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
John Nunemaker on AI Development
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
John Nunemaker returns to discuss how fast AI tooling has changed and argues tech leaders should adopt Claude Max (defaulting to Opus), set up agent-friendly local environments, and roll that out to their teams. He shares how his company built an internal “brain” using LibreChat as a self-hosted interface, enabling non-engineers to query company data and generate insights like weekly support summaries from various sources. The conversation also covers worktrees and Conductor for parallel, isolated dev environments.
01:22 Tech Leaders Must Adopt AI
02:31 Why Opus Changed Everything
04:55 Claude Max Plans and Limits
07:41 Brain Idea and Safety Boundaries
09:03 Building the Stack with MCP
10:18 Ansible Automation and Deploys
12:53 Support Data Summaries in LibreChat
15:42 Scheduling Reports to Slack
20:17 OpenCode vs Claude and Mobile Workflow
23:54 What MCP Connectors Enable
26:20 Accessing Brain via Tailscale
30:48 Team Size and PR Volume
32:32 Claude PR Review Workflow
35:25 Conductor Daily Setup
38:36 Worktrees and Isolation
50:47 Conductor Limits and Pain
51:36 App Changes and Scripts
54:45 Caddy for Local Domains
57:10 Jobs Outlook and Wrap
Notes
- John's post setting up conductor with Rails
Questions or comments, email us at railsbusinesspod@gmail.com
Today we got John Nunemaker back with us. Welcome, John. Good to have you.
John:Good to be here.
Brendan:So we, we were just talking pre-show. Um, you know, obviously a lot has changed in the AI world since the last time we had you on you know, I, we touched on AI too. You know, I remember we were kind of talking about like Vim and like some of the things that, you know, I was doing and you were doing stuff and, and Ryan, I don't even know how much you've been doing at that point, but. Wow. It's like things are so different at this point. I'm doing a lot. I know. John, you're doing a ton.
Ryan:Um, so.
Brendan:know, for, maybe I'll kind of start with like what I'm doing.'cause I also mentioned pre-show that like I'm doing stuff, but it seems like you're way past where, where I'm at. Even
John:Or in the wrong direction. Who knows?
Brendan:maybe, I don't know. That's a good point. you know, like for me, I'm trying to like, I. I'm trying to review code with it a lot, obviously write code with it. That's kinda like where it all seemed to start, but then I'm, I'm also trying to find like the gaps where, like how can I use AI to automate this or that other little thing That's not coding necessarily, but it's like tangentially related. Maybe it's the stories
John:Mm-hmm.
Brendan:So that's like a lot of different things, but. Maybe the, the place to jump off would be you had a tweet, I forget exactly how long ago, maybe like a month ago. Um, and it was something along the lines of, you know, if you're a, a dev manager or like a, a leader in tech, like you should be teaching your, you should be learning this stuff, setting it up on your own environment and then like teaching your team.
Ryan:Here.
Brendan:So maybe that's a good place to start.
Ryan:I had that one written down too. So your job as a tech leader now is to get Claude Code max, default to Opus, get your local dev environment set up agent friendly. Then do the same for everyone on your team.
Brendan:Mm-hmm.
Ryan:So like that, I mean, that's a bold statement. Why do you maybe share some of your, uh, your, your viewpoint on that?
John:Yeah, that's just what Claude told me to say, so, no, I'm just kidding. Uh, yeah, so I mean, I think where that came from, it was just one of those things that like popped in my head and I've been trying to just do a better job of like, something like I, I read things online and I'm like, oh, what a well-formed, like, smart opinion that that person had and Oh, but like most of the time it's like. Yeah, I agree with that. Like, why didn't I say that out loud? Like, why did they have to be the first one? And so like, it just popped in my head that day because I was like, I've been using Claude for like a long time. I'd been using sonnet and then like, I remember December 14th, it'll be an infamy for me. And I was like, I screw it. I'm just upgrading to Opus all the time. I'm not gonna use, you know, the lower models anymore. And as soon as I did it, I was like, this is the future. Like it was just like, I mean, within a day or two I was like. I have not, you know, generated this amount of good stuff in that amount of time period, you know, since like Rails. Um, and so I was like, okay, this is, this is definitely the way, and just, I started showing people on the, on the team. And you know, very quickly they started seeing the same thing. And that's when it was just like, you know, I don't know if I was in the shower or driving or what, but it just popped in my head and I was like, I just gotta get this out, you know? I did. And then, you know, you always get like the trolls, they're like, they're like, well, codex is better. And I'm like. I that yes, I, but like, if I just say it doesn't matter, which is better or which is worse, what I'm saying is like, your job right now is to figure out there's a new thing in town. It's to figure it out and then, and be on the forefront and understand when and where and all that kind of stuff to use it and then get everybody on your team to use it and go from there. And, and so we've been doing that at Box Out and it's. Like absolutely unreal. Uh, I mean, and obviously I'm very good with all the other stuff too. Um, but box outs were like the largest team that I'm, uh, you know, taking that approach with. And so yeah, it's been, it's been incredible. So.
Ryan:All right. So tell us a little bit about your, uh, company, AI in a Box.'cause it sort of feels like the, you know, that concept, uh, overlaps with this. I, I don't know if you call it a project or approach or however. You view it, right? But you know, what is it to you? What's the stack look like? How does it work? Like, you know, share what you're willing to share.'cause I think a lot of people are asking the same question.
John:I'll share anything. So, I mean, the biggest thing for me was like the, the, I mean, I think the, the biggest leap for us as a team, number one was, you know, Claude Max, like everybody got the, the a hundred dollars basically immediately after Christmas break.'cause all Christmas break I went crazy and I was like, this is amazing. This is the future. And so right when we came back I was like, everybody use it. I'm gonna get into the brain stuff too. And then,
Ryan:So hold on. So Claude Max is the hundred version one, right? And before that I, I, I upgraded to that about the same time it sounds like, you know, somewhere in that area. And before that I was like hitting the limit, like every, what, you know, 20 minutes it felt like, and I was so
John:yeah.
Ryan:That, you know, the a hundred dollars subscription wasn't gonna be enough. Right. And I have not hit the limit again since then. So, like anyone else listening, like, just do it. You, you can still get pretty far. And if you're a super user and you have to go to the, you know, the 200 plan, so be it. But like, don't fret for anyone else that's thinking the same thing that I was, I just wanted to throw that out there.
John:No, I totally agree. And I've actually like Wagered with a ton of people. I was like, look, get the a hundred dollars plan, get the $200 plan, use it. And if at the end of the month, uh, you don't feel like it was worth it, then I will pay for it and you can just not use it ever again. And no one's taken me up on that to date. So
Ryan:You are right.
John:it's a hundred. It's a absolutely amazing, like even. People who didn't wanna spend the money, they're like, this is crazy. I, I should have did this a long time ago. So I got the $200, I went straight to the, the biggest one because that was all about the deal, you know?
