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
The Bottleneck Isn’t the AI
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Just Ryan and Brendan this week, checking in on things from past episodes. Ryan reports back on dropping AppSignal for his homegrown monitoring (it did its job, and it changed how he thinks about sampling), and Brendan cut a $1,200/month New Relic bill to $299. From there: why AI makes small pull requests matter again, John Nunemaker's five-open-PR limit, Ryan's Grok Bot setup (the Bug Bot that works and the project-manager bot that doesn't), and Brendan's markdown-in-Git command center that now runs the podcast through the Descript and Buzzsprout APIs. We get into why "proactive" AI really means well-defined loops with limits, why the LLM isn't the bottleneck (we are), when it's fine to ship with a known edge case, and how we're each doing on AI burnout.
LINKS
- Grok
- Descript
--Sponsor--
This episode is brought to you by Flagrant. Your partners for delivering thoughtful products & applications from any stage in your project.
Questions or comments, email us at railsbusinesspod@gmail.com
Hey, how's it going?
Ryan:Not too bad. You?
Brendan:Well, as we were talking pre-show, it's daylight savings week, so feeling a little
Ryan:Yeah
Brendan:than usual, but otherwise not too bad
Ryan:Yeah, I feel, I feel you on that one. Um, just, uh, trying to go to bed early a couple nights this week and catch up
Brendan:Yep. I, I wish I was better at that, but I'm not. But, but yeah, um,
Ryan:Feels like it's been a while since we've recorded just the two of us. I don't know if that's true or false, but
Brendan:Uh, I think it's, it's been at least a month. Yeah, I think our last solo or whatever, the, the two of us episode, no guest, was, um, right after, uh… Well, actually, maybe it wasn't that long now that I think about it. Um, because it was right after RailsConf, right before or
Ryan:right
Brendan:I guess it was, 'cause we talked about the, uh, DHH's stuff,
Ryan:Right, right. All right, well, we're gonna do a little bit of a, little bit of a check-in here, right? And talk about, you know, follow up on some of the things we've discussed in the past
Brendan:Yeah, I, I took the, the liberty of, uh, asking Claude, who has all the transcripts for every episode. I was like, should we talk about this week?" And it gave me a, a bunch of stuff. Um, none of it felt like a full episode's worth, but who knows, we could prove that wrong. Um, but yeah, to your point, like, got some things that I feel like might as well check in and see how they're doing. Um, so the first one on my list was, I think it was maybe a month, month and a half ago when you mentioned… Or maybe it's been longer, I forget now. Um, I don't have that ch- uh, update here. But question is, your do it yourself, uh, monitoring, how's that going? Is that still running? Is it living up to the promise that we talked about in that episode?
Ryan:Yeah. Well, I mean, I, I think the promise was stop wasting money when I don't use it that often,
Brendan:Yep
Ryan:yeah, so to that end, I mean, it's given me everything I need. Our apps are, you know, stable. Um, you know, I have a metric now that I can sort of measure against to see if I'm going in the right direction.
Brendan:Yep
Ryan:that sort of round of work, it's generally stayed stable, and I know what the next thing to address is, right? And that was really all I needed. Um, you know, I think where I overshot is I built a lot of instrumentation that I needed to solve the problem at the time that I probably don't need now. And if I were being diligent, I could probably go back and strip some stuff away, but it's not causing any problems, right? There's no real additional server cost that it's, you know, incurring. Um, you know, so I, I think it's given me what I've, I've needed. Um, if I had time, I'd probably iterate on it a little bit more, right? Like, uh, like, so here's an example, right? Like, I have on the dashboard, it shows me the slowest endpoints, right? W- which is helpful when I was, uh, uh, originally triaging. Uh, but now, like, the next slowest endpoint is one that I can't address quite yet, right? But it takes up the top, like, 10 slots, right? So I don't know what the next thing is.
Brendan:Mm-hmm
Ryan:so if I were in performance mode and looking to, you know, improve again, I'd probably go in and say, like, you know, "Could I mute this for 30 days or, you know, exclude it until I re-include it later?"'Cause, like, I know it's a problem. I've got it logged in my ticket. I know what the fix is. It's just gonna take a while for me to get there, and in the meantime, I don't know what the next thing up would be.
Brendan:Right. Yep
Ryan:but beyond… You know, like, it served its purpose. Um, I also saw something in a newsletter recently that, uh, I think it was John Nunemaker looks like he's working on something that's similar.
Brendan:Okay. Well, he, he had
Ryan:and I…
Brendan:at one point,
Ryan:Did he? Okay, I don't remember. So, like, I, I signed up for the beta, and I need to ping him to see if I can actually get in. I, I, I'm curious to compare what he's doing with what I did and see, you know,
Brendan:Yep
Ryan:if I, if I can have something that serves the same purpose at, you know, the, a cost that makes sense, then I'd rather not even have to think about it. That, you know, that is the trade-off. But, you know, fair, to, to be fair, I haven't given it much thought since I solved the problem
Brendan:Yeah, a- and just listeners that weren't or, or didn't listen to that, that episode, can you remind me and, you know, the, the audience, did you completely shut off, um… And I forget what you, you were using. It wasn't New Relic, it was, uh,
Ryan:It was, uh, I had AppSignal.
Brendan:AppSignal, that's right.
Ryan:it is off,
Brendan:off?
Ryan:it, it's off in terms of, like, I am not sending any data to them anymore,
Brendan:You're, you're
Ryan:and my subscription, my subscription's on their free plan.
Brendan:Okay
Ryan:I left all the instrumentation, so, like, turning it on is just flipping a feature flag and then, you know, ultimately upgrading for the month. So if I need the high-powered, you know, uh, tool because what I have is insufficient, it, you know, I have it there to turn on.
Brendan:Yep
Ryan:now obviously that only helps going forward, right?'Cause I'm not sending data, so I wouldn't be able to be reactive. But if I was, you know, stuck on something and I wanted to measure it for, you know, a month or a week or whatever, and, you know, I c- I can still do that. Um-
Brendan:Right
Ryan:I'm, I'm also, and this is sort of an interesting side effect of, you know, going through the process of building the tool made-- actually made me more comfortable with like sampling.
Brendan:Okay, sure
Ryan:you know, so every time I went to, you know, whether it was AppSignal, Honeybadger, New Relic, you know, and, and, you know, w-with the q- usually with the question of like, "All right, how can I make this affordable?" Like the response is, "Well, don't send everything."
