Rails Business
Brendan Buckingham and Ryan Frisch talk about developing with Ruby on Rails and how to leverage it to build a business.
Rails Business
Leveraging AI to Enhance User Productivity
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
In this episode, we explore how AI can be leveraged to build engaging features in our applications. We discuss the development of a 'guided blogging' feature using AI, its benefits for users, and the challenges faced in implementing conversational interfaces. We also discuss a bit about the future of AI in SaaS, the role of AI in content curation, and how to balance automation with human oversight.
00:00 Introduction and Catching Up
00:41 Discussing AI feature Ryan is building
02:35 Technical Challenges and Solutions
05:49 Exploring AI Integration
14:41 Future of AI in SaaS
25:13 Human Element in AI
31:42 Conclusion and Final Thoughts
LINKS
Questions or comments, email us at railsbusinesspod@gmail.com
Hey, how's it going? It's been a
Ryan:I know how you been,
Brendan:Not too bad. You,
Ryan:you know, hanging in there.
Brendan:you know, it's, it's funny, it's like just started for my kids. Probably your kids too. But finally over and, I don't know, I feel like different about the podcast, like in a good way. Like, oh, I, maybe I have more time to do this now.
Ryan:Well, we, we can all, uh, dream of, of that and.
Brendan:Yeah. Yeah, exactly. So feeling refreshed, I'll put it that way. A little, little bit of, not, not really a break for us, but, uh, little, little bit of a break, I guess in some
Ryan:It's been what, three weeks?
Brendan:yeah, it's, it's been a little bit longer for us. Yeah, yeah, exactly. So,
Ryan:Alright, well let's do it.
Brendan:but uh, definitely. Uh, I feel like AI is the topic and I feel like you've got some thoughts on it that maybe we can start with and just kinda go from there. Sound
Ryan:Yeah. Well, so thoughts is maybe generous, but let's see where we end up. Um, I started building. Trying to build something. For our app, we have this feature that is, that we call guided blogging. Right. And you know, as it sounds like, the idea is we want to prompt users to answer questions and then, you know, turn those answers into a blog. Now before ai, you know, we have a interface that is. Sort of like a wizard, I guess you would call it, uh, with some JavaScript, uh, on top of it to just sort of move to the next step. Um, but it does. We prompt the user with a question, and it does require them to actually be a, a, a somewhat coherent writer, uh, in order to form a response. Um, you know, so an example of this would be like, um, you know, if you're a maker, you, you would have like a highlight, a creation blog post, and it'd be like, you know. Give it a title. Use keywords, you know, what did you make would be the first prompt. And then it's like, you know, what challenges did you run into would be the second prompt. Why are you proud of this? It's a horrible example, but that could be the third prompt. Um, you know, and we have, you know, probably 20 different prompt templates for different things. You know, highlight a creation, highlight a, you know, a volunteer in the community, you know, things of that nature. All right. So it works. But AI would make this. Way better. Right? Because now they don't have to be the best writer, right? They can get something that's almost ready to hit publish on by answering the prompts. The challenge I'm running into not really a challenge, I think it's more of a lack of knowledge of how to do this at this point, and I'm still sort of exploring. So I'd, I'd like your input is I want it to be conversational, so to me it feels like it should be like the ChatGPT style.
Brendan:Yep.
Ryan:You know, back and forth, but I need it to have guidance and I need the AI to actually move them along and get answers to the questions. The question is, how do I do that? And then at the, at the end of the day when they, at some point they're gonna be like, all right, I, I'm done. Right? Then I have to get AI to give it back to me in the right format so that I can then put it into the database in a way that actually makes sense. Um. So I went down the path of, you know, I, I sort of have the prompts defined and I'm using the context window to pass the prompts in with the instructions of, get answers to all of these questions and, you know, be a, you know, a marketer and use this framework and all this other stuff. But I feel like my context, window's getting large. So now I'm concerned about, you know, price, like how much is this gonna cost? So what, what are your thoughts on. This particular challenge, right? Or
Brendan:Yeah.
Ryan:even a challenge, but a particular project.
