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
Ryan Castillo
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Ryan Castillo returns to talk about pivoting Knowatoa from an SEO play into an AI-visibility tool — and why the analytics everyone wants to track are becoming impossible to trust in an AI-search world. The back half turns into a deep dive on Ryan's AI coding workflow: skipping the Claude chat interface entirely in favor of firing off cloud agents from Cursor, letting them run unsupervised, and reviewing PRs cold instead of living in the conversation. Plus: why "throw away the code, not the plan" beats hand-holding an agent through every step, and why the AI labs have zero incentive to ever let you leave the chat.
Links:
- Knowatoa
Questions or comments, email us at railsbusinesspod@gmail.com
Today we've got, uh, Ryan Castillo back with us. We, we were just talking pre-show, it's been 30-plus episodes since we recorded for us, so, uh, glad to have you back, Ryan
Ryan Castillo:Yeah, thanks. Yeah, again, congrats to you guys. I mean, 30 episodes is a lot, you know? So it's,
Brendan:Yeah
Ryan Castillo:cool to see how the show has progressed since we first talked
Brendan:Yeah, definitely, 'cause I, I think, I didn't look back before the, the call, but I think you might have been our second guest, second or third, somewhere around there. You, it w- you were definitely one of the first guests we had, so it is pretty cool to, to have you back and kind of check in, and we can just jump off from there. So, you know, last time I think Noatoa was fairly new, but, you know, sounds like things are going pretty well. If you wanna just kind of talk about kind of where things are with Noatoa and, um, you know, how you guys are doing, that'd be great
Ryan Castillo:Yeah. Um, so back then, I think we were running two products. We were-- We had Wafris, which was like a web application firewall. We had gone to a, a number of, of Rails conferences and promoted that. Um, but in the midst of all that and trying to promote it, AI started becoming a thing. And I think when we first asked ChatGPT about Wafris, it started sp-spurting out a lot of things about like Waffle House and like waffles. It was, it was basically hallucinating about the brand. And we were like, "Huh, that's, that's interesting." And, um, a lot of founder friends, we were like, "Okay, what happens if we start running a bunch of people's businesses through this?" You know, and then we started kinda building some scripts and stuff like that. And, um, that actually started rolling, and so we decided, "Hey, w-why don't we try and build an AI visibility product?" Um, and that's how Noatoa was born, and it kept on getting more and more traction, and kinda took the eye or the eye off the ball with Waf-Wafris. So I think sometime last year, we decided to kinda sunset it. Um, and, you know, there, there have been some pivots around Noatoa. Um, there's a lot of people who have thought that, okay, um, we're losing organic search, we're losing traffic and all that stuff, and, you know, AI is somehow gonna-- AI visibility is somehow gonna replace that. A-and that's just not true. Um, and some of the metrics around AI search aren't great either. Like, uh, I think the main ones are sentiment, visibility or mentions, and citations. Those are decent metrics, but at the same time, it's all dependent on you're tracking, the data that's coming in. So as we've always known, you know, garbage data in, garbage out, insights out. So, you know, that's, that's something that we've, we've been working against as well.
Brendan:Got it. Yeah, that, that's interesting. I, I think about it sometimes even, you know, like I don't have a specific product that I'm promoting, but obviously my company, we use SEO and that's kind of been one of our biggest things is always getting SEO, free SEO traffic essentially, right? And like we are still trying to navigate now when is ChatGPT or Claude bringing up our name. Like Ryan, Ryan, my co-host here, I'm not sure how his company handles it. Um, but it, it definitely hasn't seemed to be like a huge breakthrough or, or like it hasn't overtaken our SEO traffic or anything like that. Um, so I- I've actually thrown Noatoa in front of my boss to say, "Hey, this is a tool we might wanna look into." We haven't pulled the trigger on it yet, but like what, what kind of clients are you seeing, I guess is maybe my, where my question's going here is like, are you, you getting like various clients? Are you getting more tech-oriented people? Like what, what are you seeing?
Ryan Castillo:Yeah. We're actually getting people all over the map. You know, um, in our industry, everyone is targeting the agencies because the agencies are the one that are, that are most ready to buy, um, primarily because they're suffering loss of their client base, right?
Brendan:Sure
Ryan Castillo:they're seeing their metrics fall through the floor, and they're trying to justify to their clients, like, "What is our next step?" Um, but we're also seeing people come in that-- Another, like, primary client cu- customer base is people that are coming in and, um, we actually look at their keywords, and we see that, like, over half of their keywords are vulnerable to, to losing traffic to AI search. Um, so those are a big, you know, user base. Um, the third bucket are people that have zero search volume traffic at all, and they're just AI curious. They're just kind of like window shoppers, or they've started up a new brand and they don't really know how to market that brand. Um, the first two are the ones that we can help the most with. We're, we're trying to do some things to help with that third bucket, but, um, there's a lot of education, you know, across all three that has to happen.
Brendan:Yeah. Are you guys putting out a lot of, like, educational content then? Just, you know, typical content marketing types of things to help with that
Ryan Castillo:Yeah, so this, this is where things, um, of come, come into play as far as like, hey, how, how do we show up in, like, AI conversations? Like you said, you guys are trying to have that happen for your brand. Um, it is a mixed bag. Like, it's been hard for us to put out educational content because, you know, on some levels the thought is, is like, "Hey, we are not gonna actually see any search traffic for that." So in the past six months to a year, we've been really hesitant to do that, mainly because our thought is like, "Hey, this isn't gonna actually bring any traffic to the site. What this is actually gonna do is it's just gonna feed ChatGPT and people are just gonna stay and live in ChatGPT." Um, we've kind of turned around on that and said, "No, we're not gonna work ab- worry about the metrics anymore at all. You know, we're just gonna worry about our branding, and we're gonna worry about helping as many people as possible." And so within the last month or so, we've kind of pivoted towards like, hey, now we're gonna start going about, like, releasing more videos, releasing educational content, participating in more Slacks in different communities, and talks and podcasts and, and all of that.
