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
Building with an AI Orchestrator Workflow
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
In this episode, the hosts discuss building a “universal scraper” for event calendars and a new AI-assisted workflow using a long-running orchestrator.md thread.
00:12 Why Change AI Workflow
01:29 Orchestrator File Setup
04:36 How Memory Persists
07:33 Platform Constraints WSL
08:45 Git Worktrees and Isolation
12:07 Scraper Build Approach
18:26 Incremental TDD Development
20:17 Making It Truly Universal
22:39 Universal Scraper Explained
26:03 Wrap Up and Next Steps
LINKS
Questions or comments, email us at railsbusinesspod@gmail.com
Hey man. How's it going?
Brendan:Pretty good. How about you?
Ryan:Another day here at the office. Can't
Brendan:Yep, yep.
Ryan:Um.
Brendan:something.
Ryan:Yeah, for sure. Uh, all right, we're gonna, I think we're just gonna dive right in today. Um, you know, short preamble, uh, but I know you've been working on a universal scraper for LOC bowl to go out and, you know, sort of, uh, scrape calendars for events. And you mentioned that you are using a different AI process than you typically do. I want to hear. About what that is and why you changed, you know, what was the for that?
Brendan:Yeah, so there's a couple things there. It, it, and this process isn't drastically different, but it's, there's a couple different approaches and I'll just kind of start with those and then I can kind of, we'll, we'll see. It might make sense to get a little detailed as to what I was doing with the universal scraper, but let's just start high level for now. So. The, the biggest change really was that, and I, I'm not using conductor for your tool or, or for your, your app. So that's one thing that's a little bit different than what I'm doing at work, but
Ryan:good because it won't work at the moment.
Brendan:yeah. Right, right. There's,
Ryan:later, but that, I'm slowly trying to untangle that.
Brendan:yep. But I, I am trying to figure out ways to. Do similar types of things without a harness like conductor. So the, the first thing that I did, and I've been trying this with, with a lot of different projects now, and it, it has pretty good success so far, but I basically have an orchestrator. MD file, and I haven't committed this to your repo 'cause I haven't even told you about it yet, but it might be something like you, honestly, you could use it or, or not, you don't have to use it. But the way that I, I use it is I have one long running thread and the idea is essentially like I'm always gonna work from the orchestrator as if I'm gonna do planning or anything else. And the orchestrator part of the MD file is. Telling it to kickoff sub-agents to go and do the actual work. So that's one part, and that, that's working pretty well. Um, like I haven't noticed anything significant about that. That's like mind blowing, but it is kind of nice to have like the long running thread. And then I go to, sometimes I'll go to other quads and do other things and I'll come back, but there's always this one thread. It's almost more of a memory thing, I guess is the, the better way to even think about it, because. I can kill that thread, but I can come back and, you know, similar to a context window, but not, it's like a different, a little bit different. I don't know how to quite explain the difference between the context window and this orchestrator. I just know that by having this orchestrator file, it automatically knew exactly where I was. You know, when I pick it up even a week later or something like that.
Ryan:Okay, so hold on. Let's just make sure we're speaking the same language and understanding what it is. Right. So this is just a markdown file
Brendan:Yep.
Ryan:are starting a, a chat with,
Brendan:Right.
Ryan:you know, is that prompt like a use the markdown file and. Or, or is that like, is or is Orchestrator MD like a, um, you know, a, a convention that, whatever tool you're using actually knows about? Or do you have to like tell it to do the thing as part of your initial prompt?
Brendan:That's a good question. It's actually been a while since I set it up. Um. I mean, there was a prompt initially where I basically, you know, said, Hey, look at the, or, it, it created this orchestrator thing. So I had a prompt to say, I want you to be an orchestrator. I want you to use subagents. I want you to kick off processes. I want you to remember things, stuff like that.
Ryan:Okay.
Brendan:Um, but yeah, it, I'm, it's actually, I'm something, I'm trying to just think about it for for a second. Like how it's actually remembering to use it.
Ryan:Right. I mean that that was like, I mean,
Brendan:Because I'm just using Claude Code, so Yeah, like, like,
Ryan:Desktop or locally.
Brendan:It's local. Yeah. It's the terminal.
Ryan:Okay. And yeah, interesting. I'd be curious to know how it knows to use orchestrator.md. And then the other thing that stuck out to me is like, you know, it, you can go away for a week and come back to a new thread, not like a, a thread that was left open.