Brendan:Right. Yeah, I mean that,
Ryan:Mm-hmm.
Brendan:is an interesting one, I guess. So I wanna just ask a question about what you're currently doing, just to qualify something really quick. So you do with all the the, I mean, obviously you're going for a blog post, I get that part, but going through, currently today without ai, and you're asking these prompts, how are you storing them today? For instance, is it.
Ryan:Y Yeah, so I mean, it's a very simple model. It's, you know, you got a blog model and each blog has post sections, and each post section has a header, a description, and image. Um, and essentially each prompt is associated with the section, right? So,
Brendan:okay. So that it's a blog. Got it. Blog sections, and then the sections build on
Ryan:Exactly. I mean, it's just, you know, it's a blog with paragraphs. Each paragraph has a header. Each header can have an optional image. Um, you know, there's some other things that, you know, different sorts of layouts and stuff, but
Brendan:Mm-hmm.
Ryan:at a very high level, that's, that's it. There's nothing different to that. Right. But the
Brendan:Yep.
Ryan:UI doesn't, as it's currently exists, just is. Not gonna work for ai. Right. So I'm like, I'm re-imagining the entire interface of how
Brendan:Mm-hmm.
Ryan:this happens. And, and now I'm thinking, all right, well, you know, there, there's this chat interface now that, you know, prompts you to get all the answers. And then it takes all the answers and just, you know, the, the, the final instruction to, to a, to the LLM is, you know, put it in this format. Right. That I just explained that that part. I feel like, you know, with the, you know, just get, give it, say, put it in adjacent response with these attributes is probably sufficient at that point. But it's like, how do you get from point A to the point where you're actually ready to get the blog in final format? I.
Brendan:Well, one question I have too is do you need, like, like what's the advantage of having the AI help the entire time versus, for instance, doing what you do now, but then once you have all the, the sections, handing that off to the AI and having it essentially like massage it and make it look a little nicer than what you can do without ai.
Ryan:Um, good question. Uh, I think there's, so we have a, all our prompts. And all of those, you know, blog types that I just talked about are part of a framework that we've developed called Marketing 3, 4, 5. Um, and that is like the three reasons to do marketing, the four reasons people choose to work with you, and the five ways people find you. And my business partner's actually written a whole book on it, but the idea is. Like the, we have a very specific approach to how to promote businesses and local marketing,
Brendan:Mm-hmm.
Ryan:and we want to be able to suss that out of people who either aren't familiar with the framework at all, or just have trouble articulating, you know, responses. Right. I think that AI could be. A useful tool to actually get meaningful responses back. You could do what you just said and say, you know, ask 'em to fill out all the prompts and then, you know, make it look nice. But I feel like the part that helps them come up with the answers and can interject and. You know, other context. So like, we're a marketing platform, so you know, we have locations, we have, you know, product lines, we have business lines, we have all their website data. You have all their contact information, right? Like, if I can instruct the l lm, it's like in all responses, like, include this stuff naturally right now, they don't have to think about writing it. And I just think it's more powerful and it,
Brendan:Yeah, that's fair. Yeah, no, that's, that's good context too that that paints a better
Ryan:yeah.
Brendan:for me. And I mean, it, it's also making me think like, I know you haven't fully implemented this, obviously, we're just talking through it. And it sounds like, correct me if I'm wrong, like what you might be doing in your prototype or whatever you wanna call it, is basically just like maybe sending over an API called the ChatGPT or Claude or whatever, and then getting whatever you want back. But there's no, you, you don't have like, um, a rag or you don't have you're not training the model in a way that it knows your system necessarily yet. Is that
Ryan:Uh, yeah, I am not doing anything advanced. I am simply loading up the context window with the stuff that I think is relevant for now. Right.