Brendan:Got it. Okay
Ryan:interesting things there that I wanna dive into, but first, for, uh, I wasn't on the first call with you, right? And I'm guessing a bunch of our audience, uh, maybe didn't see that or know of Knowatoa. Can you just give us like a high-level view? I mean, I, I get AI search, but what exactly are you doing to help with AI search? So that's question number one that I'll kick back to you.
Ryan Castillo:Sure, sure. So, uh, and the product has shifted, right? Um, originally what we used to do is someone would come in and they would submit their brand and, um, we would take those keywords from that brand and look at intent behind those keywords and generate questions. we take those questions and we ask it against all the AI models like AI Overviews, AI Mode, Gemini, Claude, Perplexity, ChatGPT, and just to see what answers come back. And then from those answers, we would kind of look at these metrics. We would see like, "Hey, are all these like positive answers about the b- brand or are all these negative?" Um, how many times is this brand mentioned in these answers? How many time are their competitors mentioned? Um, what kind of citations and referrals to other sites are being mentioned in these answers? And so the product today shows all that. Um, and so originally we were like a visibility platform, but now what we're trying to do is we're trying to pivot that more around, hey, everything that's coming back in this is actually just a small sample size of what the internet thinks about your brand. And so as a user, what you need to do is you need to figure out when looking at this data, um, where do your prospects and customers actually live? Do they live in the sites that are being returned from the data that we're pulling? And what gaps are there in the answers that are being returned? You know, is this brand-- some- is there a feature about your brand that, that's being misrep- represented or is not being covered in these answers? And if so, we help you plan out where you need to show up and what you need to say so that you can attract your customers about
Ryan:Okay. So that's a, that's a good segue into my second question is, you know, putting your education hat on, right? Like, what are examples of what you might recommend, right? How do you show up in AI, right?'Cause y- it, uh, you, you said you were sort of hesitant to do content, right? So, like, what else is there, right, for us that are,
Ryan Castillo:Yeah.
Ryan:AI illiterate?
Ryan Castillo:and the, the clarifying point on there, it was the, the, the kind of content that we were willing to do. Um, right now, if you look at like keywords, let's just talk about keywords, and I'm gonna interchange keywords and questions a lot here. Um, with keywords, what everyone used to do in the old organic way, days is like they, they try to rank for everything. You know, they try to find long-tail keywords somewhat related to their brand, not related to their brand, whatever they could do to bring in traffic. Um, like HubSpot is a great example. HubSpot's a CMS. They had, they had pages that were optimized for like how to write a cover letter. There is nothing in like HubSpot's like product that helps people with cover letters or getting jobs, and so they were just trying to get that free traffic to see if they can convert that into an actual sale. And so in today's world, you don't really want to optimize for like informational keywords unless it's a new topic. Like, you know, Fable is a new topic. That's something that you might optimize for, but you won't… That content won't be greenfield anymore. You know, in a couple months, ChatGPT or all these other AI models will eventually be able to answer all the questions around that. So that's something a lot of like marketing departments that are focused on SEO are trying to kind of resolve theirselves against is like, "Hey, what keywords are we actually still trying to go for? Like what should we optimize for?" So, so in our recommendations it's like, hey, throw out all the informational keywords or at least change your expectations as far as what you're gonna get out of them, um, as far as traffic and how long that traffic is gonna come to your site. But instead, you know, look for all the gaps, again, in the keywords, in the questions that, you know, your content may not be answering. And furthermore, like think about where are the places that your audience lives. They don't just live on your site anymore. Like another thing some people have said is, know, everywhere is a homepage now. ChatGPT is your homepage now, so, uh, LinkedIn is your homepage now, YouTube is your homepage now. So are you putting content in all those different areas? You know, this is something that we always used to, to preach as marketers, but it was just always hard to do. It's like how can we be in so many channels at the same time? Um, now that's kinda more important than ever because by repeating that message, you're increasing the likelihood that you show up in these AI conversations.
Ryan:So it's, it's less that content doesn't work, it's that content has to be more, both more targeted and more broad, if that even makes sense. Targeted in terms of the questions or keywords you're, you're answering, uh, in broad and, you know, the chan- distribution. Um, okay.
Ryan Castillo:Yep
Ryan:All right, so Noatoa then helps find those gaps and gives you sort of a, a, a guide to where you're either falling short or missing, uh, the message in the case of sentiment analysis, it sounds like, uh, if, if I got that right. Yeah.
Ryan Castillo:Yep, that's
Ryan:Okay. right, so the third question that I have for you is, uh, I wanted you to go a little bit deeper on a comment you made, which is you, you kind of threw the analytics out, right? Um, you know, m- maybe that was said, uh, uh, you know, with a, a grain of salt, but I'd like to hear more about that.'Cause as you were talking, I was thinking like, I don't actually care about the analytics that much. Um, you know, I think, uh, Brendan, earlier on you said, "I don't know what you do at Locable." Well, I can tell you, we create content around the stuff that we care about. Um, we don't have a specific SEO strategy. We just put out things that we think will be helpful for our customer base, and we try to reference it as much as possible. But I'm not looking at like, did this keyword rank? And frankly, I don't care. Um,
Brendan:Sure
Ryan:you maybe elaborate a little bit more on, you know, why you don't look at the analytics or perhaps clarify what you do care about.