Ryan:Yeah.
John:well, it's four times for two times the price. Why would I not do that? But I, everybody that, you know, that's like working, uh, with me. I, I've just had 'em start with a hundred and I've just said like, your goal is now to try and max this out enough that it's annoying that you need the 200 and then switch to the 200 immediately. I don't care. Like, and then I've told everybody the same thing, uh, as well, like when you're at home. Like outside of work hours, if you would like to use your work Claude account to do personal things, knock yourself out. I don't care because, and I'm not trying to encourage people to like work like crazy or anything like that, but like, it's fun. I think we should all understand it. I think we should like, play with it. And I know if they do with it, I'm not, I'm not gonna force somebody, I'm not gonna say like, if you're not doing this on Saturday and Sunday, you're, you know that, that's definitely not what's happening. But if you mess around with it. And you figure something out on the weekend, that's always gonna come back and benefit work as well. So I've told everybody, I'm like, use it up, like the, the limits. They're, they're there either way. So, you know, uh, go to town. So that's, that's the first thing.
Ryan:The, the amount of code I get done now when I'm putting my kids to bed is, is sort of ridiculous, right? Like I can build like an entire feature in the hour. I'm sitting in my, my son's room, you know?
John:to entry is so low. It's unreal. It like you can just, uh, say things and like, before it would be like, like I have to wait 'cause I, I've only got, you know, 45 minutes until I gotta do something. So I don't have time to get anything going. So I'll just do something else, you know, like check my email or whatever. And now it's like, I mean, my inbox is just growing 'cause I'm like, no, if I, if I have to stop and do something by hand, that feels so. Brutal and painful. You know, like anything that I can't just like automate like that. So, um, but that's kind of what has led to the, the brain idea, which was basically I was like, you know, there's just so many things that like. I keep seeing open claw, bot, claw bot, you know, all that stuff. And I'm like, this is cool. But I'm like, I don't wanna just give away all my credit cards and stuff like that to ai. I'm not there. So other people, they can go to town, they can it, you know, burn their billion tokens and do all this crazy stuff. Like I'm okay with them doing that, but I'm like, I just, that's. Giving over your entire life and letting your AI like do all this stuff just seemed like a little too far. So I was like, well, what are some isolated ways I can do this? at the company level where it's valuable. So that's where it started and I was like, well, you know, the first thing I could think was like, what if I could just make it easier for people? You know, we run Mease, which is like a reporting thing. And so like, but what that means is, hey, I need some data. So they come to John, and John creates the report and then they have the data going forward. So they don't have to ask John every time, but I still am involved every single time and I'm like, how, how I want to get to more situations where I'm not involved. Like that's, that's where this is kind of all, you know, coming from. So I was like, how can I make it so that like. know, somebody on sales, somebody in marketing or growth, whatever, can just be like, you know, how many of this happened in the last week? You know? And I was like, well, Postgres, MCP, maybe that would get us close. So it was just like, how do I somewhere run a pre-configured Postgres MCP with a read-only replica so nobody can do any damage. has any passwords and anything that, again, damage related, that there's, there's no way really that any problems can kind of come from this. It's like, what's the easiest way to do that? So I did a lot of spelunking, you know, with, um, my best friend Claude, and we were like, at the end, you know, Lieber chat was one of the, the options that was, it's basically like ChatGPT or Claude's web interface. But it's open source and you can just, you know, it's a docker image. You can just pull it, run it, and do whatever you want. And so I was like, okay, well that looks pretty interesting. So now I'm at the point where like I just need to be able to like. Test it. And so that's why I was like, well, I'll just spin up a server and then I'll like run some stuff on it. So my first thought was like, I'm just gonna like rent, you know, whatever, a hener box that's 16 gigs of ram, something that's big enough to like do some things. which is only like 30 bucks a month. It's super cheap. And so, and then I was like, well, I'll just, I'll just install Claude and I'll let it rip. You know, I'll sign in, you know, do my stuff and I'll just have c Claude set up the whole server. But then I was like, well. But it's not repeatable at all. So then I kinda thought, well, maybe I should do it locally on my machine and do it in a, you know, like a repeatable way. So that's when I kind of to toyed with Ansible.'cause I had toyed with it before and it's, it's kind of for like a Ruby guy, it's like Capistrano for servers, you know, like you tell it to run some things and it'll just do it. And you can do the pull model, like puppet and stuff I think too. But for the most part it's just like, you run, I run a command and I can do things. I was like, okay, cool. So I was like. Uh, make a Ansible playbook to install Claude, like, it did it, and I was like, like run it against this server and it, and it did it. And I was like, okay, like lock down SSH and like, so it locked down SSH and it like threw in a couple things that I wasn't even thinking about, you know? And then I was like, uh, and I, I don't even remember all the things, but I was like, well, now I want like. code just 'cause I thought, well, if I have open code, I can run some random junk on there and maybe that can be my, like, you know, runs things when I'm away from my computer situation.
Ryan:So what? What's open code?
John:Uh, it's just Claude, it's like the open version of Claude
Ryan:Okay.
John:So, and you can run with different models and all that kinda stuff. Um, like a lot of people seem to really like it. Um, I've had some issues when I, when I used it where I've just not been a huge fan. but it's cool. And they have OpenCode Zen, which allow, you know, allows you to run. Like right now they're doing free trials of like. Kimi and MiniMax and like all these, you know, open source models, um, which is kind of neat as well. Um, but I was just like, oh, let's just make this kind of like a repeatable thing so I can script deploy. And then pretty soon I was like. I was script deploying and I had to put in a password every time and it was getting annoying 'cause I'm like typing this password or pasting it over and over. And I was like, can I just let, okay, just put the password in a file, like M type stuff, you know? And then when I do this, like just cat the file into input so that I don't have to do this by myself every time. And it was like, cool. Did it, so now it can deploy. And I was like, cool, now you're deploying. And it was like, okay. And I was like, also, now you're SSHing and so that you can verify the things you did.'cause like I don't wanna be involved. And so then it just started, it just started turning and I'd be like, do this, install another thing. Do this. This didn't work. And eventually I got to where, you know, I have, um, I, however, I don't know how you pronounce it, but Lever Chat set up and. It's got the Postgres, MCP, and I had the stripe, MCP, um, and I could say like, how many, you know, users do we have or how many graphics were created? And it was giving me right answers and I was like, okay, this is kind of cool. And so then, you know, from there I was like, well, what would be the most that, like, that's, that works, but what would be something that's actually like really usable? And I feel like we just, I have no data on support.
Ryan:Mm-hmm.