Brendan:So you're pulling in all that location and information about their company passing it along, and then are you still kind of using the questions that you have somehow with it
Ryan:Yeah, so the context window will include all the factual data about the business. Um, we have a way that we are, uh, we collect the information from our users for the four reasons people choose to work with you, which are, you know, experience, expertise, you know, likability and your sort of values and community outreach. Um. So I pass those in. I say, Hey, use the marketing 3, 4, 5 framework, which is at marketing three four five.com. And then I give it a list of prompts right from our database and say, here are the prompts for this particular post type. And then I give it, you know, system instructions. It's like, you know, get the answers to each of these prompts, you know, uh, be conversational. You're a marketing expert. I'm, I'm paraphrasing here, right? Um, you know, the goal is to come up with a. 4, 2, 5 paragraph, uh, blog post, uh, you know, of this type. And, you know, then I give it some instructions of formatting at the end, right? But,
Brendan:Yep.
Ryan:now what happens is that's all sent in the first request. You know, so the LLM comes back and, you know, at, at least fairly consistently ask the, the first prompt,
Brendan:Hmm.
Ryan:and then I can respond. And then it's like, all right, now it's a black box.
Brendan:on that, it's probably like, okay, I'm gonna go over here
Ryan:Right. But the, I think the cool thing about that is the, the user can ask clarifying questions and take it wherever it wants. And the, the hope is that the LLM will always bring them back to a final product. Right.
Brendan:Yeah,
Ryan:Um, now I don't know how that's gonna play out. It's gonna require some, some testing and iteration, and that's the part, as a programmer that maybe is a little bit, uh. Hard to reason about'cause you lose control. Right.
Brendan:Mm-hmm. Yep.
Ryan:but yeah, the, so like every time there's a response, I'm sending the entire context window plus all previous responses. Right. So like each response just grows and grows and grows.
Brendan:Yep. I, I, I see exactly where this is going now.
Ryan:Right. And, you know, I did it, you know? Right. Right. Now it's, it's, it's actually in production. Behind a feature flag with like real keys and I can use it.
Brendan:Mm-hmm.
Ryan:So far it hasn't been a huge problem cost-wise, but I also like am not like going off the rails, right? I mean it is essentially, you can ask it whatever you want and
Brendan:Right.
Ryan:I mean, it could be infinitely long chat. So like, I don't know. Now I'm just like. How do I put guardrails on it? How do I limit costs? You know, obviously I can sort of put a cap on the account, but, um,
Brendan:Mm-hmm.
Ryan:you know, I just don't know. I, I'm, I'm brainstorming ways to approach this and.
Brendan:Yeah, I mean, I, I haven't done anything quite this intricate. I've done a little bit of this API stuff back and forth, and as far as I've gotten is basically like. Here's a post, you know, gimme some feedback based on what I'm looking for. And then I, I don't really have a chat like you have, I, I haven't built one
Ryan:Hmm.
Brendan:but where my head's going is just like the way that I program with ai. And I always think about, you know, I don't want the, the context window to get too large in those cases, because. You know, inevitably it's, it seems to me like you're three or four prompts in and all of a sudden it's like it's forgotten about something in the first
Ryan:Mm-hmm.
Brendan:And it just kinda makes me wonder if, going back to my, you know, kind of qualifying questions and asking kind of how it works today, if there would be an advantage to having you, you could make it look however you want to the user, but for the LLM, if you're basically having multiple different. Conversations and then at the end you're like, okay, here's each piece, summarize it or whatever, and then say, okay, now put it into the blog
Ryan:Yeah.
Brendan:That seems like it might at least help with the context window. I don't know if may, maybe in this instance that's not the best idea. I'm, I'm no expert on this, but that's kind where I was going the whole time you were describing
Ryan:Yeah. Well, I, I keep going back and forth between sort of what I just described and what you, you know, uh, shared at the beginning, which is just ask it for the finalized output ver and like. But then I think, like, how do I, you know, what does, what does the interface look like for collecting the prompts? Right.
Brendan:Yeah.
Ryan:Um, you know, I still gotta get answers to each of them, right? So I suppose I could have like a, I could stick with a wizard and, and then it just becomes. At the end of it, be like, all right, you know, prepare for, prepare for finished version, and, you know, pass that to the, to the LLM.