Brendan:Yeah, right
Ryan:Right? And I was like, well, that felt wrong to me,
Brendan:Hmm.
Ryan:right? Because it's like, well, if I don't send everything, like what is the, what's the point?
Brendan:Yeah, right
Ryan:but even with the tool that I built, I'm not sending, you know, I'm not measuring everything. I'm sampling, right?'Cause otherwise I'd just, you know, flood my own database. Um, so, you know, arguably I could go back and be more aggressive with that and probably get it to a point where it was, you know, at a price point that I cared about. But again, you know, I don't have to think about it 'cause it's not a problem right now.
Brendan:Right. Yeah, and interestingly enough, um, maybe two months ago, it was definitely after you and I had this conversation originally, I actually went through New Relic and I think we were sp- what was it? We were spending at least $1,200 a month and, and nobody was really paying much attention to it and whatever. now I've got it down to 299 a month,
Ryan:Right
Brendan:th- some of it was, there was, and I, I wish I'd thought of it to look it up, but there, there was like a feature that was added since we originally set it up that, um, I'm trying to think of what the right term is, but it was like basically sending more data. So that was part of it, 'cause all of our costs pretty much was data. I think we've got like the one user that's $100 on New Relic,
Ryan:Right
Brendan:had $900 worth of data usage. part of it was that, but then the other part of it was really just, yeah, cutting down how much data and setting it so that it is sampled like you were just talking about. So I mean, I'm not spending nothing, but I'm also way down from where I was before, so. But I'm still interested in what John has, 'cause I know he, when he talked about it, I don't recall if that was on one of the episodes or if that was just a conversation we had with him. But I know he's even sent an email to our show inbox in the last couple months at least with like, "Hey, I'm, I'm working on a thing." So I,
Ryan:Yeah
Brendan:to reach out. We'll, we'll have to get some more and maybe he'll wanna come on here and talk about it too. So
Ryan:Yeah, that'd be nice. So I mean, he's been doing a lot of interesting things lately that I've been attempting to emulate, and maybe we can get into that later. But
Brendan:Cool.
Ryan:all right, so any other… I, I would say to wrap up that bullet point, I'm happy I did it. I'm saving a reasonable amount of money. It served its purpose. I'm still open to solutions, right, that I don't have to think about,
Brendan:Yep
Ryan:at the right cost. Like I'm, I'm willing to spend a few hundred bucks a month to not have to deal with it.
Brendan:Right. It, that's a good point. That, that's
Ryan:Yeah
Brendan:why I was asking you what you were paying or if anything, I was curious, like, what you were, your thoughts were around, like, John's tool for instance if it's, say, $300 or $200 or whatever. Like, are you willing to go back to paying? You just answered that, but it's a good
Ryan:Yeah. I mean, uh, it, it would obviously would depend on the tool and what it does, right? Like, I would feel comfortable in the one to 200 range saying yes, for sure.
Brendan:Yep
Ryan:again, I'm, I'm just gonna question whether I actually need it,
Brendan:Right
Ryan:it's like a security blanket. Um, y- you know, but at the same time, right, like I am thinking a lot about, you know, observability and how that can be used, you know, as part of a feedback loop with AI, and that's a, a bit of a different conversation, right?'Cause if I'm leveraging it every day as part of my development workflow, now the value, you know, uh, statement changes, right? Like now I'm probably gonna spend more if it can accelerate and, you know, be autonomous. Um, so, you know, I, I think that's just another example of like I'm not sure where I'll land yet. Um, but yeah
Brendan:Yeah, that, that's an interesting one, and that could be a whole conversation, and maybe we, we do just kind of put a pin in it. But one of the questions that I have is, like, how do you get rid of pull requests?
Ryan:Hmm.
Brendan:And I think that is part of the answer, at least, is that if you have the monitoring fed right back into the LLM that when code is released, it sees almost instantly that something negative has happened, and you give it the ability to act on that some way, then I think that does at least reduce the need for the pull request in the first place to try and prevent, you know, bad things from happening. Now, th- there's a lot more to it than that, but I think that is one piece that is worth considering
Ryan:Yeah, and the other thing I'm spending, uh, some time thinking about is, you know, and this is not a new concept, right? But AI has made it really easy to make these ginormous pull requests.
Brendan:Mm-hmm.
Ryan:Um, and there's, there's a mental load, uh, the bigger the PR gets, uh, as a developer of, right? Like, I really don't wanna review this. I don't wanna have to load up all this context. Like, it's just not gonna work, right? And, um, you know, again, following John's, uh, you know, blogging and tweets, it's like, all right, we kind of need to just go back to small pull requests. Like, nothing more than 500, you know, uh, you know, two or three hundred lines of code.
Brendan:Mm-hmm.
Ryan:You know, small things that you can review and say, "Yep, that makes sense." Uh, the test suite passes. You know, you've had, uh, you know, your linters run, you know, your CI is green. Um- And if you can pop in and, you know, you know, that size pull request is easy to review, right? And a lot of times you can, you know, understand it without even having to boot the app, right? But there, there's other things you can do to be like, you know, always show proof, right? If you say you're making something faster, show the before and after of, you know, the, the, the timings, right? If you're using a-- if you're updating a user-facing feature that has a view, you know, show me the screenshot. Um, and just you automate pretty much the, you know, not only the code, but then like the review process to some extent, s- to the point where it's easier to come in and be like, "Yep, looks good. That makes sense." You can understand the code, right? And you can push it. Um, and I thought the other thing that I thought was interesting is he had a, a, and, you know, maybe I'm misunderstanding this slightly, but it sounded like he had a rule where it's like you can only ever have five pull requests open at a time.
Brendan:Okay. Mm-hmm
Ryan:Um, you know, obviously he's a solo or, you know, one or two-person shop, depending on what, uh, business he's working on. So, you know, maybe that has to be slightly adjusted for bigger teams or whatnot, right? But I thought for a solo dev, that actually makes sense when you want to give, you know, an LLM like in a agentic loop, right? Be like, you know, every hour check to see if, you know, there are open PR slots.
Brendan:Mm-hmm.
Ryan:I'm, I'm paraphrasing here. This is not the words he used. Um, but if there, you know, is availability, then go look at the list of things that we've, you know, already talked about and, you know, find the next thing to work on. And he, he had a whole list of parameters that he used to figure out what that is. And, you know, the other piece that I thought was interesting was simply that, um, you know, that included bigger features, right? But the key thing was, is like, go find the next thing you can do that is a small PR.