John:for box out, I'm just, I'm like three levels away from it. I'm only brought in when there's like a severe technical issue of some sort that has to be debugged. And so I was like, I don't know, generally like what happens, what's the most common? Any of that kind of stuff. And I. I've kind of asked for it a few times and it's always just like a gut, well, maybe it's this kind of stuff. And so I was like, you know what, we have all these in, you know, crisp. And so I was like, I'm just going to, I already have API stuff set up to like allow, allow our admin to like, uh, area in our app to like view the crisp support things and do some extra stuff that Chris, Chris didn't do. So I was like, what if I just slrp 'em all into a database? got the Postgres MCP already. And so I was like, what if I just slipped 'em all in the database and then I could do things? And then I was like, what if I vectorized them and then I could do semantic searches on 'em? And then I, I was talking with Claude, and Claude was like, you don't need the vector stuff. Like just chill out and just, just get the them in the database. It's fine. So I got 'em in the database and then I was like. Build an API endpoint that allows me to like, do these few things, like search by email, search by text, um, you know, list the most recent or for a time range or stuff like that. And just to get all of the, the conversations and messages and stuff like that. So I built that API, and then I was like, I created a new directory. This is all in, like, I mean, half an hour or an hour. I was like, okay, now create a new, you know, I created a new director. I was like, create an MCP to talk to this api. so I'm just linking directories back and forth. Uh, should have just did, honestly, at this point, I just want a big mono repo for everything that I work on, but, that doesn't really work, uh, across companies and stuff, but it did it. And so it got the MCP and I was like, cool. Now I went to the Ansible, you know, brain repo, and I was like, put the MCP in Libra Chat. And it did It and it deployed it. I was like, sweet, this is really cool. So then I was like, all right. I went to Lever Chat and I was like, gimme a summary of the last week of support requests. And it was like, oh, here's the breakdown. Like, you know, 25% are account related, 25% are technical. You know, 15% are like having problems with like social media accounts and connecting and stuff, because that is. Horrible generally. Um, and you know, then it was like a, yeah, and then a few other like, uh, things and I was like, that's cool. And I was like, what's the tone of our support? Person, like, how would you classify their responses? And it was like, warm and helpful. And I was like, that's what you want to hear, right? Like, that's, you want to hear that. You don't wanna hear. Like, they're like, if, if we were gonna make suggestions for improvements, maybe a little brief. And I'm like, okay. Like that's, that's fine. I can live with that. You know, like, So I, I was like, this is really cool to be able to like, so now. I mean, I can, it's just like, what, what do I want? So now I know the next thing I want, which I haven't got into because I've been stuck on a couple of other dumb things. But, um, I wanna do like automatic stuff, like just run this every day or week and, you know, slack it. Um, so I saw like the chat, SDK from whoever made that Vercel or somebody, um, where you can like very easily run this thing and then. Posted to a bajillion different slack or, you know, discord or like anything, it's like a adapter to all of the chat things.
Ryan:Yep.
John:And I was like, okay. So like, that seems like the a, a sensible, like next thing is basically like on every week, on like Sunday, run a report that's like, here's the most common support requests over the last with like, you know, a couple of examples of it. Then plop it in Slack, so every, just automatically it's there, you know?
Ryan:Let, let me,
John:like that.
Ryan:let me interrupt you before I lose too much of a, a train of thought here. Right. But is like, so for the scheduling thing, are you literally like typing that into a LibreChat and saying, you know, AI go schedule this, or are you like, is this like a cron job? Like, do you,
John:I think, yeah, what I'm gonna do is cron it. I
Ryan:okay.
John:I think what I'm gonna do is I'll, I have Claude and open code both on there and they, you know, they have dash P and stuff like that where you can just like pass things in and then get output. Um, so I think that's what I'm gonna do is set up like a cron, just a script that runs a couple of Claude things. On top of and I'll, so I'll just install that same MCP that's in Libra, chat into Claude, and then have, tell Claude the same questions and to summarize it and do all those things. And then, um, you know, the script at the end will take the output and just shove it
Ryan:Got it. Yeah.
John:uh, slack channel.
Ryan:That makes sense. So sometimes it's easy, it's hard to see where the line is between like, you know, uh, you know, a chat, I'll, I'll just say chat generically to mean with an AI and like automation, right? Like, I think what I've come to think lately and I, I feel like things are changing so fast. I, you know, maybe change my mind tomorrow, but it's like, all right. Those are actually two different things, right? Automation and ai, right? AI can do the work in, you know, using sort of natural language, but automation needs a trigger or some sort of schedule or something to actually kick off the AI to do the thing, right? Do you, do you think of it the same way,
John:Yeah, I mean, I think it has to have something, or you have to have some system that can take natural language and then turn it into that.
Ryan:right?
John:Um, like what I, what I did is I installed like, oh, so open code. The only reason I installed it is because they have open code web, is like, you can just run a command and then it runs a chat interface that's just like, you know. Cloud code or, or stuff like that. So you can like have, you know, you can go into what directories and then you can like tell open code to run things against different models and it'll literally just make files, you know, do get things like whatever. All that kinda stuff. And I was like, again, I wanted that to exist 'cause I wanted to be able to control the server from chat I needed to. So I feel like that's the kind of thing that like you could pretty easily be like, you know, I could just clone brain on that server and then. Tell it to like, do something and then tell it to like also, you know, to like put it in the get repo also and commit it so that it, it does both. Like if I'm on my phone, I could just be like, Hey, set up a new trigger, you know, a new crown every day that does this. And open code would do that. It would just be like, cool, uh, you want this crown and this file and this is what it's gonna run. Do you want me to make a script and set dash, eat it? And all that kind of stuff.
Ryan:Yeah,
John:but yeah,
Ryan:you still use.
John:how I do it.
Ryan:Claude or OpenCode to create the thing, but it is literally just creating the script and cron job for you. Yeah.
John:Yeah. And then you get the, the benefit of like, you know, at the end of it, and again, there's a ton of people that are starting to build these automation things. Like, I feel like, uh, I've seen like a bunch of 'em in the, in the past three days. Um, but none of 'em have been, I haven't seen anybody that's hyperfocused on like company, like solving in internal company things. I guess
Ryan:there's a bunch of the big ones that are like, you know, AI orchestration, right? Uh, N eight N and
John:Yeah.
Ryan:you know that that's type, but Right. I think. I think what you just said is a good reminder that it doesn't have to be that complicated. Right. It's a, there's a cron process and there's a script and that could be all you need, right? Like to start. Right. Um, all right, so you've been bouncing back between open code and quad code in this conversation. Right. And when you say Claude Code, I just wanna make sure I understand. It's like you're, you know, you're using your subscription like you're paying for something there, right?
John:Yep.
Ryan:Yeah. Okay. Um, and then open code is sort of the open source version of that. So you, it doesn't necessarily have a subscription, but it can still hook into models. Right. So you'd have to pay for that.
John:Yes. You
Ryan:Yeah.
John:the
Ryan:Right.
John:for, or the models? Not for the um, not for the harness. Yeah. The harness is free, which is just like the, the CLI or the web interface that it runs. The only reason I'm not using open code in my daily life at all, the only reason I use open code in on this one project is because it has a web version, my dream was I want everything to work from my phone anywhere. Whenever I Dang. Well, please. And the only way I could figure it out was I, I
Ryan:Yeah.
