundreds of thousands of dollars of hours worth of hours refining and and experimenting with these tools to figure out how to How to pilot the ship. So eventually we create kind of a widget producing factory and that can provide a tremendous amount of value without a tremendous amount of effort, like with a disproportionate amount of effort on our side. And then the question becomes like at what point, at what points does that value pass that value or cost pass through to the client? And at what point is it just margin for the agency?

Yeah. And I I think most clients really understand that. I think the the pieces that in the surveys that I've seen are still true in an AI world as they were before. I clients primarily leave agencies or really any service provider partner because they stop bringing new ideas. that's the primary reason. The second reason is that quality gets gets diminished, whether it's a new team that got added to a a a different set of projects or there's turnover in the organization or whatever. Yeah, people get lazy.

usually just people atrophy and get lazy.

it's it's it's any any of those things. Those things haven't changed at all. what maybe has changed is the dance that you may have with a client. And to your point, I don't know that we have that fully figured out yet, because as these tools and AI agents are tools, let's make no mistake about it. That's what they are. As they become more and more ubiquitous, and as the agents themselves become easier and easier to prompt, I mean, think about this. Just what, nine short months ago, prompts had to be almost painfully slow. It's do this and then tell me what you did, and then do this and then tell me what you did, and then do this and then tell me what you did. Now you can give a very simple prompt, and the agents are smart enough and have learned amongst tens of millions of prompts, hundreds of millions. of prompts, billions of prompts over that time period that when someone says something like, please do this, they probably mean this and they just go freaking do it. so as you as you enter an area, an era that looks more and more like that, how does a a service provider client relationship evolve and change?

But I think it still it still comes back for me at least. It comes back to those the the old three levers, right? The good, fast and cheap.

Yeah.

I still don't think there's a world where you have all three. I think you could, but you're just screwing yourself if you do, right? The if we figured out how to provide a really good product really, really quickly, which is where kind of I think we fall or intend to fall on that in that trifecta, is I wanna I wanna deliver incredibly high quality work at a speed that nobody else can. And and do that without sacrificing the human component. I wanna do that through automation. Leveraging AI, building proper processes and tooling. And that's why we spend a tremendous amount of money investing in our own tooling and experimentation and you know, testing out new processes, all those sorts of things, which should allow us to reap the benefit of the value equation, right? If I can do something for if I can get you in market faster than anybody else can with a higher quality product. Well, it's worth more money.

Yep.

And it doesn't matter that it costs me less money per widget. Because just like, you know, the pharmaceutical industry as an as a for instance, every pill might cost me thirty cents to make. The problem is the first pill to cost me thirty million.

Mm-hmm.

So there's a certain amount of like, you know, eventually it goes too far the other way, right? Like we could get on a whole political rant about, you know,

Mm.

pharmaceutical companies and shit like insulin that should be available, but you know, we'll leave that for another day. But You know, there is that challenge of like we've we've spent millions of dollars to figure out how to do this do the first one. But now every every incremental one is cheaper.

Mm-hmm.

but inherently worth more money specifically because of all of the stuff that goes into it.

Yeah. The the interesting part about the world we live in, this continuing to talk about, you know, being a service business, like a digital marketing advertising agency that Oodle is, it's the human component is actually much more important now than it ever has been. And I don't mean that in terms of producing the deliverables. I mean around making sure that the client And the team is well understood in terms of what are we trying to accomplish? What was the process that we went through and why? What is our thinking around this? Why did we deploy these agents to go do these things? What what were what were the the learnings that we had along the journey? What why were those meaningful? And then ultimately, where did we end up? And why do we feel like that's the best approach for this particular scenario? And you know, we've gone through this iteration going back to the this idea of producing websites, where like,

okay, how did it say?

Dev when when you got to so a client involvement used to be super heavy through the design phase, right? And then development went into a black hole for some period of time and then came back and it was like, here it is. Ta-da. And I've if you use that same example in the in today's world, I think the client wants equal involvement throughout the entirety of the process because you can. And it looks very different than it did before. And so when I say involvement, I don't mean You know, they don't necessarily want to be pulling the levers of the of the widgets to be able to produce it, but they want that feedback loop to be faster and more continuous than ever before. And so that's something that I think agencies like us are still adapting to. It's it's a diff it's just different.