Brendan:Okay.
Ryan:You know, and it turns out he, he's not wrong that LL, LLMs are pretty good at doing that, right? So like, you know, here's the big plan, here's the full picture. Go find the next thing, and the next thing, and the next thing, right?
Brendan:Mm-hmm
Ryan:And what is the… You know, that, that, none, none of that is new, right? Like small PRs ship quickly. Like we're g- we're, you know, we're, we're shifting back into just like, it's like we forgot like best practices for a while. It's like now it's like, oh yeah, that there was a reason for that. And, you know, the, the bottlenecks changed, but the solution to me kinda feels like the same thing, right? Just like, let's go back to smart, you know, engineering principles.
Brendan:Yeah, totally. we, we talk about this on my team too, even whether it's AI, AI coded or not, um, and most things are now. But, um, have to look for it, I don't remember the exact link, but th- there's a some post or something out there that we talk about a lot that you, I think you said 500 lines, like this thing says like 300 lines, and if you get over that, then like humans just can't keep up with it. Like it's too much.
Ryan:Yeah, well, I, I said five initially, but I-- that when I said it out loud, I don't think it was that high. I think it was two or three. But whatever it is, it's smaller. It's not the, it's not the 10,000 that I'm used to.
Brendan:Yeah, right. Right. Yeah, I mean, th- there's definitely stuff there, and I think the, the other part of that thing that I, I keep trying to, I don't even know if struggle with is the right word, but just like gnaw on is like most of the teams that are running these agentic loops in software factories or whatever you wanna call it, they're also burning just tons and tons of API tokens. They're not on the Claude Max plan or something like that. Partially because like to set something up like that, at least the way that I've looked into it, like you, you have to have it out on a- another server or something, and like at least Anthropic doesn't allow you to even use your Claude Max plan. It's gotta be API tokens. So then the spend just gets out of control. You know, if, it, it depends on how you wanna do it, like,
Ryan:Well, I mean, I, I think there is a…
Brendan:which is
Ryan:Yeah, I mean, the- there, there is a, you know, middle solution between, you know, what, like the factory, you know, the software factory and, you know, the, the normal just like human in a loop flow,
Brendan:Right
Ryan:So you wanna talk about Grok Bot, right? And, you know, I'm having Good luck with some things, right? Like, so like the-- I have a, a bot, uh, that is a, um, bug-- I ca-bu- I call it Bug Bot, right? And I've hooked it up to a Honeybadger, and it just watches for errors, and the instructions I've given it are, "Look at this, you know, look at the code. If it's a small fix that is easily understood," and I forget the exact language I use, but you know, the gist is, you know, if, if it's easy and clear just do it, right? Create the PR, put, you know, the instructions to test it in, you know, the, the, you know, sh- write the instructions so that I can reproduce it on master, see it fail, run it on the new branch, see it pass. You know, make sure the tests do the same thing And that, that has been going really well. And it, and it-- and if it gets into it, it's like, I'm not sure or I, you know, I have questions, then I tell it to open up a Trello card. We use Trello for bug tracking. I just say, "Open up a Trello card and, you know, tag it with all the details," and I'll go in
Brendan:Mm-hmm.
Ryan:comment, right? And I, and it, I instruct it to watch the comments on the PR. So like, if it pushes something up and I- it's like, "No, this isn't right," I can give it instruction, and it'll just pick it back up in the next loop. And, um, you know, if it's, uh, on a Trello card, I had say,"Watch the comments," right? And I'll comment with further instruction. And then, you know, Monday morning, I log in and I review, you know, what's in Trello, and it's like, you know, some of them are bugs that I do wanna fix and some of them are like, you know, noise. And I say, "Ignore this. Put this on the, you know, ignore list," and it closes it out and doesn't bother me again. And,
Brendan:Yep
Ryan:you know, some things I'll say, it's like, "Yeah, this is a bug, but I'm gonna like group it with this other work," right? So like, just, you know, add it to the plan doc for, you know, event feeds, and I'll get to it when I get to it, right? Um, so that one's going really well because it allows me to respond to customer-facing errors very quickly, right? With very minimal amount of work
Brendan:yeah, a- and just so that I'm understanding, are you saying that every single one of them will either open a pull request or give you the Trello ticket, but, like, none of them are… Nothing's merging without you still, right?
Ryan:I, I'm not there yet.
Brendan:Yep.
Ryan:Right. And I, I think I'm--
Brendan:Yep
Ryan:Yeah, I, I'm not there solely because I don't have the instrumentation or the guardrails that would make me feel comfortable,
Brendan:Mm-hmm.
Ryan:yet without, you know, my eyes on it. Um, but that's fine, right? Like, I don't… It's saving me time, and I'm fixing bugs, uh, most of the time before they're reported, right? Which is like… And I've made it sort of a point as if it's, you know, if it's a, a customer, f- you know, a paying customer bug, like I then ping customer support and say, "Hey, reach out to this person. Tell them, you know, we saw this, we saw this error report. We've addressed it." Right? That, that buys goodwill, right?
Brendan:Yep
Ryan:Um, so that, that one's going really well. The other ones either I have not had a chance to sort of tune,
Brendan:Mm-hmm
Ryan:or have just become… Like, I have a product manager bot that I wanted to sort of help me prioritize, uh, and that is not going quite as well as I wanted it to.
Brendan:Yeah. A- anytime it's like, me my to-do list," or, "Give me my priorities," like I've found that to be a major struggle so far
Ryan:I, I have ideas of what I can do next. I just have not come back to it.
Brendan:Yeah. The one thing that I've been trying l- recently is trying to tie it with, like, roadmap things. So if I've got a bigger roadmap item, then that's, like, the way to give the LLM a picture of, like, what the priorities should be. Like, if it's on the roadmap, then it's gonna be a priority. Now, granted, there could be 10 things on the roadmap, how do you prioritize those? W- we, we generally don't have that many, so, like, we usually have maybe five, six things on the roadmap at a time. We do it quarterly. that at least helps, and I can also then narrow it down a little bit and say, "Okay, of these six, pay attention to these two or three right now."
Ryan:Yeah
Brendan:completely fixed it yet, but that's been the thing I've done in the last, say, two weeks, and it does seem to be helping more than anything else I've tried without something like that
Ryan:Yeah, it's definitely… This goes back to maybe one of the, the, the first or second episode that we recorded when we were talking about how I organize my tracker,
Brendan:Yeah. Mm-hmm
Ryan:right? And that's evolved a little bit, right? But now I have, you know, I don't care what you call them, epic, sprint, whatever, a group of t- related tasks.