Ryan Castillo:Yeah. So give more into my background. So before all this, you know, we all had different stops in our career. I, I was working at a, um, you know, like a Fortune 500 company. You know, they're trying to get to a billion dollars, and I worked on, in marketing and attribution there. And, um, you know, everyone, just to kind of illuminate attribution for the audience, is that attribution is kind of the study of trying to figure out, okay, hey, we spent this mon- this money on X, you know, and it resulted in Y sales, or Y leads, or Y conversions. Um, and can we track every single marketing action or marketing touch to this spend or action on our side? Um, and that was very challenging at that job because we were having sales cycles that lasted, like, as long as two years or a year plus, you know. And these were, these were deals that were, you know, six-figure deals. And so in all of that, you know, I would say like, "Oh, you know, the first touch for this user might have been this Google thing." And then the sales side will be like,"Well, we've been, we've been prospecting this deal for six months, you know. Why should Google get any kind of credit?" Or, "Why should the, like, PR side get any kind of credit in this kind of stuff?" And it, and it, it all kind of at some level becomes, comes kind of hand-wavy, like, you know, just made-up
Ryan:Yeah
Ryan Castillo:to kind of paint the picture that you, you want to see because of course departments have to justify budgets and that kind of thing. Um, but the end of the day, um, it's just like what you said, Ryan. What's most important is, like, being useful and helpful to your prospects and your audience. And now with the privacy stuff, the privacy stuff in email, the, you know, it, it's getting harder and harder to track, like, how many times you touch a user. There, there's no way of doing that in ChatGPT. There's no way of doing that in Claude, like, of knowing, like, all the answers that I ask on a given day so that, you know, something that I've bought after I've researched Claude, there's no way they're gonna be able to tie that back to me.
Brendan:Yeah. I, I'd even add just like the iOS has gotten so much stricter and like Private Relay and even just Safari in general, like just strips a lot of that attribution stuff out. Yeah, we, we've dealt with some of that in our industry as well. It, it is a challenge
Ryan Castillo:And so analytics is important, attribution is important. Like when you're running ads, you know, it, it, it is smart to say like, "Hey, you know, we spent $1,000 on these keywords. We got, you know, 100 leads from that." That, that's important and that's interesting. But you also have to realize the flaws in that approach as well. It's like now you can't just say like, "Hey, we're optimizing for all these keywords and now we're getting all this traffic," because that's kind of broken on some level. You know, you don't see… More things are showing up under direct and like plausible than, um, than organic or social or those o- other- all those other kind of things. And so when you kind of try to look at your efforts for your business or for your marketing and stuff like that, what you have to really ask is like, what are we really trying to accomplish here? You know, are we trying to increase leads with this campaign? Are we trying to educate? Are we trying to do these things? Is it really important to look at these metrics? Do these matter? You know, your, your gut comes into play, your experience comes into play, the, the numbers come into play, and you need to kind of be able to, you know, figure out like what works best for you. I know that's kind of like a,
Ryan:Well,
Ryan Castillo:it depends answer, but
Ryan:I, I wanna
Ryan Castillo:is
Ryan:dive, dive into that a little bit more if, if you're, if you're okay with it 'cause… All right, so, um, I have a publication business that I, I run. It's magazines, all right? Uh, they're quarterly publications, um, high-end sort of coffee table style, glossy, thick paper, will sit on a table for, you know, years sometimes. you know, I lose sales because people are like, "I, I don't know if it's working. I haven't gotten any leads from it." And I'm like, "Well, how are you tracking that?" Right? There's no actual… And, and, you know, you can ask. That doesn't mean they're gonna respond with, "Oh, I saw your ad." Like, most people wouldn't say that. Um, and even if they say, "Oh, I came to your website," that doesn't mean they didn't see the ad six, seven, eight times consciously or subconsciously the way, right? There's no way to track that. And I think that's-- there are very few mediums, including, like, the digital stuff, where you can actually get from the beginning end in right? And, like, the closest I think we get is with, like, perhaps social media attribution channels, right? But even that is flawed. yeah, so what, you know, what do you, you know, g- given your background, what would you people, right?'Cause I, I'm, I'm curious from a business perspective, but, like, know, the other side of that is simply just like, all right, uh, the way I would look at it is like, all right, I'm, I'm spending X amount of dollars, right, in all my advertising, it is leading to some, you know, revenue that comes in at the end of the day. And those are really the two numbers I care about, right? Spend and dollars. And if that is in the right direction, right, like spend is going down, revenue going up, or at least staying equal to where it was, then I'm happy. If it start, you know, if costs are going up and revenue stays flat or it goes down, then I'm unhappy and I want to adjust, right? But that's, that's about the level that I'm willing to go in terms of, like, analytics. I'm curious your viewpoint given, you know, where you came from and what you're working on.
Ryan Castillo:Yeah, and, and this came down to the conversations that we would have with the CEO at the company. It's like, "Well, we can't attribute all this stuff to all our other efforts, you know, of our inbound marketing. You know, it's very easy to attribute all this stuff to our sales efforts, like our in-person events and conferences and all that other stuff. you want us to just give up on all these other things? Should we just, like, fire everybody and, you know, kill these departments?" And, you know, the response was always no, because a lot of those efforts are branding and are not trackable. And so, you know, usually the people that are wanting to cut are the people that, you know, don't understand the value of that effort. You know, like you said, that magazine is gonna sit there on that table for years and decades. You know, like, in some industries, that's super important, you know? Maybe, like, there's a jet ski magazine that's been sitting on that table, and
Ryan:Right, right
Ryan Castillo:that goes into that barber shop, they're like,"Oh, man, this jet ski is awesome. Let me see if I can go find it on Craigslist or something like that." Or, you know, even those m- those, those magazines like 10 years old, you know, "That's the jet ski I want," or whatever, you know?