Brendan:Yeah, I, I guess where the reason I'm feeling like might want to have a wizard or something like that, or like the separate. Interactions with the AI is kind of the guardrails. That's, that's one way I see of doing it. that you're not allowing it to be just free chat.'cause you know, going to cost. Like what if one of your clients is using of these chats to answer one of these questions and they're going back and forth like 20 times. You know,
Ryan:Yeah. Or, or they could just realize that, hey, I can come in and just use this as ChatGPT and I don't, you know,
Brendan:Exactly. Right.
Ryan:it's not ever, it would be kind of annoying 'cause it, hopefully the LM would at least try to correct them back to course, but
Brendan:Yeah,
Ryan:reason.
Brendan:and I'm sure there's
Ryan:Yeah.
Brendan:that, but
Ryan:Yeah. It's interesting. I mean. So the, the blog piece is where I'm focusing first, just because I feel like there's a natural need for it there. Uh, but there's other like, activities that I'm also considering. It's like, all right, well, once a blog is completed, you know, generating, you know, social shares, you know, for various networks, um, you know, putting together a newsletter, you know, from the blog or several blogs or our calendar or whatever. Um. You know, coming up with alt text for images, like any of those things that,
Brendan:Yeah.
Ryan:um, people should do but don't want to because it requires a little bit of thought. And if I can make it a little bit easier, you know, that seems like a, a win for, for our users. Um.
Brendan:Yeah. And, and, and one of the things that I'm just thinking of now, 'cause one of the tools that I use with AI quite a bit is for this podcast we use a tool called tool called Descript. And it's not all ai, but some of it is ai. And like for instance. we're talking about is gonna be transcribed and then I can just go in there and I can say, you know, gimme a summary, or gimme a podcast description, or whatever it is. And it's just making me think like, that's not a wizard. It's, you know, it's not even in advance as what you're trying to build, but it. It has to have the, the some sort of context about what it wants to build, right? Or, or what it wants to give you the user back because it knows, obviously that there's the transcript and it knows that, hey, this is, they're gonna either do a podcast or you can do YouTube or whatever. But it's limited down and, and I guess maybe where I'm going to is, I've also been thinking a little bit while we've been talking that. Well for price for one, like think you could run your own model potentially. Like, this might not be the easiest thing, but one thought is could run your own model, like, uh, what is it, llama or whatever. That's like the, the open source one, and you might be able to train that on your own data that it's like not part of the context window for one anymore. But then it might, might also help you stay on. You know, on track with it. I'm not saying that's easy, it's just kinda another option I'm
Ryan:Yeah. Well, I mean, I, I think there's, there's a lot of pieces to this. Like right now I'm just sort of experimenting. Right. And, you know, I, I, I think the blog idea is cool. I just also think our users would, you know, the, it's some sugar, right? Like for the sales team, um, it's like, oh, you have ai, right? But like, the easiest place to start would be to do either just like polish the blog at the end or do the social sharing stuff, or
Brendan:Right.
Ryan:these other smaller. Pieces is probably what we'll end up rolling out first. But then, you know, I'm thinking long term right a little bit too. It's like, all right, what does, you know, what does the future of SaaS look like with ai? Right? And there's a lot of different opinions out there, but one that you know, sort of sticks with me is this idea that, you know, you don't necessarily. Need a front end anymore. Right. Um, you know, so I, I, I'm, I'm, I'm just thinking like, what would it look, what would our app look like? Like there's still value in connecting dots, you know, and integrating systems. That would probably require our own model at that point. You know, I, I don't really have any interest in figuring that out right now, just 'cause it's
Brendan:No.
Ryan:too time consuming. But, you know, there's, there's that, and then, you know, in that case it's like, I always come back to, you see stuff all over the place. They're like, people are automating systems with ai. I just don't, I don't see it. Right. Like, and then I question, I'm like, am I. Missing something or is it just hype and overblown? Right. So like, you know, obviously AI saves a ton of time, you know, with stuff like you just described with the podcast and descrip, like getting a description or a summary, like those things are great and that has saved a ton of time. But like as for like fully automated systems, I haven't seen that yet in anything that we've done it. Do you, have you seen it? Do you, what is your feeling on that?'cause.