Brendan:Yep
Ryan:And I have like the next 10, you know, columns that are like that, and they have a, you know, a name that is like of the group and… But each o- each one has now, you know, depending on what it is, it could have five cards, or it could have 50,
Brendan:Sure
Ryan:right? And, and it's a mix of bugs, it's a mix of features, it's a mix of ideas. But I've instructed it to be like, you know, here, you know, treat those columns as sort of the roadmap, and that helps a little bit. But, you know, Grok is like it-- I have it set, I think, to check three times a day, right? Like, I don't need like constant on, just like somewhat responsive. Um- It has a hard time understanding that I might have three or four things going at once, right? And, you know, may- maybe that's my problem and not Grok's, but, um, you know, if, if, if I've said, "Hey, I, I was working on this yesterday, and then today I, I do a little bit of that and something else," it, like, it gets mad at me. It's like, it's like, "Oh, y- you still have this other thing outstanding." I'm like, "Yes, I'm, I'm well aware of that, thank you," right? But I've got these other new priorities now that, you know, things have come up. Like, that's the job.
Brendan:Yeah, right
Ryan:um, and, you know, I've, I've just gotten tired of correcting it, so I've sort of stopped, right? Um, but where I do find it useful is when I'm like, "All right, I've got an hour. What can I get done?"
Brendan:Yeah
Ryan:Right? Like, to triage my, uh, open PR list. It's like, g- give me a PR that is doable. Like, could, could I get this shipped in an hour, right? That, that's what I got. So it's good at those things or, um, you know, give me something different to work on today 'cause I need a break from what I've been working on. Like, it's okay at that kind of stuff too.
Brendan:And for that, you, you called it project manager earlier. Is that basically all you told this bot? Like, "You're my project manager, I need you to keep track of the roadmap or Trello or whatever"?
Ryan:Yes. Um, I don't remember the exact prompt, but the, you know, to paraphrase, it was like, you know, help me manage my priorities so I don't have to remember, you know, what's on the, you know, what's next up on my to-do list, right? Because I'm tired of writing it down every day and then just like I get to half of it and then I, you know, I'm OCD, so I gotta, you know, relet- rewrite my list every day and I'm like, this is a waste of time. Like complete waste of my time to just rewrite lists first thing in the morning.
Brendan:Yep
Ryan:so I was like, all right, like I don't wanna forget it. Like a lot of this stuff isn't like super critical, right? But it's taking up brain space, so it's like, just make sure it's there so if I ask later, like, you know, if I, if I need to ask like, "What is that thing that we talked about in November about XYZ?" It has better recall than I do. Um-
Brendan:And, and that's exactly where I've found the value with mine too. Like, it is that part. like the, the second brain idea of, like, storing things for later and having it help find it. It's really good at that. horrible at prioritization.
Ryan:Yep.
Brendan:So
Ryan:Yeah, I mean I think I started like, 'cause I've got Like, you've got your roadmap, you've got your tickets, you've got, you know, company priorities,
Brendan:Mm-hmm.
Ryan:And then I've got, as of yesterday, I think 180 open PRs, which is ridiculous. Right? Um, you know, I'm, I'm closing, you know, 20 or 30 a week, and it just still is going up, right? So that's a different topic. Um, what was I saying? Uh I completely lost my train of thought. What was I talking about?
Brendan:Oh, we were just talking about prioritization and how it, it's good at… not good at prioritization, it's good at, uh,
Ryan:Right, right.
Brendan:yeah
Ryan:I have all these things, and I wanna be able to say like,"What are like the blockers?" Right? Like, help me unblock it,
Brendan:Yeah
Ryan:'Cause like with, with that many PRs and that many tickets and that many things, like there's so many interdependencies. It's like, help me find the next thing that I can meaningfully push. And that was-- Like I think it's okay with that when I ask it like a specific question, right? It's like, you know, "I need to get this thing done. What's in the way?"
Brendan:Mm-hmm
Ryan:and, and, and much more quickly than I could be like, "Oh, well, you need to merge this PR before you can do this thing."
Brendan:Yep
Ryan:what I probably should just turn off is like the daily, like here are your priorities. Like that's, it's, that's just a complete waste, right? But when I need to be like, "Go look at like… All right. Here's what I need to get d- I know I need… This is, this is the priority today. I need to ship this feature. What's blocking it?" Right? It's very good at going and looking at all the things and being like,"Well, if you ship this other PR first, that's gonna make the merge cleaner. And if you do this other thing, you know, and oh, you know, before you, you start doing your code, like make sure you send the, you know, the writeup to marketing in order to make sure that they're started on the, the email that needs to go out in conjunction with the deploy." Right? Like, it's good at that, and that, that's helpful on its own. It's not what I set out to build.
Brendan:Right. Uh-huh
Ryan:but you said second brain, like my-- the, the last bot I have is like a chief of staff. Um, and what I'm slowly doing is it, I'm feeding it, uh, more and more context about the business objectives.
Brendan:Sure
Ryan:Um, but I'm at a point now where I'm not sure I wanna continue with Grok or if I wanna try to use ChatGPT or Claude or something else, mostly because the rest of my team all use ChatGPT, right? So like I wanna train it on company stuff, which means I want them to be able to sort of get the benefit of it. And then I, you know, not to, to come back to John any more than we already have, but like he talks about like, all right, you should own your context.
Brendan:Yep
Ryan:really wanna train up Grok and have it like, you know, it's a black box. Like it's storing something.
Brendan:Mm-hmm
Ryan:now I'm like, all right, do I have a lot of MD files in a Git repo that I control? Do I go down the SQLite path and, and… I don't think SQLite's where I wanna go. I think I'm gonna end up in GitHub for the short term with just a bunch of files. But then I immediately was like, all right, well, does… What lives in, in the product repo and what lives in my repo?
Brendan:Yep.
Ryan:Um,
Brendan:And that's exactly the world I'm in,
Ryan:yeah.
Brendan:my command center stuff all lives in its own Git repo, which by the way, because I've got a Mac Mini that looks at that and I've got a computer that looks at that, it, it's getting better, but it, it constantly was, like, hitting conflicts in Git to merge things.'Cause I have it, like, basically auto-sync, so everything it… Every action it does almost, it, like, syncs it and pushes it, so then it gets into… And, and maybe that's the, part of the flaw. We don't need to analyze that, but I just wanted to point out that, like, that's the system I have, and it works for the most part. I'm
Ryan:Well, let's get into it. T- tell, tell me a little bit more about, like, how you structure it, like what your thought process has been, 'cause, you know, I'm trying to make some of those decisions, and I'd rather not start at zero.