Ryan:Yeah
Ryan Castillo:but I get it, you know? And this kinda goes back to another one of you guys' previous episodes where, where someone was talking about, like, how SaaS is dead. Um, the reason why they're talking about SaaS being dead is because the way in which they've marketed their SaaS and, like, how they've tracked that has changed drastically. Everyone was dependent on organic as this free marketing channel that, you know, Google would just bless you with traffic, and then if you cracked that formula, you would somehow… Your business would grow. And now that that's been taken away seemingly overnight, people are trying to, wrestle with that. And I think more and more people are kinda gonna kinda have to come to terms with, like, how you've been looking at the business, Ryan, is that, hey, all of this stuff is not trackable. So all we can do is try to really think about where our audience is, where they're experiencing the pain that they're experiencing or where they're trying to find information about the job that they want to do, and make sure that we're in those channels with the stuff that they need.
Ryan:Right
Ryan Castillo:the more we think about that as an industry, you know, that's, that's kind of the human, human advantage in all of this.
Ryan:Yeah, I, I, I think that's a, a good point. Um, you know, I would even go so far as, you know, I'm gonna transition the conversation here a little bit, um, and, you know, maybe we can dive into some AI stuff, uh, despite my best efforts to stay away from it. Um, but the… You know, from a business perspective, I'm most interested these days about, all right, how do I take and use AI to make it repeatable, right? A- and in the case of, um, know, what we were just talking about, you know, the, the first step would be, all right, like, what metrics do I care about, right? Can I get those metrics and then feed that back into the AI to like, all right, you know, what's the next step? Um, you know, so the example here would be, all right, we know revenue and we know cost of, of customer acquisition or CAC or whatever you wanna call it. Um, you know, give me a report, you know, each week. Are tho- are those numbers going up or down, right? then I'd like to trigger off of that and say, "All right, uh, costs are going up and revenue is going down. What do we do about it? Give me an analysis. Where are the gaps," right?"Fill in the holes. Give me a list of things that I can try this week to try to move the needle." And I think it's easy to talk about that, And it makes sense. It's very hard to implement that, um, at least with AI, right? You can do it manually the way we've always done it, and I think that's a skill that a lot of us, uh, including myself, fail to, uh, hone on a day-to-day basis. Um, curious, are-- what are you seeing… Uh, uh, we talked pre-show a little bit about your AI stack. You know, so maybe this is just a good segue into how you're feeling about AI and what you're using it for. You know, is that, uh, something that you would consider, um, or are you in a completely different mindset?
Ryan Castillo:Yeah, I, I think pretty deeply about… And, uh, I think there are a number of like, uh, talk submissions I've done. I'm actually gonna do one locally about everyone, not only, you know, SaaS owners or founders, but everyone in- with a technical background that is working with AI, they have to understand where the human perspective and the human experience is most important and where the AI can help and not to blur that boundary, you know?
Ryan:Sure
Ryan Castillo:one, one great example that I'll give as far as like, you know, using AI is, um I think one thing that we kind of talked about is like, hey, one of the moats is your customers, your prospects. You know, as long as you're serving them, as long as you're helping them, you know, your business will or, or be around for a long time. Um, so I try to get as, on as many customer calls as possible. I try to do all the support tickets, I try to do all the emails, because I, as the founder, have to understand where our prospects and customers are coming from. I have to understand that deeply. So, um, I will jump on call with anyone just to understand, like, where they're coming from. where the AI can start to take over is like, okay, I've, I've had, like, 20 Zoom conversations this week. Okay, now let me submit those 20 Zoom conversations to Claude. Let me put them in various projects of the three different buckets that I mentioned before, the agency bucket, the losing, like, click traffic bucket, and then the zero traffic bucket. Okay. Well, now that they've been put into this, these buckets, all right, now I can start having some conversations about, like,"Hey, what are the patterns that you see? What are, what are the things that we talked about in all of these, um, that are in these conversations?" Okay, well now, like, I have those conversations, and now I have another Claude window opened against our code base. Okay, what things are actually doing? Like, how are we serving these, these needs? And what, what pieces of content do we have published to actually answer these questions? And so at some point, you know, the AI starts to co- to take over, but then I come back into the place, and then I'm like, "Okay, well, well, I think that this feature is best for this user or for this use case. I think this article needs to exist for this user and needs to be explained in this way." I don't wanna just rely on it generating AI slop. I know intuitively human to human what is the best for you because I've listened to you and I've empathized with you. Um, and so that's kind of, you know, that's one example of, of how we use AI in our process.
Ryan:Yeah, I, I think I agree with you, right? There, there's gotta be sort of guardrails, right? So you avoid the slop. Um, but I think a lot about, you know, so that, that process that you described, right, totally makes sense all the way through. Is any of it automated? Right, so that, so that you don't have to, like, do the prompt to, like, aggregate the data, know, uh, y- y- you described three or four steps there, and I, I, I won't do it justice to try to repeat them back to you. People can rewind 30 seconds. But that's great, but it takes time, right? How much of that could you on autopilot to get you to the point where you need to, you know, then have your human, uh, intervention?
Ryan Castillo:Yeah. So, um, a number of things are automated in that. So I have skills for extra-extracting insights into those conversations. Those skills map the data into other files. Like, I think Brendan had mentioned like,"Hey, everything should be in Markdown." It…
Ryan:Sure
Brendan:Yep.