Brendan:Yeah, I
Ryan:To be able to do, but I, I just, I don't get it.
Brendan:yeah, I'm kind of with you where I've, I'm seeing it as well, but I'm not seeing how to do it yet. Now, one thing that I've started looking into, I don't have much to say about it besides, I'm, I'm looking into n8n, I dunno if you've heard of that one. It, it's an open source tool, I believe, and I think they have a cloud version you can pay for, but it, it feels more like a developer. it's kind of, I think you can write some, maybe it's JavaScript. I forget what it is, but you can, you can write code or you can do like a drag and drop interface. what it is that I can tell from the little bit of use I've done with it is it's kind of like chaining things together. So one instance is, you know, if I wanted to, from a development manager perspective, if I wanted, this will be pretty simple and maybe it's kinda a dumb AI one, but like I could have it send me alerts like, okay, someone's opened a pull request. Maybe I have it run through an LLM or I, I have it, you know, make an API call to gimme a summary, just make it easier to digest it. And then maybe I have an interaction and then it kind of like chains together, but most of it's automated in a way that it's saving me time. That's, that's an idea. That's not probably the best example. And I think if you go on their website, they have just like a ton of examples. Um, but that's the kind of thing that I've been seeing a lot and, and I've even been in meetings with like. Our sales team and stuff talking about like how do we do, you know, uh, marketing automations and things like that, that would also relate to that. And you know, for me, one thing I'm playing with a little bit is how do I automate, not completely, but how do I like use AI to help with code reviews because I've just got so many of 'em. How do I get that summary? And it doesn't have to be this n8n thing, but like been doing it with Claude Code, I've been trying it with like Cursor, which obviously uses different, um. LLMs, you can pick whatever you want, but it does a pretty good job in that instance of just looking at the code, comparing it, and so far I haven't done it a whole lot, but I have seen where it'll actually give me back that I probably would've found anyways, or like maybe I already looked at it and I'm like, oh, I already saw that. And so it's doing a pretty good job. Also pointing out things that I maybe would've missed. And so that's a good starting point, but it's definitely not a, I don't trust it to do all the work, and then all of a sudden I don't have to do anything.
Ryan:Yeah. Well, and, and I think that's a good example of where I feel like the line is right now. Right? Like it, and. Like, and maybe this is just, we don't trust it yet. And that may be for good reason or bad reason. Right. It doesn't really matter. Um, but that is like, all right, it saved you time in doing the initial analysis. It gave you a summary. It pointed out some stuff, but it's not autonomous yet. Right. Like it didn't try to fix it and push a new pr. Right. And that would be, to me, that would be like where the real value. Could start to, to show itself because, you know, theoretically, at least in the code space, you could get to a point where if you have all the systems, like the CI systems, a full test suite, you know, all your stuff set up, that it could iterate itself, right? That's the the dream, right?
Brendan:From, from things that I've seen and heard, like you can do some of that stuff that you were just describing, but I don't know that I would completely trust that either.'cause you know, just in normal coding with it, it goes off the rails enough that I would kind of worry that you would get back something that's just not even on the right track
Ryan:Right.
Brendan:Um. there was a tool called Vibe Kanban I saw in like a Hacker News post recently and what it, like, their demo video was basically showing you start with the user ticket and then, you know, clot or whoever what, whatever LLM picks it up. It does the work, it opens the PR and that. I forget now though, if it did the code review, like all the steps you just said, but I, I've seen that in other places.
Ryan:I mean, cursor can do it too with, with Claude, right? Like in, uh,
Brendan:Yeah.
Ryan:mode.
Brendan:Right.
Ryan:for some reason it thinks my rails app's not configured properly, which I don't fully appreciate. So I haven't been able to fully test it out yet. It just, every time it tries to build, it says Rails isn't available. Um,
Brendan:Mm. Mm-hmm.