Brendan:Yep.
Ryan:where to pick off. So, uh, a pickup, uh, so memory, right? Like is that being stored in the orchestrator file? Is it
Brendan:Yeah. Yep. And I mean, I can even give an example of that because I guess it was like last weekend you had the, the Docker spec file thing and you asked me to take a look at that. So I switched over to that and um, and then my computer actually restarted as well, so like I had a couple things where I was like moving on to something different. And everything got wiped from where my terminal was the last time I worked on the universal scraper. So I came back to it and I was able to say, okay, I, I forgot this and this and this. Can you remind me where we were? And it pretty much brought all that stuff back. It even like saved. Uh, like test JSON and we can get into what that means in a little bit if we want to, but it, it saved a bunch of different things, specifically about the universal scraper. It was able to tell me, yep, we've built it this way and it's doing this and this and this. Now I also haven't, I haven't got to the point where I've, I've started like multiple different, uh, branches, so I'm not sure how it would work with that sort of thing. But so far with the one. You know, feature the universal scraper. It's working well to be able to just like, keep a context or keep a memory of what's going on. You know, with, with that,
Ryan:yeah, so basically you've made your. Local interface operate like the desktop version where it has at least some history. I don't know how it works, but it clearly, I can go back to a, a chat thread I had three months ago and it picks up wherever we left off. Um.
Brendan:I, I guess the, the, the similarity to what we talked about, I, I think it was the last episode, was the command center thing that I was using for productivity. And I still use, like, it is sort of similar to that idea. It's not what made me decide to do this. It was actually a different tweet that I'd seen where someone was like, Hey, you should try this. So I tried it and it seems pretty cool, but it is similar where like the command center also has a bunch of files where I can come back anytime to that as well and just be, be like, Hey, you know, you're, you're the command center. Like, tell me what I need to do or here's these different things. That is one thing I've been doing with this orchestrator thing is I do, the couple times I've had to come back, I have reminded it like, Hey, you know, you're the orchestrator. This is, or, or like, where were we? Things like that.
Ryan:Mm-hmm.
Brendan:I don't know if I have to do that or not. But that's something I have done.
Ryan:Okay. And what, and what was the, I mean, was this that you saw a tweet and you wanted to try it sort of thing, and now you're just sharing what you found? Or was there some other driver of like, Hey, the way I was working before wasn't working, or it was too cumbersome or. You know, was there a problem that you were trying to fix or did you fix a problem you didn't know you had?
Brendan:Yeah, there wasn't really a problem except like, like the development work that I do for you, I do on a different machine. It's actually a Windows machine that runs Linux for via WSL,
Ryan:I'm sorry,
Brendan:yeah. Right.
Ryan:that's
Brendan:Yeah.
Ryan:these days, but that used to be a huge thing.
Brendan:Yeah. I mean it, it actually, that part works out pretty well except. You know, conductor for one thing is only on, uh, Mac anyways, and there's other harnesses, but it seems like all the, the, the hot cool harnesses are, are only on Mac. So there was that limitation right there where it's like, okay, I'm, I'm in this totally different environment, so maybe this is a good place to try this orchestrator idea to just see. Can I get some of the benefits of, and I haven't, you know, gone to running multiple work trees and multiple branches and all that kind of stuff. I imagine this probably could do something like that if I wanted it to, but I haven't even pushed it that far. Um, but that's kind of the, the biggest thing was really just that I'm on a different platform than I normally am that doesn't have the ability to, well it doesn't have the things that I'm used to using on Mac, so I'm trying something different here on, on this.
Ryan:that makes sense.
Brendan:Yep.
Ryan:I. Yeah, I, uh, one thing I've realized in the last couple weeks is how, uh, I am on certain, I call it technologies or. of developing, right? Like, so I, I watch AI or Claude, you know, a lot, and as it's going through its thinking process, I try to follow along as best I can and it's like, all right, you know, the way it searches with grep and glob and the way it uses certain rubies and taxes and the way it uses work trees has made me realize that there is a lot deeper to go in certain areas, especially like get, you know, I'm, I would say I'm a. Average GI user, right? But I
Brendan:Yeah.
Ryan:very deep the
Brendan:Mm-hmm.
Ryan:see, uh, doing, it's like, oh, I didn't even know I could do that. Like, I didn't know what work trees were until, you
Brendan:Same here.