Brendan:Oh yeah, no, the, I'm okay getting into that part of it. I guess what I was talking about is, like, whether the, the Git commits and the auto-syncing is a problem. Like, I don't, we don't need to figure that part out. m- maybe, maybe. I mean, maybe it'd be helpful. But, um, so what I have is, yeah, it's just a Git repo. It's all… I told it everything has to be in an MD file so that it's easy to search, easy to create, whatever. Um, I let it determine how it wants to store things, so if it wants to put it in folders or, you know, how, subfolders, whatever, like, it, it does all that. Um, I basically give it, you know, "Okay, I know Ryan, and I do podcasting with Ryan, and I do, uh, some consulting with Ryan or whatever." So it knows about you, and it, like, can remember things like,"Oh, what about this project?" Or, you know, "What about this invoice?" Or things like that. it knows about basically every project I do. I've… The, the personal one I've got it s- actually no, both of them I've got it set for my calendar and my, email. So, like, on the personal side, I have it check my emails for, like, are there school emails? Are there, you know, hockey emails for the kids or anything that they're in? Stuff like that. the personal side, it's like my boss, anything that he sends in Slack, which, which is only in the, the, uh, work side. anything that he sends is, like, top priority. At least I've gotta respond to that right away. And then there's, like, tiers there, and that's again back to the priority thing. It's like it's not, it's still not perfect with that. Um, but yeah, I guess, like, I'm just trying to think, like, what are the other interesting parts?'Cause it's, it's very similar to what you're already doing, but to your point, like earlier, it's, it's less of a black box because I've got all the files in Git.
Ryan:Right
Brendan:And, you know, I, I interact with mine almost completely through the terminal still, although it is… I've been using the remote control stuff on Claude a lot, that's helpful, but you still have to have your computer open and stuff like that. So it's like I don't… By doing it this way, I've lost what you have with Grok Bot, which is the, the app interface because, you know, it's, it's all on a local computer. Um, so it, it
Ryan:But,
Brendan:on how much you value that sorta thing too.
Ryan:Right, but like, I mean, all that is is a cron, right?
Brendan:Oh,
Ryan:So it's like, I mean, you, you could easily run that without Grok with not too much effort.
Brendan:no, that part, yeah, and I have that all set up too. Like, technically speaking, has its own thing that works like Cron. It's not actually Cron. Um, and I forget the s- it's been a while since I set it up, but, but yeah. It's basically like I set up Cron jobs for certain things. Like I have it go out, and I look, it, it looks at Rollbar twice a day or something and tries to find, hey, do you have new errors that you need to look at now? Um, it looks at GitHub for p- for pull requests, and it does, it does like initial reviews for me, and I, I could get better at that part too. Like, I'd like it to get a little better at like scoring things, and even blocking. Like, I haven't really had it look into that. It can, I just don't have it automatically doing that stuff.
Ryan:Well, the Git-- the reason I wanted to do Git is like with Grok, like I have it connected to Drive,
Brendan:Yep
Ryan:docs, and it's in a folder that, you know, I've-- it's either called Command Center or Second Brain or Brain or something to that effect, right? So like it, it's… But the problem is it doesn't do version control,
Brendan:Yeah
Ryan:So then, you know, then I have to give it instruction to be like,"All right, every time you make a change, create a new file and archive the old one into an archive folder." And it's like, well, I mean, okay. I mean, it works, I guess, right? But my storage is gonna go up.
Brendan:The, I, I thought, thought Google Drive did have version, n- not
Ryan:It, well, it
Brendan:but I thought there was versions of files
Ryan:It, y-yes. Have you ever used it? Like, if I ever had to go back and, like, find something, it's not great experience.
Brendan:Yeah
Ryan:right, like I, we, I use Google, um, Sheets for quite a bit of things, and if someone changes a cell, like y-you can kinda figure out what happened,
Brendan:Yep
Ryan:but all you can really do is restore back to that point. And it's like, well, like I don't wanna lose all the stuff that I've done,
Brendan:Mm-hmm.
Ryan:tr- I don't trust it. That's the bottom line on…
Brendan:yeah. Yeah, no, I mean, a- and for what it's worth, like my setup, if anything gets messed up with Git or if I wanna even go back into Git history, like I just ask it to do that. It doesn't require any, you know, anything from me, and it doesn't require even the Claude agent to be like jumping through hoops or connecting to APIs or
Ryan:Yeah, 'cause it's just, it's just Git control.
Brendan:Exactly.
Ryan:Yeah. Um, so, you know, maybe that's just because I've been wired now to think of it as, you know, small changes to files over time. But, um, I do not like how the existing Google version control works from a UI perspective. So,
Brendan:Yeah
Ryan:um I, I wanna just be able to easily s- like see the progression,
Brendan:Mm-hmm.
Ryan:when I need it.
Brendan:Yep
Ryan:Um, so yeah, I don't know. Uh, I'm thinking I'm, I'm thinking I'm gonna try the GitHub route, but I still have the… I really don't want two for work and personal.
Brendan:Yeah
Ryan:Uh, my, you know, everything's so intertwined, right? I just, I need it all in one place, but then, like, how do I keep that, you know, separate at the same time? I, I don't know
Brendan:Yeah. It, it's tricky. I mean, partially for me it was like, well, I'm just an employee at my, my job, so I don't really wanna intermix those things. But I don't, I don't know if it really matters. It's not like it's… I'm not building software with it or whatever that's not company software, stuff like that. And I don't think that's what you would be aiming to do either. Um, uh, there was one other thought I just had that was… Oh, here's an interesting thing that it's a little bit of a tangent, but it's still related. It's all still the same system. like, for the podcast and editing, I've, I've actually got it down now to… it's taken me almost two years now to do this, and I've wanted to do it forever, and it, it's kind of stupid k- how easy it actually was. But, like, for the longest time I would, would get done recording, I would download the file to my machine, I would, upload it and whatever so I could listen to it, and then I'd do edits and all… Like, I had all these steps of, like, very manual, very manual. then it wasn't until, like, a couple episodes ago I finally figured out how to just import directly from Descript where we record uh, Buzzsprout where we host. then once I did that, I was like, "Wait a second, well, they both have APIs, so now I can hook my command center up to both those APIs, and it
Ryan:Hmm.