Ryan Castillo:and then, so that's capturing all the data there. Um, when I go to, like, building out the feature or writing the blog posts, um, I used to be very big into Claude and, like, living in Claude, but then I, I found that, you know, I would just kind of argue back and forth with Claude all day, and then I'd have to, like, open all these work trees, and it, it just didn't seem like an efficient process. Um, so, uh, I'm good friends with Jesse at Bento, and I've essentially copied his process. And what he does is he uses Cursor and cloud agents. So once he kind of figures out, like, you know, from the conversations or from the insights like, "Hey, this, this post needs to exist," or, "This, um, this feature needs to exist," then I go to Claude, or not Claude, I'm sorry. I go to Cursor and I say, even with this mic, I use Super Whisper and I talk into the conversation. I was like, "Hey, this thing needs to exist. It doesn't exist right now. Go spin it up in the cloud and then open up a PR for it." And then once the PR is open, that's when the human intervenes again
Ryan:Right
Ryan Castillo:out like,"Hey, is this built in the way that I want it to be built?"
Ryan:A- and is that true for content too? Like you have it open a PR, or is there a different workflow for content creation? And, you know, I'm leading
Ryan Castillo:That, that's still-- I'm still honestly, you know, I've mentioned that we kind of are pivoting on, on content right now. Um, so lot of me is trying to kind of like find my voice as far as like how I want us to be represented and to be useful. Right now, the flow is, um, I'll just record a video. Like someone will ask a question, I'll record a Loom, I'll record a video. That video goes to Descript, that transcript goes to Claude. like says like, "Okay, do we need to add more point of views that you're trying to express to your point of views file? Um, okay, how do we refine this down to like a tweet? How do we refine this down to a LinkedIn post, to a newsletter, and to an article?" Um, we have some stuff in place that automates all that, um, but I think it could do a better job. And especially, uh, as far as like the handoff to the human side, because I think what we were doing is, we were pushing it a lot to buffer and publishing it in all those places. But what that really loses is the channel rules. Like,
Ryan:Hmm.
Ryan Castillo:we would just be pushing stuff to LinkedIn, and when we pushed to, to LinkedIn, we weren't considering LinkedIn's algorithm or like their standards as far as like how people should interact with a platform. I think LinkedIn has like a rule of, um, four posts max a week, and then we'll, we'll promote those posts. But those have to be well thought out, deep, you know, useful to the community posts. Or, you know, if you want to show up on Twitter, you know, you should actually participate and have conversations with people, and then as they respond to you, they, they need to… you need to be ready to respond. So, um, as far as publishing, you know, we're still, we're still trying to figure
Ryan:Yeah. All right, one, one last question then I'll cede the floor back to, to Brendan. Um, You, you talk about, you know, uh, you, you mentioned point of view file and other markdown files. How are you organizing that stuff right now, right? Is it, uh, do you have like a, a, a repo that is separate that you put all that stuff in? Are you putting it in your, your code base? Do you have some other system, right? Like, you, you're sort… Brendan and I talk a lot about building a second brain or a command center for the business, right? And you see people talking about like, "Oh, I put everything on Google Drive in this folder." Um, or, you know, a lot of people are like, "Oh, I have a, a skills repo, uh, on GitHub." How do you think about that?'Cause it's something that we're, I'm trying to figure out and I'm just collecting, you know, more and more insights of people who have tried things
Ryan Castillo:Uh, we put it all in our code base. And, and the reason why is because when we wanna publish a post, you know, we do have Claude skills and other skills that are just like, "Hey, know, I forgot to create updates for the last, you know, 10 or 15 features we released over the last couple weeks. Can you, can you just generate posts for this?" And Claude will look through the, the Git logs and everything like that, look through the code, and it'll, it'll come pretty close to, to releasing, like, a feature update for that. Um, the other reason is, uh, by doing that, it allows us to interact with, like, various MCPs, even our own MCP, and use the content of, um… Like for instance, with Noatoa, we, we, we, we give you the ability to choose, pick and choose the questions you wanna track, um, and we give you suggestions and build content for you. Um, if we want to weigh that against our code base, it's-- I'm getting very meta here. Um, "Hey, here's the suggested post. How does this stack against what exists inside the code base?" And so-- Or the content, like what content ex- already exists for this topic that's been suggested by Noatoa.
Ryan:Right
Ryan Castillo:we use that M-MCP in the same Claude session that has access to our code, and then it makes the, you know, it makes the post itself that much stronger
Ryan:That's a, a very interesting insight that I'm gonna write down. Th- this idea of weighing, you know, the suggestions against the code base or existing content. Th- that sort of feels like it needs to be in the list of things that you do regularly. So you for that one. all right, what else you got, Brendan? I've been talking long enough.
Brendan:No, it's all good. This has been great. I, I've actually really enjoyed having this tangent off onto the, like, marketing and SEO and all these things that are very much related to what we're doing all the time, but not, you know, the AI coding of what, of it all. So now I'm gonna take a, a hard left turn and go right into AI coding of it all.
Ryan Castillo:Sure, sure, sure
Brendan:Uh, I, the biggest thing that, that jumped out from what you were just talking about, you know, you touched on using… g- getting out of Claude and getting into, uh, Cursor. And, and I think I understand what you were saying with, like, using the, the cloud agents and stuff, but I'd like you to just elaborate for my own benefit and, you know, the audience. Like, uh, what, what does that workflow look like in whether it's the marketing stuff or even just coding? Um, 'cause you guys are still using Rails, correct? Yeah, so, so what's… L- maybe let's just start with, let's go into the code side now and just talk about, like, a, a new feature. Like, what, what does it look like? What's your workflow look like using Cursor and Claud- eight- or, uh, cloud agents and stuff like that?