Ryan:and a half ago.
Brendan:Yep.
Ryan:And I tried one out the other day and I, I didn't even realize it was like a get thing or it
Brendan:Yeah.
Ryan:I don't know if that's actually.
Brendan:I, I think it is built into gi. Yeah.
Ryan:it's a requirement or just a feature, but
Brendan:Mm-hmm.
Ryan:somehow I got into a work tree and I was like, I don't even know how to get out of this thing, or commit or clear, like back to, you know, ma, you know, master main, whatever.
Brendan:Yep.
Ryan:and I was like, oh, I, I, this is confusing.
Brendan:Well, I mean, we can touch on that real quick. Maybe you got the answers already. I assume you did, but you know.
Ryan:I, I mean, I figured it, I mean, GPT is great for
Brendan:Yeah, yeah.
Ryan:you know, a feature that I don't know how to
Brendan:Right,
Ryan:so
Brendan:right.
Ryan:know, I, I quickly got out of it, but I didn't, it took me a while to wrap my head around it,
Brendan:Mm-hmm.
Ryan:oh, this is like a branch, but it's not really a branch
Brendan:It's more like a copy of your repository, is the way I look at it. It's not, I mean, you can use it.
Ryan:it's a namespace. So you can have
Brendan:Yeah.
Ryan:workspaces, but it, at least the way it was doing it. Didn't do it that way. I, I don't know. But
Brendan:Mm-hmm.
Ryan:it quickly, intro, the reason I brought that up is 'cause it quickly introduced the issue of like, all right, well they're both using the same database.
Brendan:Yeah.
Ryan:Right. And that was the part about conductor that stood, stood out to me. It's like, all right, can I actually isolate the entire environment? So like, I have two databases running, I have two, well, one database was separate tables, right?
Brendan:Yeah.
Ryan:or. uh, one elastic server was separate. Indices and name spacing, all of that. And that is where I immediately ran into trouble when I was, you know, this whole, I tried to get conductor going, the amount of work it was going to be to make my, make the code base. Um, compatible,
Brendan:Mm-hmm.
Ryan:name spacing, database tables, Elasticsearch, memc, you know, all the things to make sure that they're actually isolated environments. It's like, well at this point I could probably just use Docker and it'd be faster.
Brendan:Yeah. Mm-hmm.
Ryan:I've been resistant to Docker 'cause I just. I don't know. I'm used to, I've, I've always developed locally. It just seems easier to me. Um, but anyway, uh, that was a little tangent. Where were we?
Brendan:No, I mean, I, I think I was, I was pretty much done talking about the orchestrator part of it, and, you know, the, the, the next part that I've been doing a little bit differently here is, and part of it just might be the, the way that the universal scraper, I guess, needs to work, or at least I felt like it needed to be built. You know, I, I need to essentially iteratively go through and say, here's the page that I wanna scrape. I'm gonna take one element at a at a time. And it's probably would be worth talking about how we want the universal scraper to work to make this make sense for people. But what we're trying to do here is build A-A-J-S-O-N as the the config, so you don't have a bunch of columns in the database. You've got A-J-S-O-N blob, we'll call it. Yeah.
Ryan:that was the word I was gonna go for. So
Brendan:Yep. And we want it to map to, you know, there's a key that is, you know, what, what's the thing we're trying to put into Lable? What's the, the, um, the, the name of the Yeah, thank you Attribute in Lable that we want to map an element in the HTML to
Ryan:Right?
Brendan:and. That part's working out pretty well, but I was doing it one at a time, so, you know, let's just say we got the, the page title. So I do the page title and then I'd have Claude go out and look at the page and figure out what is the attribute or what's the element, or it actually found a lot of JSON ld, um, elements, I guess. I don't even know what those actually are, but
Ryan:Schema for.
Brendan:yeah, that's what it seemed like. Yep.
Ryan:right? Essentially.
Brendan:because it, it loved those.
Ryan:more reliable than
Brendan:Yep. Yeah, that, that's what I gathered when it was telling me, oh yeah, this is great. I found all this stuff here. Um, but yeah, like the, the challenge is making it universal though, right? Because in so far I've only really been able to push it against. This one website, I wanna do it against a bunch. But it, it was pretty nice to be able to go to my orchestrator, say, okay, here's what I've got. And then I think what it was doing, and this is kind of where the, the little extra power came, is it would send one agent to go and do the web search on that page, and then it would take the other agent, it would look at the code, and then they would kind of combine, you know, report back to each other and kind of finish the feature.