Brendan:do everything." And so
Ryan:Nice.
Brendan:now it's just basically, like, I tell my command center, "Hey, I'm done recording this episode. I want you to do X, Y, and Z," and it, like, literally does it all. And it takes
Ryan:Yeah, it's amazing
Brendan:fi- yeah, it's awesome, and it takes me like five minutes to do, like, everything that used to take me hours.
Ryan:Okay. I won't feel as guilty now
Brendan:yeah, right. I mean, I mean, I… The longest part now is just listening to the episode and doing the edits, and that's honestly not that big a deal usually either. So
Ryan:Yeah
Brendan:pretty, pretty pumped about that one. Um, but th- you know, it's, it's the same kind of idea though. It's like I keep wanting to find things like that, whether it's with coding or not, or, or managing, you know, priorities, all that kind of stuff. Like, how do you close the loop, and how do you make the AI be more proactive for you? Like, like an employee. You know? Like, like Jesse Hanley was talking about with Grok Bot back in the summer when it first came out, and he happened to be on the episode with us, and it was like he's got five employees or whatever he had. I don't feel like… Like, I've been trying to do that with my command center, and it just doesn't ever feel like it gets there. It's always waiting on me
Ryan:Yeah. I mean, it'd be interesting to go back to him and see how he feels now, right? And we have obviously not talked to him since, so I don't know how he's feeling. But what I can say from my own experience is I don't think about it that, like… I don't think about it that way anymore, right? There-- You label-- You can label them whatever you want, right? You wanna call it a, you know, marketing assistant, you wanna call it a product engineer, uh, I don't really care. The, the tasks that it excels at are things that are well-defined,
Brendan:Yeah,
Ryan:right? So like, you know, there are tons of things in a business, right? Like I think the podcast example is perfect. Like you knew what the steps were. You do the same thing every time, right? And we should create, you know, what-whatever you wanna call it, an employee, quote unquote, that does that one thing really well, right? So like… And if you, you know, it's defined, you know what it is, you know, you may or may not do it on a schedule or kick it off manually, but it doesn't really matter, right?'Cause it's like, you know it's defined, right? And that's what AI is. I think where we get lost is when we try to go to the undefined things, right? It's, it's-- There's no ac- You know, like it's, it's-- To me, its intelligence is based on the fact that it can do more more efficiently than we can.
Brendan:Mm-hmm. Yep
Ryan:Like, you know, I, I know people argue about, you know, are, is AI intelligent, not conscious, but like smarter, right? Like
Brendan:Mm-hmm
Ryan:not really, right? Like it's not capable of making a choice,
Brendan:Yeah.
Ryan:right? Like someone still has to tell it. Now, people are like, "Well, I can, you know, t- give it an objective and it does make choices." Well, that's true, right? But, you know, it's, it's taking inputs and burning through tons of tokens and making like a, you know, basically a probability theory guess on
Brendan:Mm-hmm
Ryan:and then it's reacting to data that comes back.
Brendan:Yep
Ryan:Well, that's all we do, right?
Brendan:Yep
Ryan:you know, I, I don't know, you know, to your point of like you want it to be more proactive, well, I'm not, I'm not actually sure we want that. What I want is it to be really good at the repetitive things that we do all the time, right? So like, you know, the, the podcast thing is a good example. I think bugs are a good example. I think performance and security on the engineering side are good examples that where you can sort of define a loop, give it data, and let it go Right? I don't really need it to make, like, priority choices. I want it to help-- I want it to aid me in making a priority choice, and there's a, a subtle distinction there, right? That is, you know, when I'm ready to sit down and be like, "All right, what's the next thing we wanna work on?" Like, I want it to help me figure that out,
Brendan:Yep
Ryan:it to be autonomous in that regard. Um, not yet.
Brendan:Right. But I'm the same way. Like, I, I don't want, want it to do all the work. Like that, for one, sounds kinda corny, but like it would take all the fun out of it. Like, why would I even wanna Then I'm just sitting there, like watching. Um, but, like, I, I guess when I meant proactive, it's more like if we, if you hire a human, you expect them to do certain things, whether it's a repetitive task or thinking, and I'm thinking more repetitive tasks types of things. Like, that's where the proactivity falls down for me right now, is that I'm basically, like, trying to get it to even do repetitive things. Not the, the podcast thing is one example. On the work front, it's like kinda get back into that territory to what you were even just saying, like, and maybe this is why it fails, is because I'm expecting it to be able to do more of a open-ended task than something that is determined, uh, pretty well. Or, or, or a- able to follow it, right? There's no steps to it necessarily. maybe that is part of my problem, but there's still that proactivity from my standpoint that's less of… I mean, yes, it's an automation, 'cause it's a cron job, again, to your point. It's kicking off the, the bot to go do something I don't have to kick off manually. That, I think that's the key, and even with the cron jobs, it's like I don't get enough value out of it. And maybe that's my pro- maybe that's…
Ryan:Well,
Brendan:The framing is
Ryan:I, I think there, there's two levels of it, right? Like, if it's just-- Like, I'll stick with the podcast example for level one, right? You know that you, uh, periodically do a podcast. Now, we try to do it every other week. Sometimes it's every three, sometimes it's every week, depending on, you know, our schedules a-and whatnot, right? So like, there isn't a specific, like, cron that you would, you know, know. But you could say, like, you know, we usually record Thursday or Friday, right, in general. So you could have a cron that says, you know, every Friday at 5 o'clock, check to see if there's any podcasts this week via the API, and if there are, you know, do the SOP for preparing the podcast for, uh, publication,
Brendan:Sure. Yep
Ryan:That, that's sort of level one, right? If there's nothing, it just, it checks, there's nothing there, it, it dies, right? Level two, I think is, is, you know, you go up a level and you say, "All right. Here are the things that I do regularly."
Brendan:Yep
Ryan:it's a blog. I'm just making stuff up. There's a blog. You know, I do social media. I have to check the PRs. Um, you know, I want to update my KPIs. You know, whatever it is, right? And each one of those has to have a level one thing
Brendan:Yep
Ryan:you, you set it up and the instruction I would imagine would be something like, "Here are my responsibilities. You know, y-you can do, you know, three things at a, at a time. Every time you check in, which I want you to do every hour, I want you to pick the next three things that need to be done and move it to the next step." You know, if it's something that's fully automated, then you can say, "All right, report back when it's done." Or if it's something that needs my, you know, human in the loop review, be like, "All right, you know, tee me up with what I need." Right? That's level two. Now you run-- You still run that on a cron,
Brendan:Mm-hmm.