John:now, was that without vibing a whole bunch of stuff, so
Ryan:So Claude, that's, that's a frustrating thing for me right now with the Claude Mobile app, which is like, I can do it, but then I can't pull it open on my desktop and continue like, so I, I have to like start and finish.
Brendan:I, I just saw something
Ryan:Okay.
Brendan:like yesterday that they, if you're on the max plan, actually I think you get,
John:The remote.
Brendan:it. Yeah. It's like the remote one.
Ryan:Alright, well there we go. I'm
John:ha.
Ryan:later today.
John:It's, well, I think they're, they count any max, so 102 hundred, I think both Count. I, it's
Ryan:Well,
John:that, those are both the max plans. Just one, one super and one's regular,
Ryan:well, maybe I'm missing something 'cause I, I have the max hundred plan and I can do stuff on my phone, but I cannot figure out like how to get that. On my desktop may, I guess,
John:So,
Ryan:well, let me rephrase Now that I'm, I'm saying it out loud, right? I can probably pull it open in the, the web view and see it, but I can't then access it from, you know, I, I use the VS code extension in my, you know, vs code editor to use Claude, right? So maybe I, that's a disconnect. I don't know.
John:Well, you can tell it, uh, from your phone or from the desktop. You can tell it to commit to GitHub and it'll make, do that and make PRS and stuff, and then you can clone it and so you're all good. So that, that's
Ryan:Well,
John:before, but
Ryan:I do that, but you lose the chat history, right? Like I wanted to be able to just like pick up where I left off.
John:You
Ryan:And sometimes that's useful and sometimes it's not, right? Like sometimes it's better to just nuke the context window and start from where you're at. But, um,
John:Yeah.
Ryan:I'm working on a very specific problem that is like, it took a while to build the context and it's like, I don't really wanna start over. So I guess I'm gonna sit here and type up my phone now.
John:Well, I think what you can do for those scenarios, I've started like realizing this, that like, uh, these things are not. The way that, like, I, I didn't realize this before, but like the way uh, Claude works when it's going back and forth it literally sends the whole chat every time. That's why we have compaction and stuff. Like they are not intelligent. There's no memory on the other side. They're sending the whole thing every time. So like, that's why I've started the last like. Four or five days, I start over way more often. Um, I think it just, you actually end up with better results because, you've went through the other side of it, like your head has, has distilled the problem down and so then it, it can be a lot more succinct to redescribe the issue. and then sometimes it doesn't have the baggage that the previous conversation had.
Ryan:Yeah, so I've noticed that as well. And now I am, as a result, I open more tabs and have separate conversations. Typically in parallel. And I wanna, I wanted to ask you about sort of parallel work in a second, but to wrap up sort of your, your Brain AI company in a box I thing the Postgres MCP server. Is sort of the MCP to talk to your read replica database, right? Like it's the connection to that. And
John:Yeah.
Ryan:Stripe, MCP is the connection to Stripe data on Stripe, I assume, right? That's how is that the right framework of thinking about this thing?
John:Yeah, and you can turn all all of them on and talk to all of them. So I could be like, here's an email address. Tell me everything you can about it from the database, stripe and support. It would gimme a summary on them, like it, and it would pull all the data and then, you know, it would use the LLM to like analyze it and then give you a report back, which is really cool again.'cause like what kicked it off was like, one of our salespeople had a list of attendees to a conference, um, that they, that they had went to and they were going through and trying to figure out who was a box out customer and who wasn't. And so who should they follow up with?
Ryan:Yeah.
John:were like, 'cause we don't just throw a thing in and send a mass email that's not. The way we operate. I everyone hates that, so why would we do it? So instead we do personal stuff. And so he spent three hours of one day going through like a hundred emails and he was like, this is really slow. And he told, you know, um, Robbie, who's like, um, co-owner and whatever, and, and so I was like, oh no, we can totally solve this. So in like 10 minutes I made a script that would just hydrate. An email with, have they ever been a subscriber? Are they currently a subscriber? You know, never been a subscriber. That kind of stuff. And then like if they were, here's some other stuff. And so it was just like, take a spreadsheet, hydrate it with a thing. And I was like, this should be ai. This should not be, I mean, I was using ai, but he should be able to do this on his own without my assistance. That's kind of where the whole brain idea kicked off is like, let's solve that one. And
Ryan:It feels like
John:thing that
Ryan:you're almost there if you're not already. Right.
John:Yeah. Yeah. Like, and that was.
Ryan:Described.
John:Few hours of work, maybe over the course of like, you know, a couple days. Um,
Ryan:Think about how much time.
John:that is the hope. We're really big on automation and like trying to not add humans to do things when we can make, you know, we've always been this way from the beginning. Um, so we're very small compared to like, competitors in the same size and stuff like that. Um, but even more so now, so.
Brendan:Uh, just another question on like the interface. You mentioned Slack before, like the, kinda like a digest it sounded like, but how are the people that aren't you interfacing with this, this brain?
John:Yeah, so right now I have tail scale on it so that, you know, it's only accessible via tail scale. So what I've started doing is I just, have been inviting each of the people that can use it and that to the tail scale, uh, tail net or whatever they call it. And then they just download the Mac app, and then they go to like a, I do tail scale serve, you know, for Libra Chat. And so they get like a URL that they can go to. So basically it's like bookmark A URL. And just have tail scale running and you're, and you can access it and you're fine. Um, and so that, I think that's, that's the way right now. Does that, does that answer the question? Was I getting the right?
Brendan:I'm not familiar
John:Um.
Brendan:tail scale, but I kind of get the gist of what you're
John:It is, it tail scale's. Like a company VP is sick. It's so cool. Like I, I have not used it until this project, but I was like, everyone keeps talking about tail scale for this purpose. So I was like, I'm just gonna figure it out. Uh, AKA Claude. And so Claude was like, yeah, just download it, do this. And then I, I just take screenshots of the UI and I'd paste him into Claude and I'd be like, what do I do next?
Brendan:Yep.
John:and like, you know, if you've been doing this for 20 years, like I have, like you have a sniff test and you know what's legit and what's not, and tail scale is, is really sick. You end up with a little menu bar item, a Mac app, iOS app, and then it just creates this like network that anyone who's running that and correctly off can, can load infor, you know, information
Brendan:Yep.
John:that is running on that. And anyone who isn't, can't. So it's like a VPN, you know, like if you're at any big company, it's like a VPN. You just add a person, tell 'em, you know, set the permissions of what they can access and what they can't, and just install Mac app and log in with an email and that's it.
Brendan:Right. Is the interface kinda like a web interface though, and then it's
John:Yep.
Brendan:basically tunneled through this VPN basically.
John:Yes. Yeah. So Libra Chat is a web interface. Yep.
Brendan:Okay.