Ryan:How many times does it, does it say, uh, I'm tired of waiting for the subagent. I'm just gonna do this. it finishes and then it says, oh, I just got the results back from the subagent and everything checks out.
Brendan:Uh, it hasn't done that to me, so
Ryan:I,
Brendan:Zero so far.
Ryan:like daily.
Brendan:Yeah.
Ryan:I mean, when I'm mu Yeah. I'm not using an orchestrator, so maybe that's different. But like when I'm in Claude desktop,
Brendan:Mm-hmm.
Ryan:you know, working from my iPad, that happens all the time and I'm like, why did we even need the subagent apparently,
Brendan:Yeah. Right. Brendan tokens for no reason. Yeah, that, that's an interesting one. I don't, I don't know if you have anything else you want to talk about on that, but is that similar to the, I feel like we talked about something else in one of the episodes about some bug. Is that related to that bug that you had talked about?
Ryan:so the bug was, and I still see this regularly, and actually today I saw it for the first time, um, in, you know, when I'm using vs. Code extension for Claude. But the. The permissioning. So it sometimes respects it and sometimes doesn't, right?
Brendan:Oh, that, that's what it was. Yep.
Ryan:be a prompt where it's like, do you want to run this migration? And, you know, sometimes I walk away, right? Or, you know, change screen
Brendan:Yep.
Ryan:and I come back and it's like, oh, I just. It just went, didn't wait for an answer. I just decided, Hey, I, I'm gonna run this.
Brendan:Yep.
Ryan:happening a lot. This is the first time I saw it locally in VS code, but it happens a lot in desktop and it was usually in subagents. And what would happen is it's in a subagent, there is no interface to the user.
Brendan:Right.
Ryan:Right. So I think it eventually skips it, and then it would sometimes get, you know, once the sub agent completed it would, some, it would get caught in this weird loop where like, then all the prompts would come up, but it's already done all the works. Like, what are you prompting me to do? Like, I, I can say yes or no, it doesn't change anything.
Brendan:Yep.
Ryan:annoying. Eventually I started a new thread, right, because it, it can't seem to get out of it, it that, that's gotten a little bit better lately. It's concerning that now locally, that it just completely ignored the. boundaries it's supposed to have. But that, that, like I said, that was the first time I'd seen that happen.
Brendan:Yeah, that's interesting. I haven't seen. That happened at all? Um, I've actually seen the opposite, where if it can't run a command, this might not be the opposite, I guess, but it's like what I've seen. So I clearly have screwed something up in my WSL setup so that everything has to be run through pseudo.
Ryan:Okay. Yeah.
Brendan:mostly it's like, uh, docker commands and stuff like that. I don't know how I got into the state, but it's fine. It's not that big a deal until I tell Claude to do it and then it can't run those commands.
Ryan:Mm-hmm.
Brendan:but it's not even that it's asking for, for permissions though. It's just like, I can't do this. There's no way for me to allow you to, to, to say yes to this. So go run this command for me, basically is what it ends up doing. But, um, I. So, I don't know, maybe it's not exactly the same kind of thing, but it, it is a, an interesting thing where it's like for me, instead of it, you know, waiting around or just skipping it, it'll just say, no, I can't do this. So here you get it back and maybe it would, that would happen if it was the same type of thing in your situation as well.
Ryan:Yeah. I mean, when I'm on my iPad and I'm just using the default cloud environment, it often tries to run like the test suite or boot up a server and it's like, oh, there's a Ruby version mismatch. I'm like, why can't you resolve that?
Brendan:Yeah. Right?
Ryan:not that hard.
Brendan:Mm-hmm. Yep.
Ryan:I can't install Ruby. I'm like, why? Why can't you install Ruby? It's not a destructive action.
Brendan:Yeah. Right.
Ryan:so things like that are interesting. Obviously there's workarounds, you can set up your own environment and do all that stuff. I just haven't gotten to it. um, I was gonna say back to the scraper, so. You said you were going out and doing sort of one attribute at a time. Was that just because you were going through the development process and trying to like, you know, figure out what each attribute was and like how it was gonna work? Or is that like the actual process of the scraper where it does one at a time?