Ryan:You know, you, you have to give it priorities, right? And, um, you know, John said he gives his lanes, right? I got a big lane and a small lane, right? A big lane thing, you know… And, and his reasoning I b- if I understood correctly, was because he's running locally and his computer can only hold, you know, do so many concurrent dev things at a time, right? So like I can do one big thing that requires like a big test suite run, and I can do one small thing, right? But I, I think you could apply it to, you know, just, you know, throughput, right? Like if we just tell it to go, like all the time, and it's like, always do the next thing, always do the next thing, it's gonna be overwhelming. Like, we can't keep up with it. And because there is always stuff that at this point in the evolution of LLMs requires human review, right? You just end up with, you know, in dev world, you'd end up with a million PRs that are never gonna get merged, right? But by giving it the constraint of you can only do three things at a time, or there can only be three things open until I say it's closed,
Brendan:Mm-hmm.
Ryan:but then telling it to like, "Here's how you make a choice. Like, here are the priorities. Like, here are my business objectives." Like, I think that's level two that we need-- we're trying-- we're both trying to get to here is like, all right, let it have some autonomy and choice
Brendan:Yep
Ryan:within the system or prompts in this case that we define, right? But it's still, it's still just a prompt
Brendan:Yep.
Ryan:with context on a cron in whatever schedule that we want, right? So it's like, I think we can get to what you're saying. Like that would be-- that would appear to be proactive, right?'Cause it's picking the next three things every time. Now, I still have questions about whether it could pick the three things that you think should be the next three things.
Brendan:Yep
Ryan:but that's a limitation of the prompter, I think, not the AI.
Brendan:Yeah. I think that's all right. Um I, I like talking about this though, 'cause like in my head it's, it's just kind of been like the proactivity is more of an amorphous thing now that I think about it more. But, you know, it's really just like you're not doing enough for me is kind of how I always think about it. But I, I think this is a good… You're laying it out in a, a way that is helpful for me to now think about, okay, I'm clearly not giving it enough, 'cause I've tried this like seven times, and it doesn't work.
Ryan:Right
Brendan:And, and even the AI will come back to me and be like, "Well, we've already done this like three times or four times or whatever." Like, changing it one more time isn't gonna fix it. Like, there's something else going on. Um, and part of that is on me, because I don't always take the time to actually figure it out. It's like, okay, I wanted you to do this right now. You didn't do it. Now I'm… Now my window's closed, and I can't work on it anymore. Um-
Ryan:Well, I'll also push back on the productivity comment too, right? Because on one hand, you sit there and you say, "How do I close more PRs?"
Brendan:Yeah
Ryan:Well, I, I can tell with, uh, you with some certainty, even though I don't know for sure, that those PRs were opened by, you know, some sort of LLM model chugging through a bunch of stuff, and the bottleneck is actually getting them over the finish line.
Brendan:Yeah
Ryan:right? So, like, productivity of the LLM is not the problem.
Brendan:Right
Ryan:it's incredibly productive, uh, more so than we could ever be. The, the bottleneck is on the other side, right? So it's like we have the feeling that we need to be more productive, but it's not the LLM, it's us, and it, it's relative, right? Like, I, I think if you were to look back and, you know, give it, you know, some thought, you'd probably realize that you actually are quite a bit more productive than you were 12 months ago in the amount of work you're actually getting done. It feels like you're getting nothing done because you see the possibility.
Brendan:Yep
Ryan:until we get to full autonomy, I think the key thing that I've been wanting to experiment with is, like, giving the LLMs, like, artificial limits of what it can do at any one time and, like,
Brendan:Yep
Ryan:to the point where it's like, "No, dummy, like, finish the work that you started before you start the next thing," sort of parameter, right? Like,
Brendan:Yep
Ryan:just so easy to open the next thing, and there's a, the, the, like, the, you know, the addiction, the dopamine hit of getting another thing done is great. But if it doesn't get into production,
Brendan:It's not actually
Ryan:like, what, what actu- like, what, you know, yeah, I felt good. Like, it's like, "Oh, yeah, like, I-- that would be a cool feature. Like, I could get that done." And, you know, I get 85% of it done, and then the last 15% sits there for six months. Like, that's not helpful. That was a waste of time.
Brendan:Right.'Cause then can get in all that, but it's just typical, like, oh, well now I lost the context on that. What was this even for? How do I review it? the things you talked about earlier. Um, a- and this is also reminding me, like I… It was maybe a month or two ago now I, I was not intentionally, but, uh, as I look at it now, I was like, I guess I was sort of using my Claude agent as a therapist
Ryan:Hmm.
Brendan:ways. Un- unintentionally, but I was basically, like, trying to analyze habits, I guess. And like, okay, it, it, I never feel like I get anything done on a personal side. Like, I'm always trying to, like, work on something and get it done. And the, the analysis that it basically came up with is like, "No, you don't have a shipping problem. You have a problem, and, like, actually being, like, the businessperson problem. So you're incredibly productive from building things. You just never put it out into the world to actually anything with it."
Ryan:Yeah
Brendan:an interesting thought, 'cause like I already know I'm not a great salesperson. I don't love that part. But I thought I was better, better at it than, than what my… this is all based on the f- the inputs that I've been giving to it. It's not like I was like, "Hey, analyze me specifically." It's like, "Here are the things that I've been doing with you over the last nine months." So it could just easily look at them and be like, "Yep, you suck at this part, but you're really good at this part." so,
Ryan:I don't know if I wanna ask Claude that about me.
Brendan:Well,
Ryan:I'm scared of the result
Brendan:It was kind of involuntarily. I was, like, trying to get to a different point, and then it was like,"Well, actually, if we talk about this, it'll answer your question." like, "Okay, great." Um, but yeah, so be careful what you ask. But, uh
Ryan:it, it, maybe I read into what you just said, right? But you said on the personal side, does that mean like, you know, uh, you know, household tasks? Or does it mean…
Brendan:n-
Ryan:mean?