John:tail scale. Also, both of 'em have web interfaces. Yeah.
Brendan:That
John:So tail scale.com is you access outside of the tail scale network. Like that's just like go to a thing, like any SaaS and sign in and do your thing. Um, Libra Chat is running via tail scale serve, which basically specifically creates like a subdomain for your tail net. So it's like Libra chat tail, whatever kind of a thing. Um, and it's kinda like eng gr or, uh, you know, any of those kind of t tunneling things. Yeah, yeah. That's the best way. The mental model is eng gr uh, it's like eng gr for your company, of. Uh, you can do Eng Rock. I could do all the same things probably, but tail scale seems like the cool kid on the block. So I was like, I'll try 'em out. And they're as advertised. It's pretty good. You get a, I don't know, a hundred some devices for free. I mean, you
Ryan:So why do you need that? Like, couldn't you just have like a, is it to avoid having to run your own server to off and then show the chat interface? Like what's the benefit?
John:so Libra Chat actually even also has auth as well. Mostly I was just like, this has literally, this box is going to have literally everything related to the company on it. I would like it to be as locked down as possible,
Ryan:so it was more of a security concern than a
John:yes.
Ryan:challenge that you had to overcome.
John:I mean, I don't want somebody trying to hack, you know, email addresses on Libra Chat or, you know, I don't know, like any number of attempts to, to get into it. And so I was like, just completely lock it one extra layer.'cause all the stuff is gonna be centralized there. So that creates, you know, a little more of an, uh, attack vector if anybody were going to. So that, that was my thought. But,
Ryan:You know. That's fair. I was just, I was like, all right, well, why do you actually. You know, outside of security, why do you actually need it? You could just have a, a webpage somewhere that has a password, you know, authentication system and be like, all right, but yeah, that, that makes sense. You're putting pretty much your, your whole business in that. Let, let's make sure it's secure.
John:All the repos. It has read access to the database. It has, you know, a bunch of API tokens for Claude, for, uh, codex, for, open Code Zen. Like, there's just, there's a ton of stuff in one spot. So I was like, you know what? Let's just lock it down. Even SSH can be locked down to only if you're on the tail scale. So like. That's kind of cool too. It doesn't even matter if you have the right keys if you're not on the tail scale network. Also, you can't get in, like you can really kind of lock it down, which is kind of neat. So,
Ryan:Yeah, that's cool.
Brendan:Well, I, I'd like to shift gears a little bit to, you know, back to the, the actual day to day like coding with Claude and like
John:sure.
Brendan:doing. Like how, how big is
John:So.
Brendan:I forget if we talked about that
John:So, uh, coders, let's say five or so, so there's probably like five of us that are actually like code and then, you know, the entire company is like, you know, including contractors that are borderline full-time. Basically we're probably 15, 20 people, something like that. So.
Brendan:Yeah, so, so quite a few people. just like the, paint the picture here for me. So if you guys are all pushing Claude to the max no, no pun intended, but you're pushing your max plan to the max,
John:Mm-hmm.
Brendan:and. What does that mean exactly? Is that just like tons of prs and you guys have all gotta be managing these with either Claude reviews or your own personal reviews or like what does that look like?
John:there's new problems for sure. Um, so like before the problem was like getting the code built and all, you know, testing it and all the things and the whole process and then getting it merged and like I would say our, we have very minimal process. Like our goal is like give people a lot of autonomy to do things like. kind of knows, like if you're gonna migrate the database, like check with John, you know, like, um, because I'm the one that probably cares the most about that and has the most experience with it. So from that standpoint, you know, there's bits of it. And then, you know, if you're going into like the producer screen, okay, talk to Steve. If you're going, you know, to like various sections of the app. There's just like, um, don't know, heard, heard. You know, everybody just kind of knows this is the spot. Like if it's, if it's a template related thing, we talk to Philip. So it's just like everybody kind of knows that already. Now everybody's just slinging stuff like crazy. Like before, like using Philip as an example. He was more, we brought him on specifically for like React and for HT ML and things like that. And now, I mean, he's consistently making PRS to the Rails app that are. Good enough that I'm like, yep, let's merge it. It's, it's good to go. And maybe, maybe like a, like the first couple that he did, without Claude, I was like, Hmm, I see what you're doing here, but you've got like, you know, camel case, we're underscore here. Like this isn't gonna work. You know, like, outta my code. No. And um. But like now, I mean, with Claude, it was like the first one he did, it was really close. And I was like, well, I can see some resiliency issues. You know, there's a loop and inside the loop it's doing an HTTP request. And I'm like, what if one of 'em fails? You know, like, um, what, how we, we'll end up in a partial state. So I was like, at Claude, like, what would you do for this scenario? Like, mostly as like, 'cause I thought, well, if he sees it, so we have the GitHub integrations for Claude as well, so. That we can say at Claude on any, any PR and anything like that, and Claude will it and issue comments or actually commit the code and make the change or any of that kind of stuff. We set that up as well. It's, it's great. And so, um, we, I did at Claude, like, you know. Uh, I see some resiliency problems there, like in review the rest. And so then, you know, and so Philip saw that because it comes in as a comment,
Brendan:Yep.
John:he like makes some changes based on that. And he is like, you know, uh, what do you think now? And I was like, at Claude review the changes. And he was like, oh, I see what you've done there. Like, and so, so now he puts it in a spin cycle before it even gets to me. You know what I mean? Like, so he'll do the PR. He'll do the review locally in conductor. He'll find a couple things, review again in conductor, find a couple things, review again until there's nothing. And then when there's nothing, he creates the PR up there. And then he says, ATLA review this as a staff engineer with specific skillset of blah, blah, blah, blah, blah. And then, and that's just another thing that I've. I useful. And so I've started, you know, broadcasting to everybody. so he'll put that on and he'll go through all the fixes for those. And then once he gets through all that, he's like, what do you guys think? And then it's like, yeah, stamp of approval most of the time. Like it, generally it, there's not there. You know, once it's went through the spin cycle locally and also remotely and each of those spin cycles have no previous context, they're all like new context reviews. So they're not like, you're in this thread and you've got it done a whole bunch of things and now you're like, Hey, review yourself. It's like, no, this is a new thing coming in that has no emotional attachment to this code that was written. And it's just like,
Brendan:Right.