Brendan:That was my, that was just me developing so. I, I've, I've messed around with some of these other, you know, I've, I've built not a universal scraper for you, but we built the, um, parser I guess, or whatever. Yeah. And I did give that one a chance to just. Do the whole page and it just came back and it was a mess. And it's like, well, so this time I decided to do incremental that for that reason alone, like just knowing that it's probably not gonna get it right anyways. And this also gives, gives me an opportunity to, oh, that's the other thing I didn't mention. So I also told my orchestrator that you always have to do TDD. So it gives it an opportunity to do that sort of development and, you know, write the, I don't know if it's actually writing the test in the that order, but. Because it does it so fast. It's like, okay, well we're done, but here's your test. So it's doing that. But for, at least for me, then it gives me the opportunity to run the spec, make sure it's passing,
Ryan:Right,
Brendan:it myself quick, make sure it's passing, and then just feel confident that, okay, we've at least got a, a safe point where if the next thing breaks that, then we've gotta go back and fix that too. Right? So.
Ryan:That makes sense. What other challenges have you come across in thinking about this? This is like a very meta feature,
Brendan:Yep.
Ryan:Um, you know, so sometimes it's just hard to reason about it. Um,
Brendan:Yeah.
Ryan:what have you found difficult? has surprised you? Anything worth sharing there? Or is that too in the weeds?
Brendan:Um, well, I honestly, I mean. Once you and I just hammered home the concept of how it should work, and I got that understanding, like I got past most of the, the hurdles so far. Now again, I've only worked with one site to scrape so far, so now I'm really interested and I still got a little bit of like one tiny little thing I gotta fix there where like everything was working, but then when I put it into the actual. App, it's running into two, um, uh, validations basically.
Ryan:Sure.
Brendan:So gotta fix that. But yeah, I'm really interested to see what happens when I give it the next website that has a completely different dom. And, and then this goes back to the JSON blob. Again, I will need Claude to help me build out what that looks like, but. Is the code actually gonna be universal? And that is something I did run into now that I think about it. Like there were a few times where going through my IT iterations, I don't remember what specific elements, but there'd be an element and I would tell it to do, you know, make it, it already knew to be universal, but I would tell it, you know, go get this element and then I would actually do a little code review. On just that, that commit. And I would see something in there that looked a little too specific. Like, why are you calling this HTML? Like how do you guarantee this is gonna be universal on another site? And of course it'd be like, oh yeah, you're right. And then we would, what we would essentially end up doing is it would start stripping things back out and putting more into the JSON blob and less into the code, which is what I wanted all along. And I, I do feel like there was a point now that I think about it too.'cause this has been like over a month of, of me working on it. So some of it's just like flooding back to me right now. Um, there, there was a point where I feel like we finally had an under understanding where I had told it it needed to be universal. Yeah. Basically.
Ryan:Okay.
Brendan:Pretty much. It was like, I, I don't remember what I said specifically, but it was something along the lines of, okay. This is the way it actually needs to be. Like, you're doing it this way, which is close, but like we, what we really need to do is this.
Ryan:Mm-hmm.
Brendan:And then it was like, oh yeah, that seems right. And it's actually been better ever since.'cause that was maybe like halfway through. So the rest of them went a little smoother because we'd been able to iron out that understanding there. Um, I, yeah, and I wish I had the prompt 'cause it was pretty, it would be pretty helpful I guess to share that. But it's not that big a deal, but.
Ryan:All right. Well, for everyone listening, the, the idea of the universal scraper is essentially that you have a, a, a Jason object that holds the definition to point to all the elements a calendar that you would need to then programmatically sort of loop through that calendar. extract specific event details to store in a database, right? So you might have things like, all right, here's the main calendar, URL, and here's the next page link. And the previous page link. And, you know, there's a, a section of like, all right, um, on each page you need to loop over all the event objects. So like you might have a definition of what the event object would be. And then within that object, know, here's where you'd find the title, here's where you'd find the, you know, start date, here's where you find the location, and so on and so forth. And the challenge is abstracting out the page specific stuff from the definition, right. And, and making sure there's a clear separation so that. You know, we know that the definition is gonna have this form and need these things, and then it is a matter of, you know, ideally AI going and being able to fill in the blanks. And even in the short term, it can be, you know, a programmer or even a, you know, customer success person that can, you know, inspect a page and find the, the attributes or the CSS selectors or the whatever selectors we're gonna use to. the pieces, right? But then you just store the definition and then the rest of it is really, has nothing to do with ai. It's
Brendan:Right.