Ryan Castillo:Yeah. Okay, so, you know, a- as a founder of a small team, there's only so many hours that we have in a day. You know? And there are a thousand things that come up all the time. You know? And again, going back to the whole SaaS is dead thing, uh, that's why I'm a firm believer against that. There's just so much that we have to do in a given day to, like, keep this thing going. Um, so what that looks like now is whenever I get an email, whenever I get a, a support ticket, whenever I get an idea, what I do is I go open up Cursor. I literally go and I open up a new, um, cloud agent. I pull the microphone to me or I, I'm ready to type it, and I'm just like, "Hey, see if you can make this feature exist." And, and Cursor comes… There are a couple of plugins I think I use. I think I use something… What is this thing called? I'm a… I think it's called Thermos. Oh, wait. Uh, Ther- Uh, there, there's basically this thing that I use after I give it the description, what it does is it spins up multiple, um, agents, and they kinda can argue with each other about the code.
Brendan:Okay
Ryan Castillo:and I let that happen in the background. After I've done my dictation and submitted it, Cursor is, does no longer exist in my brain. I go back to doing all the other founder things.
Brendan:Yep
Ryan Castillo:then at some point, if it's the same day or if it's a couple days later just because we're so busy, um, what I'll do is I'll say like, "Okay, now it's time for me to do a PR review." And I go back in Cursor, I have skills for this. I say, "Okay," like, "look at all the open PRs." And that feature will be one of those open PRs. And at that point, it's time to start interrogating the feature. Tryin'- time to start kind of figuring out like, "Hey, does this actually work the way that I want?" Like, "Is this actually popping up in the place I want it to pop up? Is this actually the user experience so that I want the user to have?" Um, or like, "How does this code look?" Or, "What, what decisions did it make that, that I like or don't like?" Um, and then still within Cursor, you know, I'll do a number of like security checks or like, you know, code quality checks and stuff like that. And then I issue another skill that just, uh, merges it and then ships it to production. And then, at that point, if it was something that, um, a user requested or a support ticket came in for, then you know, I, I send off like the,"Hey, we, we built this thing for you. Thanks for, you know, rec- recommending it," or, "Thanks for reporting it."
Brendan:Yeah
Ryan Castillo:so that's the typical workflow. Some of them
Brendan:Got it.
Ryan Castillo:like pretty complicated. You know, I've kind of been pushing the envelope with it. You know, I
Brendan:Okay
Ryan Castillo:pushed the envelope. Like there's a pretty big architecture change that I want to do. Um, and I know everyone's been talking about Fable and throwing like huge things at Fable. Um, this has just worked for me. Like I don't want to live in the conversation, you know, for my entire day. I got other things to do, you know? So I just want to live in these conversations needed.
Brendan:Yeah. I mean, that's definitely interesting to me because, I mean, my role is not a founder, but, you know, as a manager and having all the different things coming at me, whether it's s- people on the support team, people on my team wanting code review, you know, my CEO looking for things. Like, I've- can never really just focus down on code. Now, I, I will write code similarly to what you just talked about. Now, Ryan and I, we had John Nunemaker on earlier this year. He put us… Well, I've been using Conductor less and less, but the workflow you just described is sort of like what Conductor does. It's like you get your own little work tree, and you kinda, I kinda kick it off and let it kinda run, and then I'll come back, and it, it does reviews, all those kinds of things. And they're basically all skills, but of course, because Conductor's its own harness, you just like click a button rather than using a skill. But that, that does sound pretty similar. Um, but yeah, I, I think… Go ahead
Ryan:I was gonna say, I mean, uh, to, to me, Conductor is slightly different, right? That the problem it solves is that, you h- you know, how many of those concurrent conversations can I have in my local development environment without clobbering, you know, the
Brendan:Yeah
Ryan:in process, right? You know, you solve that, it sounds like, by just doing everything in the cloud, right? Cloud spins up its own, you know, environment, everything is isolated, at least to the point where it goes to a PR, right? And then you have to merge it back in, right? Conductor, which I, I just came back to this week, Brendan, because I'm hitting this problem again, is like, all right, now that I've got a workflow for me that is working with, all right, ask a question, build the architecture. I have it save it as a MD file, uh, in the code base, right?'Cause sometimes I'm just planning, and sometimes I, I just need to remember for later, but I'm not actually gonna do anything with it, and sometimes I'm like, all right, I just need the plan for AI to follow, um, that, you know, that I've approved. What happens is I, I… once I start going through that implementation plan, right, I want it to verify, right? So I want it to run the test suite, and I want it to, you know, open up, you know, Playwright or Selenium or, you know, whatever, and actually click through the UI. And I quickly realize all of a sudden that I can only do one conversation at a time that way, right? Because otherwise the second one says, oh, the test, you know, the d- the test database is already in use. Like, and then it, it goes down this tangent of like, oh, let me try to spin up Docker. I'm like, no, no, no, no, no, just wait until the, the port is open or use a different port, right? Like, it's not that hard, right? But Conductor solves those problems in my mind, uh, uh, you know, for anyone… That's more for anyone listening than it was a question, but if… Go ahead. It sounds like you got something you wanna say
Ryan Castillo:I, I, we never used Conductor, but I think we built something similar and, and it was this whole idea behind work trees
Ryan:Right
Ryan Castillo:like that. You know, prior to this, I had never, ever used a Git work tree, you know? The--
Brendan:Same here. Yeah.
Ryan:Yeah. I, I had no idea what it was until a couple months ago
Ryan Castillo:e- even when my co-founder had like brought it up, I was like, "I don't, I don't ever wanna use that." And, um, we went down the rabbit hole of like pushing the envelope with it to the point where I think we were like generating screenshots and, and using those screenshots to test, um, like click, doing like click tests
Ryan:Mm-hmm.