Ryan:which. Dunno, 'cause i's clearly running. Um, but yeah, I mean here here's another example that I feel like is almost autonomous, right? And this is again, back to our platform, right? But I wanna, we have a calendar and the hardest part about we maintaining a calendar is getting events for your calendar. And we do a lot of stuff to try to solve that problem within a community. But what I want to do is. I want to use Action Mailbox and have like an exposed email address where I can subscribe to all the newsletters that are in the community and have them automatically come into Action Mailbox. I want AI to pick that up and extract event information and like, you know, tee it up for, you know, either, you know, as an event in our system, you know, either automatically or with a review step. People are. You know, want that ability. And I, I can see a, a, a path in a not too distant future where it's like, all right, you get that, they get approved, and then AI again comes in as soon as they're approved, starts picking those up and like teeing up a newsletter, it's like, all right, you wanna send this? Like, that's about as close as I can imagine,
Brendan:Mm-hmm.
Ryan:autonomous and, and that even that would be like really awesome. Right.
Brendan:Yeah. Right.
Ryan:mean, huge time saver if you could get it to extract the information, right. Which I think is probably the hardest part about that whole thing. Um, but, you know, I, I feel like you could get a, a fairly reasonable end-to-end system running either fully autonomously or with those, you know, a few checks right before, you know, publishing an event or sending the newsletter where you can sort of spot check it and make sure it's not, you know, doing something it shouldn't be doing. Um.
Brendan:Yeah. Yeah, that, that, that's the, the check-in is a thing that I'm always thinking about with. With coding primarily, but I think with what you're describing there with a newsletter would make sense too. You don't want it to just throw something out. You don't review it, you don't, you, you still want that person. And that's a, a strong reason why I feel like, you know, developers, the rule might change, they're, they're, we're not going
Ryan:Right.
Brendan:Because you've gotta still be able to make sure you're not throwing a bunch of slop code out there, but also from marketers or, you know, uh, people, people doing newsletters, whatever. I wouldn't want to subscribe to something like that if it's just completely AI generated myself. Um, so that review step is still something that I think is really important
Ryan:Yes.
Brendan:this stuff.
Ryan:Yeah, I, I, there were a couple things there, and I'm trying to remember what they all were, but. The importance of, all right, so you said developers aren't going anywhere. I agree with that. Right. Um, I do think, you know, with the proper tooling, you could prevent code slot.
Brendan:Yeah. Yep.
Ryan:think, I think. It's probably not quite where it needs to be. And B, most people don't have that sophisticated of a system set up, right? Like, you know, you need all the pieces, right? Testing, style, guide,
Brendan:Right.
Ryan:you know, linting
Brendan:And
Ryan:and, and that's complicated, right?
Brendan:Right. And I think a lot of that just lends to the fact that we as developers think
Ryan:Mm-hmm.
Brendan:But you know, a vibe coder or someone who's never coded before isn't gonna be thinking
Ryan:Right.
Brendan:And so unless the tool just comes in automatically knowing all that, and maybe one day it will, like, I think that's the barrier right there, is that. Uh, yes, we could get there to what you described, but we've gotta think about it as a developer. What are the, what are the pain points that we want to fix and have the AI help with or remember, or whatever.
Ryan:Well, and then the second piece of that was. I was gonna say, all right. Those things I think can be solved with technology
Brendan:Mm-hmm.
Ryan:and maybe this next sentiment could be as well, but I feel, feel like it's less likely is like the role of developer will be about the, the greater architecture of everything,
Brendan:Yeah. Yep.
Ryan:that is. Subjective, first of all. Right? So it's like, all right, it's the, you know, you can't apply, you know, a specific rule to all situations.
Brendan:Right.
Ryan:Um, you know,'cause then you're gonna end up with like an app that's all service objects, right? Like, it's just, it won't,
Brendan:Yep.
Ryan:know, it's not ideal. Um.
Brendan:Callback callbacks
Ryan:Right. Uh, you know, so yeah, I, I think I agree. Like I don't think developers are going anywhere. Right. It's just, I hope, my hope is that we just become more efficient and we can continue to focus on, um, more challenging things and not have to deal with the mundane stuff quite as much. Um.
Brendan:Yep.