Ryan:just loop handle rate limits, you know, uh, uh, good citizen sort of stuff.
Brendan:Yep.
Ryan:right. And, you know, the, you know, the big challenge with any event scraper is essentially you have to build some sort of integration. Right. You know, sometimes there's a feed. A lot of times there's not, if there's a feed, it's a little bit easier because, you know, that doesn't, there's a, there's a more structured interface.
Brendan:Yeah.
Ryan:but when there is no feed and you have to scrape, you know, it might work for a while and then someone changes the page, right? They update design, they change website providers, they do something
Brendan:Yep.
Ryan:it breaks. So the idea of this universal scraper is that it can ultimately be. Self-healing. So like we can detect when there's a change in the quality of the data that we're scraping, can go out and say, Hey, something, something's not working the way it was working. Go find the new selectors. Or, you know, the, the, thing that needs to be updated to fix this and, and heal, uh, the definition.
Brendan:Yep.
Ryan:Um, and that's where we're going with that, is to get an idea of, can we put. These scrapers more on autopilot and less, uh, reliant on a developer to come fix it when it breaks. Um, and also
Brendan:Right?
Ryan:just knowing that it breaks in the first place, which,
Brendan:Hmm.
Ryan:you know, sometimes is obvious and sometimes less obvious.
Brendan:Yep. Yeah, exactly.
Ryan:just a quick description.
Brendan:Yeah. Yeah. I think that was good.
Ryan:Um, all right. What else we got?
Brendan:I mean, for that, that's kind of the whole process, and it's, it's really nothing, you know, revolutionary, but I feel like it's worked pretty well. It's at a point where I'm, again, it's, it's easy to come back to for one, if I need to take a break or I just don't have time since, you know, this is just a part-time thing. Anyways, it's like that, that's been really helpful and yeah, I, I would say that. At least with this site, the process has been really good. You know, we'll see if it works the same way for the next ones. Hopefully it's even easier 'cause ideally it's just tell me what the JSON blob is and I don't have to do any more coding, but I have a feeling that there's gonna be some more iterations either way.
Ryan:Well, I found out this week that there's actually providers who entire calendars with JavaScript and
Brendan:Yep.
Ryan:dom at all. And I was like, why would you do that? I don't understand. Um. But for now, I'm not gonna support those. So, um, sounds like a different problem for another day.
Brendan:Yeah. Yep, exactly. See, I don't have much else to add on it, but I mean, it's been a cool project. I've enjoyed working on it and it's, yeah, I've definitely learned things that have been, you know, useful going forward as well.
Ryan:Cool. Well, I'm excited to
Brendan:Yep.
Ryan:I haven't had a chance to look at it yet, so you can, you can let me know when it's ready and I'll take a peek.
Brendan:Will do
Ryan:Yeah. Um.
Brendan:anything that you wanted to talk about?
Ryan:I mean, we always, there's a long list of things we can always talk about. Do you want to keep going or do you wanna call it here for today?
Brendan:Uh, yeah, I mean, I think this is probably not a bad place to wrap it up. Um, I don't know if you had anything else around this sort of thing you wanted to cover quick or if.
Ryan:Hmm. No, just the, the normal AI's moving fast,
Brendan:Yep.
Ryan:hard
Brendan:Yep.
Ryan:up, but, um. You know, if there's anything, uh, listeners you guys want to want to hear, we, we said coming into this, we're gonna talk a little bit more about business and then we talked about AI again. So apparently we're not very good at keeping our own promises, but we'll try to do, you know, try to provide some variety. Um, of course, if you wanna, if you like the AI talk and you want us to keep going on that, let us know that too, and we'll, we'll pivot accordingly.
Brendan:Yeah. Yeah. I mean, pre-show, we, we were just talking about how it just dominates so much. It's, it's kind of hard to talk about anything else, but yeah, I mean.
Ryan:thoughts are consumed by, you know, all things AI and how it affects workflow, how it affects the business, how to, how to take advantage of it, how to do all the things. So, yeah, I, I sat here for 20 minutes thinking. I was like, what else could we talk about. Like, it's like now that's, that's what I'm thinking about. So, oh, well.
Brendan:Yep. yep. And I think a lot of people are there too,
Ryan:All right. let's call it there.
Brendan:Yep. Sounds good.
Ryan:All right.
Brendan:All.
Ryan:I.
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