Ryan Castillo:the screenshots, and it just got too cumbersome. then even, even with that, the pattern that I was noticing is that like a small feature would become an, an, an epic. You know? Like if you're living in the conversation and you're just having the conversation, then you're just like,"Oh, I need to add this other thing. I need to add this other thing. I need to add this other thing." And then nothing ever ships. And so, or it just becomes too big to ship. And so, we still use those same principles with this Cursor workflow that I'm describing. You know, now I'm not living in that conversation anymore. Now I, I'm, I'm living in, know, a proposal mode, or I'm living in review mode. And then when I'm in review mode, you know, then it says like, "Okay, I wanna test this out," and it says, "Okay, well, we'll use the work tree, work tree stuff that you had set up and all that stuff." So I think Conductor kind of fits into this workflow that I'm describing. I've just never personally used it. But, um,
Ryan:Yeah, I guess… Yeah, I think my point was more like I, I look at Conductor as a harness more so than, you know… And, and that harness is exactly what you just described, which is, you know, use work trees and then has some setup scripts to get around, you know, the, the shared database problem, right? But it just, it abstracts all that away and gives you a nice UI to try to track the conversation. But if you're not in the conversation, to your point, less value. You don't need, necessarily need Conductor, unless I'm missing something, uh, to do the, the end-to-end workflow. It just does give
Brendan:No
Ryan:UI to see where you're at.
Ryan Castillo:Hmm
Ryan:but yeah, interesting. Like I, I don't necessarily want to be in the conversation. You know, once I have a plan that I've sort of stamped, um, I find that it, it, Claude still gets caught asking questions even though I'd say, "Go build the thing," right? So it's like I'll say go, and I'll come back, you know, two hours later and it hasn't even started 'cause it was like, "What do you want me to do here?" I'm like, "You have the plan, just do it. Like we've already done this. Why are you asking? Like continue, please." And I've wasted, you know, two hours of wall time
Ryan Castillo:and, and that's, you know, a lot of people are big about, like, living in the conversations and using these tools to ship content for people and stuff like that. But a lot of these conversations just, like, really frustrate me. You know? Like,
Brendan:Yeah
Ryan Castillo:uh, I'll, I'll, I'll be talking to Claude about, like, a, um, "Hey," you know,"we just ran this service contract. I need to send, send the invoice. You know, I just want you to, like, double-check, like, what I wrote here. You know, just give me a, a thing." And then Claude will say stuff like, "Oh, this is where people always get it wrong." Or they'll, like, they'll introduce all these, like, random quips, and I'm like: Wait, do people really get this wrong? Do… Are people actually doing this? Like, how do you know this? And it's just like, Why are you just making up all this other fluff for me? I don't wanna see all that. I just wanna get the job done, ship the thing, and move on to the next thing. Um, so, um, just my personal philosophy. I mean, the AIs, like, help enhance the stuff that I'm trying to do, but I'm not trying to live in them all the time. I got, I got li- a lot more interesting things that I wanna do outside of all this
Ryan:To, to clarify, just'cause you just gave an example of using it to like, you know, spot check copy for an invoice it sounded like, right? Which I totally get. I wouldn't wanna live in that conversation either. I'm just looking for a quick gut check, right? But that is a very different, um, sort of use case than building a big feature, right? Um, so I'm cur- like do you, you don't like to live in either conversation or just like the conversations where you don't feel like you have anything to add?
Ryan Castillo:No, uh, I, just want to get them done as quickly as
Ryan:Right. Yeah, I mean, that makes sense
Ryan Castillo:yes. You know, um, it-- The, the living in the conversation and, and, and adding more and more features, that was me, like, maybe four or five months ago. Like, uh, you know, I just kept on getting stuck. I kept on just adding more stuff, and that's just, that's not useful.
Ryan:So what you're telling me is, you know, basic engineering principles still apply. You know, small
Ryan Castillo:100%.
Ryan:often, right? Okay, cool.
Ryan Castillo:100%, yeah. 100%,
Ryan:Yeah. No, I, I mean, I, I find myself in that trap often. Um, you know, both, you know, feature bloat, right, and getting too big to ship, uh, which is something I'm personally, you know, cognizant of and trying to work on. You know, at the same time, there's this… I think this is a human problem, not a, a machine problem, that it's like you want to-- You have this reaction where you, i- if the AI is gonna go off the rails, uh, no pun intended, um, you wanna catch it early and correct it early. But the truth is, I think that, you know, or at least I'm finding, is that throwing away the code is easier right at the end. It takes a lot less time. So just let it do what it's gonna do, and then correct it at the end. And if that means you throw away half the code, so be it, right?'Cause, like, you didn't write it. No blood, sweat, and tears there, right? Like, but that is just so anti how we think, I think, as people, it's hard to let go. So, you
Brendan:Yeah, and it's funny you say that though because just yesterday I was working on like a little side project that I started, I don't know, s- three, four months ago, something like that. And I actually decided to take Fable and say, "Hey, take a look at this. This feels like it's going on way too long. It should be done." And it looked at like, "Hey, you've already invested this much time in coding." And I went back to it, I'm like, "No, you did this. I, I don't really care. Let's just throw it away." And it's like, "Oh yeah, good point."
Ryan:just had this conversation with, with my business partner earlier
Brendan:So it even thinks that way.
Ryan:estimates and I'm like, half the time it says, "Oh, this is like a three to four-day feature." I'm
Brendan:Yes. Yeah.
Ryan:10 minutes, so I
Brendan:Right
Ryan:know where you got that." And then on the, on the flip side, it comes back and says, "Oh, this can be done in seven to eight days." I'm like, "You're full of crap," right? Like,
Brendan:Yeah
Ryan:you know, normally this would be a six-month to a year feature. You're not doing it in three days. Like I, AI is good, but you're not like that good, right? Um,
Brendan:Mm-hmm. Yeah
Ryan Castillo:and those kind of things are infuriating to me.
Ryan:Yeah.