John:figure this out. You know? So that's, um, that's like a good example of like a typical work workflow that's helped a lot because. At first I was like, oh my gosh, what are we gonna do? Like everybody's pumping out, you know, several PRS a day and I'm now a blocker and I don't wanna be a blocker. So now clog can kind of help iterate on that a little bit. And then for me, you know, like my day to day is like, I mean. I, again, today I'm having some issues with conductors, so I'll, I will, I'll stick to the positives before the last, like four days, five days, it's kind of scrambled my brain the last four or five days with some issues. But conductor is awesome 'cause you get this whole left side. I don't, I don't open editors anymore unless I just wanna see the whole file with like. A particular syntax highlighting scheme. That's literally the only reason why. and so it's on the left side. It's like all the workspaces that I have across all my repos. And then on the right upper corner is like your GitHub flow. So it's like, here's all the GI diffs and things like that. If you wanna click on 'em, side by side views, et cetera. And then here's all the like CI checks and stuff like that. And then here's, you know, like some individual to-dos if you want to add extra things for like, once you finish the code part of it, like I need to go create a Facebook, you know, social app for threads or things like that. So you can have extra things like that too. And then it's got the PR linked at the top and I can click that and then it's got the whole flow of like, create the pr. And then once it's made, run the ci, fix the errors, you know, that whole get up flow. And then the bottom is like a terminal. Um, and then the middle is just as many. Claude tabs as you want for that individual workspace, and of which you can have lots of them. And so that's, that's been my flow for probably two months now. It's going really fast, but, or beginning of January, um, in December, I was still using Claude in ghostie with and tabs and all kinds of stuff. because Claude, uh. Cloud code in the terminal or with the, like the actual Claude Code. SDK is a different beast than in an, in an any editor. I used it in Cursor and which is basically vs code, et cetera, and it's just, it's totally different in the end, end result that you get in my experience. Maybe they've caught up. In the, in that, in the last little bit. I don't know. but yeah, so that's day to day me and I got right away. What I did is I was like, I'm a salesperson, so my job is I figured out, you know, Claude Code and Opus and I went to everybody and I was like, get the a hundred dollars plan, get the a hundred dollars plan when you max it out. Get the $200 plan. If you need my credit card, let me know. Like all that kind of stuff. And I was like, here's how I do it. Show 'em like three or four things, get 'em going. Then it was like two weeks later, I was like, conductor's better. Here's the thing. And, and what's cool is because the first time there was some uptake, you know, that like it was a little slow 'cause it was like. Because everyone's like, whoa, I'm not the editor. This feels weird. Everything's different now. I'm in terminal. Where, you know, but like now with Conductor, it's like everybody like went through that first boost and so they're like, uh, yeah, I would like the next hit of dopamine. Please. Like, and so they, when they switched to Conductor, they're like, everyone's all in. Um. And again, I went through first, so I figured out like, how do we make sure that like the entire app runs on a different port, you know, like, uh, and that it can run on eight ports at the same time, like eight d different instances can all be spun up. And then that, you know, you can run the test at the same time because all the databases are partitioned apart from each other. And you can run the main app and you can, you know, seed and like everything is just like completely isolated so that every time you spin up a work instance, it's a brand new, fresh thing, you
Ryan:All right. Talk, talk about that a little bit more. How, how does that manifest? Because one of the things I wanted to ask you about, I've seen you talk about it a few times on Twitter, is just like the parallel agent, or even per, it's not even parallel agent, maybe it is, but it's like parallel processes of the same thing.
Brendan:Yeah, be. Before you answer John, I just want to add a little bit of what I've learned and see if it's similar to what you're gonna answer, but I've been using the Claude desktop app a lot, one thing I've learned there is that it kind of defaults to a work tree,
John:Mm-hmm.
Brendan:guess I'm curious if that's what you're using or not.
John:Yes. Under the hood it's just work trees. Yep. It's turtles all the way down. So, so, yeah. So under the hood, uh, conductor makes a work tree. It gives each work tree a name. Like a city name, like Shanghai or whatever. And so that's the unique name for it. And then, um, it will rename the branch once you type the first prompt in, you know, so if I say like, build this new feature, it says, okay, this new feature is now the branch name. So it'll automatically, it doesn't ask you when you create it. It's like, do your prompt, we'll name it for you later, which is really nice. It's like. Tiny little non friction thing. And then, uh, you get the work tree. And the first thing you do once that work tree gets created is it if you have a setup script, it will run the setup script. And so the setup script is like anything that's get ignored in the main repo, like link it or copy it, whatever your preference is to the workspace, because you're gonna need your dot m file, you're gonna need your key files for rails, credentials, you're gonna need act, you know, you might need active storage, local files. You might need like. things that are not in the repo.'cause the work tree is only gonna get the repo. And then, you know, after that it's like, okay, now you're set up and you're good. You can do like, bundle, exact rake you can do run, spin up a server, all that kind of stuff. Um, and one of the things that it does in there is like, it will, um, I create a database for each, um, you know, you just add like a suffix. Or whatever you wanna call it in your YAML file. if that suffix is there, then you know, it'll create new databases for you. So the core database might be box out sports product, you know, development. Um, but if you, you know, use it with conductor, it will work fine without conductor. But if there's a conductor workspace name, then it's like box out development under Shanghai. Um, and then, you know, when you archive. Like when you merge, it'll archive the workspace, or if you archive it because you decide not to do it, and then that will drop all those things and reverse them. So, um, in the words of Missy Elliott, you know, put your, no, I'm just kidding. Um, but there's, uh, like I, it just popped in my head. It's been a thing with my wife and I lately. We're like song words, key, key off songs, and then we start rapping. Um. So the setup and tear down are the two things you need. And then once you have those things working, you have no contention, you have isolation, all that, that's cool. Then you can run, you wanna run the server so you can run the server. And then what they do is they automatically inject a conductor underscore port environment variable. And so then what I do is I just say, cool, I'm gonna use proc file. foreman 'cause that's what everything I deploy with uses Heroku Railway, et cetera. And so I just spin up a and a proc file. Um, I just, uh, the, whatever the script is that starts it. So bin dev bin start script server or whatever you want to make your thing. It says, Hey, if conductor port is present, set port to that. Otherwise. Uh, just use port, you know, 3000 or 8,000 or 5,000, whatever you wanna do. And so then from there, now boom, you've got multiple instances running and everything will break because everything is based on the original one. After that you have to, you know, get that port into the cookies, uh, so that your sessions are separate. And then, you know, you gotta get the port into like. Action U, you know, URL, generation and Action Mailer, URL, generation and like all these other things. But like, if you look on my site, there's, you know, uh, john imaker.com. There's a post about it and it, if you just feed that into Conductor or Claude, it'll figure most of that stuff out. Um, and then you run into really weird browser issues because everything runs on local host, and so the, all the cookies get stored in the same place and you hit the max cookie size and then everything barfs. So then you have to start deleting your cookies every so often. There's everything is like new problems, but that's the general flow for me, and that's what I've like, helped everybody else on the team get going, and then I do things halfway and just get it working and then they make it better. You know, Garrett comes in and. Adds like four new scripts that are way more resilient and better than what I, you know, hacked together the first time. Um, but that's, that's our flow.
Brendan:Yeah, that's cool. The conductor, I'm not real familiar with. I, I've seen you mention it a few times on Twitter, but I hadn't, hadn't looked into it. It wasn't familiar with it, so that's. Like, is that a paid tool or is that open source?