Ryan:And then I, I was gonna comment on your other, uh, thing, which was, you know, you don't wanna subscribe to a newsletter that's AI generated. And I agree with that too, which is why the human component is important. Um, like AI could probably be trained to some extent to do better curation, but the doing so I think will be prohibitively. Expensive, um, to get it to where it would be useful, where humans can just sort of do it quickly without a lot of cost. Right? So, but, but if I can, again, take out the part where I have to read through a hundred emails and each one extract the events and manually type them into the database and then manually. Create the newsletter. Like if I can get rid of all that and all I have to do is like swap in, you know, swap, you know, remove this event at this event, you know, maybe move the order around a little bit. You know, put it into to sections that, you know, make sense for whatever the newsletter is. Write an intro, even if that is partially AI generated as well. I, I think you would get to a point where, you know, that is a newsletter that somebody would want to read. Um.
Brendan:Yep.
Ryan:if you rely too heavily on AI to do the writing, it becomes, uh, transparent, uh, that the AI wrote it. And then there's this, uh, lack of authenticity that people, you know, ascribe to it immediately. Which again, I'm not, I'm not sure is a good thing or a bad thing yet. Um.
Brendan:Mm-hmm.
Ryan:You know, I saw an interesting thing. I'm gonna, uh, uh, Neil deGrasse Tyson, I follow him on Facebook and he put up a post about AI and fake news the other day. I'm probably butcher it. He's predicting that the end of the internet will come when people that were against fake news can't tell it's fake news anymore and then lose faith in the, the, the factual nature of the internet as a source of truth. I was like, yeah, that's kind of already happening, I feel like, and you know, one of two things will happen. Either that'll be true or people will. You know, there'll be some sort of paradigm shift, which I think is much more likely than the internet dying, um, where, you know, sources can be more easily verified or, you know, you, you see stuff where, uh, you know, the true decentralized internet, like every author has to sign there. You know, page with the blockchain or things of that nature, like I, I think that's probably where we end up in order to verify that it came from somebody, but who knows how long it'll take us to get there.
Brendan:Yeah, I mean that, that, that also, I don't wanna go too much longer on this, but it, it's making me think of another thing that I've heard of, I dunno if you've heard of this, but model collapse, is basically if everything is AI generated at some point. All the AI is doing is consuming AI generated
Ryan:Right.
Brendan:and how do we know if it's any good anymore? And then eventually the model will collapse, basically, and it's not, you can't trust it anymore,
Ryan:Or at least it doesn't grow. Right? Because at some point, like if, if humans stop creating content, then there's nothing new to go off of.
Brendan:Right,
Ryan:I've, I've thought that too, and it's, I don't know how that
Brendan:I,
Ryan:is solved, but I'm sure we'll solve it in some way. I.
Brendan:yeah. Right. So. All that said, I'm, I'm still in favor of keeping the humans around to help out with some of the tasks, but definitely making our lives easier would be be nice. uh, mean, I think that's kind of the point of where we started, right? You were, you're looking to build features for your users that are sure AI's in the mix, but the reason you want AI in the mix is because you wanna make it easier to use the platform to give you, be more productive, get things done
Ryan:I, I wanna save 'em time. That's our whole promise. So we're gonna allow you to do marketing in 15 minutes a day instead of,
Brendan:Yeah,
Ryan:know, hours. Yep.
Brendan:Yep. So unless you have any other thoughts, maybe that's a good place to just, uh, stop and, uh, I'm sure we'll have lots more thoughts on AI as, as things progress, but
Ryan:That sounds good.
Brendan:All right, sounds
Ryan:All right.
Brendan:for listening. See you next
Ryan:See you.
People on this episode
Podcasts we love
Check out these other fine podcasts recommended by us, not an algorithm.
Code with Jason
Jason Swett
Standing in the Fire
Very Good Software
Remote Ruby
Chris Oliver, Andrew Mason, David Hill
IndieRails
Jess Brown & Jeremy Smith
Startups For the Rest of Us
Rob Walling
Vibing with AI Code
Brendan Buckingham