Yeah, I think kinda we've meandered off of course a little bit, but kind of pulling us back to to center on the AI slop topic, I think that You know, a couple other things that I would probably layer in there. is I don't th I I think that I mean obviously there's there's plenty of poor work that exists. But I think going back to that same idea that Jake and I explored, you know, kind of together when we're at Rails World That isn't exclusive to AI. It's just that we can do more of it faster now. Right? I think I think we've said this before, even probably here, that AI is a force multiplier for what what for whatever you already have a lot of. So if you are producing a lot of garbage, it's a force multiplier for producing garbage. If you're producing great work, it can be a force multiplier for producing great work. And I think that on the other end of the spectrum. You know, there are folks who have taken the stance that like everything AI generated is just garbage. It's all bad by default. And I think, you know, I kind of choose to take the opposite stance where it's like I'm going to assume that everything is is good by default. And then explore it and if I find that it it is bad in some way, not necessarily gonna chalk it up to being just AI slop immediately. Maybe the person who was piloting it wrote it by hand. Maybe they didn't maybe maybe the person piloting it truly didn't understand the ask and you know, meandered off course. I don't know. There are a lot of variables at play. I do know that just dismissing someone's work just outright, is really bad. And demoralizing and probably something that should be avoided, especially if you want to create a collaborative team. it's a challenge that I have with you know, working on projects, especially public projects where like Omarchy, where other people are going to push things in, there becomes this challenge. And Mitchell Hashimoto talked about it at one point when he was working on creating vouch, where he was like, you know, if If you've pushed a pull request that you worked on, like you actually put effort into I'll spend time with you going back and forth and help you get that pull request merged and help you understand the things that were wrong if it's wrong, right? Like if it's just hot fucking garbage. I'll spend however long it takes to help you get that merged. The problem is things like Cursor have created one button that you can click that'll open a pull request on a project. And those pull requests produce this like massive diary of like here's an issue and whatever. And they'll be completely off base and it'll create some patch that makes no fucking sense. And as a maintainer, you're not really sure who you're arguing with here. Like, if if this is just something where you clicked one button and you actually don't know anything about it, and you're not gonna be offended if I say, Hey, actually, this is completely off fucking course, I'm just gonna close this, I'll just do that. But the the challenge is I don't know when you've done that and I don't know when you've actually put effort into it and just missed the mark. And like that becomes the trickiest piece. That's where people have talked about even disclosure of like, you know, some projects require you to disclose, like, hey, I've used AI for these portions. but also that, you know, I've also manually reviewed it and done certain things. Like everybody's trying to figure the same problem out. And I don't think there's a good solution. I think that I don't think there's a one size fits all solution. But I think the good solution is, you know, to kind of follow best practices that you would from a a normal collaborative standpoint, both when producing something, to ensure that you're not just dropping, you know, an AI slop grenade in somebody's lap, but also from the receiving end to make sure that you're potentially not accusing someone of of just you know being lazy and not producing good work when Perhaps they poured their heart heart and soul into that and they just missed. you can lose a lot of respect really quickly that way.

Yep. So the TLDR is yes, AI slop exists. No, not everything is AI slop. Be kind to people and give them the benefit of the doubt is still a real thing, even in the age of AI. So hmm. Strange.

Weird how that works, isn't it? There's still people piloting the goddamn ship. So maybe be aware of that. It's actually really funny, and I I'll maybe we'll close with just like a small little story from it. But you know, Omarchy has been no stranger to the AI conversation, especially now, as we've kind of taken the stance of it being a malleable OS and being very agent forward. and that's led You know, anybody who doesn't like Omarchy to just accuse it of being AI slop and they'll point to certain things in the code. There's things that they pointed to the code that I fucking wrote my hand. And I'm like, that's not AI slop. That's just they didn't know any better at the time. Like, you know, just like ever you know, all this artisanally written code that we've had for years that's been bug free. it turns out that's that's all a fallacy. Like there have been bugs in every bit of code forever. You learn over time and you go, shit, I didn't know that this was a problem. So we we make those changes. Ironically, some of the things that have been critiqued the least are the pieces that have been AI assisted. And probably because, you know, those AI assisted pieces went through adversarial reviews and you know, lots of honing and refining to make sure that they, you know, they didn't miss anything. But you know, the The one that cracks me up the most is it's a security vulnerability that got called out by some folks with regard to the original Docker state decision, which is that we added the user to the sudoers group, which technically does open up a an escalation hole, but is literally how Docker tells you to set it up, and also how every fucking developer that I've ever worked with has their computer configured. we just happen to do it out of the box. So it gave people something to complain about. But if you round up a hundred devs that use Docker on their machine, I guarantee if you run docker ps, every one of those motherfuckers lists out a bunch of box boxes and every one of them is running this sudo. Maybe a couple of them are using passwordless passwordless Docker, because that is officially supported now. But it's just comical. That is a decision that AI did not make. That's a decision that conventionally We made because that is that's how people have operated. That's how developers have operated for a long time. Now we're bringing this operating system to less technical folks now. So we have to have kind of a heightened security stance from where it was before when you're just operating with developers. But again, just one of those things that people, you know, are quick to jump to that and be like, Well, AI just wrote it and you know, did stuff like, No, it didn't. I wrote it by hand.

I think that's a good a good story to end on.

Till next time.

Till next time.