John:First hits free on it. Um, it's basically like, um, I don't know what's the best way to describe it. Like, I, they're a funded company, so I, I know, I think I have a pretty good idea of like where they're going. Um, but right now that Mac app is just free. Um, and so, and they, you know, ship, ship nightly on it and stuff. It's, when it works, it's, it's as big of an unlock as AI is. in general. So it's like taking, you know, the, the leap from writing your own code having AI write your code with Opus. And that actually is like, good, that leap. It's that, that leap on top of it when it works and when it doesn't work, you just feel like an angry old man. So.
Ryan:And is that, is that leap in the ability to have multiple, you know, isolated environments running Right. Because like, I mean that, that otherwise it's just sort of one dimensional, right? You're just sort of being able to stack and stack confidently that you're not gonna leak context or leak database or leak, whatever, and.
Brendan:Well, I mean, one, one thing that I do right now without conductor. Is because of the work trees alone. Like I can have Claude Code, at least go write the code, but it's not gonna do all of the things that John said with environments. So that, that's a bigger step. That's pretty awesome sounding.
John:Yeah. And it, it is, and it ports to any work tree solution. So you can use it with work trunk.dev. You can use it with like anybody's homegrown bespoke work tree thing. Like you can automatically just like, what I could do is just say like, don't even rely on conductor Jason as a config file to tell conductor what hooks and scripts to run. Like, just put it in, like get hooks of when a new work tree is created. Run this script, you know? And if conductor Port exists, great. If it doesn't. Then automatically, you know, ha do a consistent hash of the workspace name to, you know, like a port or a port range is what we do. So we do like a, a 10 port range that way we have room for like v dev and random other little things you might need to run and stuff like that. but like, I, I, I'm saying all this and I, I feel like sometimes people are like, this is magic. You, you must be way ahead of me. figured this out like a couple weeks ago. You know what I, I mean, like, I, I'm not, like, I'm not a, we're all. Way behind on everything right now. So, um, there's not, like, it's not magic. You literally just like download conductor and like try the, like, it'll immediately make, add a repo, it'll make a new workspace for you, and you'll be like, that didn't feel good. Why'd you do that for me? And then you're, you realize like, oh, that's actually what I wanted. I, I did kind of like that. And then you start, you know, getting it working and once you can spin up multiple of them, the great thing about it. To answer that question, um, looking down, because you're at the bottom of my screen, is, um, the, the great thing is like you end up with, oh, I just got a new honey badger error while I'm, I'm working on these like two things. Get a new honey badger error, export is marked down, new workspace, paste it in, go back to the other things, you know, oh, got another one. Paste it in. Go back to the next thing. I got a bug report from support request. Copy and paste the chat history, you know, put it in there again, I'm reading it. I'm not just like prompt injecting myself, but, um. Like, you can just quick scan it and look at it and be like, okay, make a new workspace, work on this while I'm doing other things. Um, and so I can stay focused on the things I'm doing, but things are getting done in the background. And then at some point you go back and you click on it and you say, oh, okay, I see what's going on. Um, and they give you like nice notifications like Mac os it's, it is Mac macOS only right now. And you, you know, you get a notification, you click it, you go right to the thing. It's, it's a really great experience. Um, there's just like a couple of bugs for like, once you go crazy with it and you start having like, I mean, I, I would have, I right now will have, have 30, you know, one workspaces, but like if I had across like 10 or 13 repos, if I, if performance would allow me. And I wasn't hitting all these other issues, I would have 50 or a hundred in there. Like I would, I would completely go nuts. And if it worked on my phone, I would have 200. You know, like it's just, I'm limited right now by how good it can perform. So that's the thing that's kind of holding me back on it. But I
Ryan:So today's the first time I've heard this notion of work tree, either one of it sounds like both of you have worked with it. Can you explain to me what it is?
John:I mean. Imagine just like, uh, gi you just say GI work tree, whatever. And what it does is it literally says, cool, I've got your whole repo cloned, so I'm just gonna make a new branch over here in this other directory. All right, see you later. Uh, that's it. Like, it, it just, it literally just like, so like conductor uses like home j new maker, you know, conductor. And then inside of there would be like the repo like box out sports. And then inside of that would be like Shanghai.
Ryan:Okay.
John:you know, Baghdad or Granger or whatever the city is. They, they pick cities and then sometimes you get, they, some of the cities are rare. I know you, you want to know this, but, um, I've collected one of the rare cities and so I feel pretty special about that. Um, and then they have a whole map of where you go in the world. Lots of cr, like really good
Ryan:That's funny.
John:a work tree is, you could, you don't have to use a work tree. A lot of people will literally just get clone again in another directory. I mean, that's the easiest way to think about it. It's just another place where you can commit totally separately. A branch, you can only have one branch in one repo, so this just gives you N branches.
Ryan:Yep. Now it makes sense why that can be like portable, right? Because that, that's not, and so you have the work tree that handles that part. And then conductor, if I'm understanding this correctly, is just the isolating the environment. Yep.
John:it helps with that and creating the new workspace and then I, and then the isolation setup and tear down and then, you know, command R to run the server and, you know, give you a little button to open it in the browser and that kinda stuff. They don't have a browser in it yet, but, there's a ton of other things, maybe, I don't know what you're using Brendan, but like, there's like, um. Um, trunk.dev is like a, a command line, only one that's just the work tree and the hooks. Um, there's seam which is built on Go lib, ghosty. There's dm, which is built on UX and li maybe Lib ghosts. No, I think it's just built on T muck, so you can use it in BYOT, bring your own Terminal. Terminal. Um, but yeah, there's a bunch of different ones that are kind of on top of it that do the same thing.
Brendan:Yeah, I'm, I'm just using the Claude desktop app right now myself.
John:Nice.
Brendan:you use the Claude Code, you know, version of it, 'cause there's also cowork and chat and all that
John:Yeah.
Brendan:It has a little checkbox that says, do you want to use a work tree? So, but that was my first exposure to it.'cause I mean, I remember, you know, we LA last year, Ryan, we had, uh, um, I forget, there was a couple people who were kind of asking this question like, how do you have multiple things going? And that was my unlock. I was like, oh, it's just gonna use a work tree. What's a work tree? And yeah, just kind of figured it from there.
John:And you just ask Claude and Claude tells you, like, and Claude from the CLI has like a dash w and it'll just make a new workspace automatically with a random name. Or you can actually like, you know, Claude dash w, my new feature and it'll do that. Um, but again, then you need to get hooks to, or you need to manually run the setup and tear down and things like that. So it's not perfect gravy like that. But conductor is, it's flow. I mean it's pure flow. You can work on three to five things at once, move 'em all forward. And it's pure flow, like you don't feel kicked out of it in the good space. Um, when you start getting to 10 workspaces and up. when it starts to get really painful.'cause they haven't solved, uh, some problems with, uh, performance and with, uh, rate limiting on GitHub everything goes, I mean, it goes, dude, I, it like scrambles your brain 'cause things are just bouncing around in the sidebar instead of saying consistent because. It hit the rate limit. And so it can't tell if there's a, a PR for this branch and they don't store that in SQL Light locally, like they store everything else. And so then gonna, uh, again, I'm not gonna go into it right now, but there, there you can run into pain. And if you wanna know the pain, I'll tell you later. So
Brendan:Yep.