Ryan Castillo:uh, we, we know like one of the… I know one of the CTOs that just started working with Amazon and, know, in the last six months and, and one of the things that she was doing is like she was looking at big and seeing like how AI has actually really impacted their productivity. You know, like what, what it has, has it actually done? You know, have people been able to ship more PRs and stuff like that? And, and the truth is yes, on some levels, like yes, more PRs are getting out there, but now it creates this, this other bottleneck where someone has to approve all these PRs and stuff like that. It… Yes, it's increased or it's decreased the time to PR, which is like a, a success metric for, for engineers at a lot of, at a lot of cor- corporations.
Ryan:Speaking of metrics you wanna throw away
Ryan Castillo:yeah. but, um A lot of these problems are still human problems,
Brendan:Yes
Ryan Castillo:And a lot of us are, have become like overly reliant on the AI to argue back with it. And so you have to really like be super critical of the things that are coming back, you know? Like you have to be ready to argue with it about the stuff like, "Hey, why'd you tell me that this is gonna take like 10 days when you can do it in five minutes?" And stuff like that, you know? Like,
Ryan:I don't even bother arguing with that. I just, "Just go. It'll be done." Right?
Ryan Castillo:go. Like,
Ryan:I think you gotta, you gotta pick your battles of, like, where y- you mentioned earlier, like, where can a human provide, like, the intuitive value that an AI is gonna miss, And that, that's where I'll focus my argument time, right? And that's usually at the bigger… Not always, but usually at the bigger architectural level of des- system design, right?'Cause, like, it-- AI is very good at building a feature in isolation, especially if it's greenfield. But if it's part of an existing application that needs to interact with the other parts of the app, it tends to miss those, at least in my experience. And that could just be a prompt problem, but, um, I feel like that's where I can add value, you know, in the conversation or at least in the planning phase. Um, yeah. You know, statement, not a question.
Ryan Castillo:No, and you know, it-- This comes in, back into the marketing side. By design, the future of all these AIs is they want to keep you in the conversation as long as possible. They do not want you to leave the platform. They-- Google wants you in AI mode, ChatGPT wants you in, in that chat, Claude wants you in its chat, and it, it just wants to continue to provide you value and provide you with that loop. And what is, like a lot of people are starting to notice is this is just exhausting. You know?
Brendan:Yeah
Ryan Castillo:back in the day when we were, when we were pairing all the time or pairing was a, a big thing in the, the, the software development side. Like, yeah, it's cool to have two people on a project and pair and, like, you have a second pair of eyes, but that is just an exhausting endeavor to, like, pair and look at code all the time. And now, now we are in, like, five pairing sessions at the same time while trying to juggle our other work and then, like, you know, trying to just push it more and more and more and like, "Hey, how can we make this thing do stuff overnight? How can I just burn more tokens? Oh, I need to get to my Fable limit," or whatever else. And what's the point of that? You know? Like, the
Brendan:Yeah
Ryan Castillo:the, the, the point should be in all of this is like,"Hey, how can I make the most use of this, this tool so I can go out and live my life the way I wanna live my life?" And, um, I think that's kind of lost on a lot of people right now.
Ryan:ironic that the companies, the AI companies actually wanna keep you in the loop considering how much it costs for them to keep you in the loop. You think they'd optimize for, you know, outcome and efficiency. Um, but not my problem to solve.
Brendan:Yeah. I mean, this has definitely been interesting. Um, I know we're getting close to time where we might wanna wrap up just, uh, to respect your time here, Ryan. But, um, is there anything else, like, around the, the coding side of things or even just, you know, Noatoa business stuff that you wanna talk about before we do start wrapping up?
Ryan Castillo:Hmm. trying to think. Uh, no, uh, you know, I think, um, I guess one thing I think, and I've already touched this, is there's gonna be, there's gonna be a bounce back from all this, right? You know, a lot of agencies, uh, especially, uh, in the marketing side, and this is just gonna come back just to like all of our working experiences side, is like everyone has become over-reliant and like so AI, know, optimize that I think the counter is gonna happen. You know, everyone's saying like,"Oh, SaaS is dead," that everyone's relaun- launching products. But now you're gonna see like a lot of these product, products that exist and they're like, "Hey, we actually need an experience engineer scale things, who can figure out like what platforms we need to push our data to, how to take, you know, some of this data that we're just pushing to SQLite or SQL or whatever else and push it to Redis and like that." You know, the, the human experience still matters in all this. And, you know, that's what I would encourage everyone to think about instead of like, "Hey, doom and gloom, AI's gonna take my job or AI's gonna kill my SaaS," is to think about like,"Hey, what have we actually done on the human side that the AIs can't do?" Like, let's give ourselves some credit and, you know, let's like double down on that and, um, you know, be hopeful. So
Brendan:That's a good message. Yeah. That's a good place to, to wrap up too. Um, where do you wanna point people to before we sign off?
Ryan Castillo:Sure. Uh, people can find us at knowatoa.com. That's know, K-N-O-W-A-T-O-A.com. Um, the, the landing page is still being updated, you know. Um, some of our tooling is still being updated. Uh, I'm start- gonna start, uh, being more and more present on Twitter. I know people call it X now. I still call it Twitter because I was
Brendan:Same here
Ryan Castillo:the, the, the fail cl- the fail whale and all that stuff. Um, but yeah, you know, I'm gonna start posting videos. I think I'm about to do like a, a 30-day, like, publishing content challenge, you know, just to kind of share like what we're doing the AI visibility company to kind of like promote our brand and stuff like that. So
Brendan:Awesome.
Ryan Castillo:there.
Brendan:Cool. Well, thanks for joining us. It's been great having you again
Ryan Castillo:Yeah. Thanks, guys.
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