Ryan:Well, let, let us, uh, or at least let me play with some of the stuff you've already said, and then when I hit the pain, we'll have you back on a third time to talk about the pain. Um.
John:yeah, hopefully it'll be gone by then.
Ryan:Yeah. How, how much did you have to change the actual apps in order to get this to work? Right. Was there, was there anything that you had to change? You know, like I, I know in our app we, we have, we use a, a custom port already,
John:Mm-hmm.
Ryan:and I'm already thinking like, all right, is that gonna be a problem with, you know, mapping the ports for the different workspaces? Right. But was there anything that you ran into that way, or did it, was it mostly plug and play?
John:It is, honestly, it's very easy, like I would say on, you know, we've got, I'm on like four or five, you know, big rails apps, and I would say it's fewer than a hundred lines of code. It's typically like five or 10 lines of app code and then a bunch of scripts.
Ryan:Yeah.
John:end up being the longest part of it, honestly. And they're the easiest part.'cause Claude just writes 'em,
Ryan:Right.
John:you know, there's no for that part. That's, I don't know. Is that, do you feel like that too or, um, for the, or I guess you're using the Claude work tree stuff, so I guess it's a little different.
Brendan:yet, but
John:Okay.
Brendan:that sounds,
Ryan:But I feel like you can kind of TDD this, right With, with Claude is like, all right, you know, you do the thing, it doesn't work. You'd be like, here's the problem. Go write the script to fix the missing environment variables. Go write the
John:Yep.
Ryan:script to do X, Y, Z. Right? So, I mean, to that extent, I imagine it wouldn't be that difficult, but I have a feeling, at least for one of our apps, I'm gonna have to change application code in order to make it compatible with that, you know, maybe. I mis fundamentally misunderstanding how the port mapping stuff works. But that would be my only hesitation. I'm sure Claude. I'm sure Claude can fix it too. So it's like whatever. But.
John:Yeah. Yep. Definitely can. And that's, I literally all the original, like what I started doing is just making like a bin conductor dash setup script and conductor dash tear down or archive. Um, and so like, and then we eventually did like a conductor run, I think as well. Like before it was like, you know, bin dev or whatever. Um, but once in a while it's like. I dunno, they can be the same scripts. I mean, you could literally just make it so that like bend dev when you run it, like it just, you know, knows how to do it or script set up. Just knows if conductor's there, do some different things. Like, that's basically what we're doing is just like, if conductor's there do some different things um, you're good to go. Uh, it's, and it's really not a lot of things. Uh, I mean, the first time can be a little painful. It can take a couple hours. Um, if you use the article I wrote, that'll make it less. but you, you know, there's, every app has weird little dumb things that it does that, you know, I probably haven't done the that dumb thing yet. And so those will be the things that you'll, you'll start using it and you'll be like, why doesn't this work? Oh, yes. And you know,
Ryan:Yeah,
John:that.
Ryan:well, we, we have a, a web builder that allows you to host a website on a custom domain. And that always, you know, it doesn't matter whether it's like, you know, doing docker container or whatever, like, that's always something you have to deal with no matter what. Right. Because like this, you just have to spin it up on its own port. Right. And, and I'm sure that's easy to do, but you have to write the script to do it.
John:I got it. Fix for that now too though.'cause a
Ryan:Okay.
John:does the same thing. And I was like, okay, what? How do I solve this? And I ended up using Caddy and it's awesome.
Ryan:Okay. I, I've seen Chris talk about that on Go Rails, uh, or, uh, remote Ruby rather. Um,
John:It's so good. You just put it in your proc file. Again, like another thing that spins up when you start the app
Ryan:okay.
John:then you have a caddy file and then you can just tell at Star Fireside, local host goes here, you know, and then you get HT PS. So you can test that. Um, you can do custom domains. Um, you can do anything you want. So you could have like my custom domain, local host. different than like, you know, food bar fireside, local host and all that. So we ended up, I think I put caddy in all of our apps now just so we'd have http s 'cause it's so lightweight and, and we just run each one. You can run it globally and just do API requests to like mod the fi, the config. we just run it in the pro profile of each one. I just prefer like everything is either running or it's not running for this app that it needs in one isolated place. I don't like global stuff'cause then you get weird. Global state where something didn't clean up. Right. And now everything's broken. So, but yeah, caddy Caddy definitely solved that for it's brew installed caddy and tell Claude to do it and it's
Ryan:Yep.
John:I mean, it's really cool.
Ryan:Yeah, that makes sense. I've been wanting to do something like for, for whatever one of our apps uses, uh, lvh.me as the local development domain, which I've been wanting to get rid of for a long time.'cause it's like now local hosts can do SSL. Right. So it's. But yeah, CA Caddy will help with the custom domain thing, and then I can stop having to manually add everything to a host file in order to
John:Yep.
Ryan:view, uh, the page, right, which,
John:I was just doing that until like a month ago and then I finally decided to try Caddy and I was like, oh, life is so
Ryan:yeah.
John:And even just like then cores is an issue, so then you gotta star any port and cores and there's, yeah, but
Ryan:Oh, that's a good, that's a good tip. I'll have to look into that one.'cause, uh, it is not so bad when it's just my, you know, local development, but then you had to deal with like, uh, CI and GitHub actions and it's like, all right, well now I have to like, spin up a specific thing just for testing and that feels wrong anyway.
John:Yeah.
Brendan:Well, maybe this is a good place to wrap up. I wanna respect your time, John. It's been great chatting with you. But, uh, anything else you wanna add before we, uh, sign off here?
John:No, I mean the, I guess the only thing I would add is everyone who thinks that the, that all of our jobs are gonna be gone. Uh, like look at the jobs report. I said I just saw some, I dunno if they're true or not, but I saw some jobs reports that showed that like software engineering is like the only thing that's like going up kind of, it's like up four point half percent month over month or something. And I was like, that's wild. I think everybody's nuts for it right now. Now maybe that's just because everyone's like, well now we can do even more. And so they're getting. Bullish or they see that there is an arbitrage over the next six to 12 months before everybody picks up on this good stuff. Um, I mean, that's the way I'm looking at it, but yeah, I'm not sure. So I guess try to stay positive about it. Definitely learn. Don't get behind 'cause that's probably the worst, worst place. Um, so.
Brendan:Yeah, definitely agree with that. All right, cool. Well, we will link to, you know, conductor and some of these other things we talked about in the show notes, um, and your Twitter or anything anywhere else you want to share.
John:No, doesn't, that's fine.
Brendan:All right, cool.
John:Yeah. Thank 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