Brendan:yeah, good question. It, it's more projects, like things that I want to do, things I'm trying to build that maybe would become a product that I could try and sell or something like that. Which it- there's varying degrees of why I don't do some of that stuff, or I get to a point and it's like, "Oh, okay, I'm, I've built the product, but I'm not gonna, like, give it to anybody else." But, you know, it, it, there is a, a pattern here, and it's also told me, like, the way that I send emails and stuff to people. It's like, well, you, you, you are, um… Oh, I forget the, the terms. But, you know, it's basically like I'm self-deprecating a point where it's like you're undercutting yourself and all this kind of stuff. And it's like, well, you're, you're, you're not wrong, I guess, but I just don't think about it until somebody points it
Ryan:Yeah. I mean, it, it would tell me that I'm short and can come across as rude, so I get it.
Brendan:Yeah,
Ryan:But it, it just-- to me, to me that's efficiency.
Brendan:Right.
Ryan:Or like, yes or no. You asked me a yes or no question, I gave you a yes or no answer. Like, maybe ask the question that is actually you wanted responded to.
Brendan:Right.
Ryan:Uh, no, I, I've, uh, realized that, uh, LLMs are much better at writing more empathetic emails than I've ever been able to do on my own, so I am leveraging that quite a bit these days.
Brendan:Mm-hmm
Ryan:I'm sure it's obvious to other people, but it comes across as friendlier, I think, so
Brendan:Yeah. I, I mean, for instance, one thing that it was, guess, coaching me on it, it was basically telling me, like, "Well, if, if Ryan wants you to do more work, you should ask for more money." And I was like,
Ryan:Yeah.
Brendan:I
Ryan:Yeah, good call. Good call on that.
Brendan:one."
Ryan:Yeah.
Brendan:And so the… There, there's things like that, I guess, it would be
Ryan:Yeah.
Brendan:But, uh,
Ryan:Well, hopefully we've rectified that one. But,
Brendan:no, we're all good. We don't have to dive into the specifics here. But, um, uh, I don't know. I feel like we're, we're a little bit off on a tangent. It's all related, but H- how do you wanna land the plane here?
Ryan:Let's land it with AI burnout. How you feeling these days?
Brendan:Yeah, I, I think better in a lot, in some ways, I guess, but I have to think about that for a minute, 'cause like I, I do so much stuff with it, like sometimes it feels really productive working with it, and other times it feels like kind of what we were talking about earlier with prioritization and never feeling like it's working right, and then trying to get it to fix something that I don't even know how to fix. So that, that's the frustrating part. But like
Ryan:That's interesting
Brendan:Yeah. Uh, know
Ryan:I feel like I don't, like, have that reaction anymore. Or like, uh, very little do I feel like it can't point me in a, a direction that is worthwhile to explore.
Brendan:Yeah
Ryan:not always right. I'm not saying, like, it always gives me the, like, the best answer.
Brendan:Mm-hmm
Ryan:there were times where it used to be like, "Do this." And it's like, it would try to do it, and it would just be completely wrong. Like, what just happened? Right? Like, I don't run into that anymore,
Brendan:I,
Ryan:don't think. But I'm not su-
Brendan:Especially around code. Like, I couldn't tell you the last time it wrote code that I was like,"Oh, this is completely wrong. Like, what are you doing?"
Ryan:Yeah
Brendan:not the case. It's more that prioritization of things, that it too much information and not enough context kind of thing. but
Ryan:Well
Brendan:though, how's it going with being basically the sole developer and that many pull requests open? Like, how's that affecting you? How, how are you feeling?
Ryan:Uh, I'm fine. Uh, I'm good. Like AI is not the thing that stresses me out.
Brendan:Okay, that's good
Ryan:you know, I feel You know, I'm-- There, there are nights that I'm tired and I don't wanna look at the screen anymore.
Brendan:Yep
Ryan:still, like, just always… I, I love figuring out problems and how to build them and plan it and execute, right? So that, you know, no shortage of that for me, right? And I'm totally fine with that. The, you know, the stress of, um, you know, I think now meeting new expectations is what gets me, right? So, you know, I, I can both move much faster and do much bigger things, but the expectation is now that I move that fast and do bigger things all the time.
Brendan:Mm-hmm.
Ryan:Um, and if anything's gonna burn me out, it's, it's that, right? But AI itself is not the cause of that. That is, like, you know, human-to-human interaction and, you know, uh, HR
Brendan:Yeah
Ryan:stuff, right? Um, you know, so just being better at managing expectations and making sure not to overpromise things, um,
Brendan:Right
Ryan:always been the case. It's just easier to overpromise now,
Brendan:Totally. Yep
Ryan:yeah, I, I don't know. Like, I, I don't feel I don't feel burnout from any of the AI stuff I'm doing. Uh, you know, with my publications, I just closed the winter issue, and I was ready for a break, Right? Like, and, and that was… You know, I was doing sales, you know, I had the number I had to hit to, you know, cover all the expenses for the quarter. And, you know, there's sort of a, there's a longer gap between winter and spring, right? So, like, I have to stretch it a little bit further.
Brendan:Sure
Ryan:kinda uncomfortable, especially if you don't hit your number. And I was just like, I got to it. I got to within 1,000 bucks of the, the, the number I needed, and I was like, "You know what? I'm done. Can't-- I don't have anything left for this quarter." I needed a break. So, like, you know, I've got burnout, but I don't think it was… It had nothing to do with AI. It had nothing to do with development. Um, and, you know, I've been able to refo-- you know, I've, I've taken a break. I don't need to do sales for a couple weeks. And I've, I've shipped some big features this week that were… have been outstanding for a while, and that felt really good, and I'm gonna ship another two or three in the next two or three weeks before I have to jump back into sales. So I'm feeling good. Um, I don't worry too much about the future of AI, um, or what it means for the economy, the market, the jobs.
Brendan:Yep
Ryan:I think it's all gonna sort itself out like it always does.
Brendan:Right
Ryan:but I'm still pretty bullish on what it can do and, you know, if, if anything, I still want to explore more. I wish I had more time to do that. But, um, I've also just had to be like, "All right. No, I've got, I've got two businesses. I got a family." I don't have a lot of extra time to explore new things, even though I have, you know, a bazillion ideas a day. But I, I made a choice. This was pre-AI. But I, I think I made a choice about 18 months ago, and it's like, I'm gonna stop chasing side hustles and just focus on what I got.
Brendan:Yep
Ryan:Um, and that has its trade-offs.
Brendan:B-
Ryan:But yeah
Brendan:narrowing down the AI to a couple lanes, like it's, it's good to do
Ryan:Yeah
Brendan:self too, for, for sure
Ryan:Yeah, and that was You know, I, I think part of that is circumstantial, right? The publishing business, you know, fell into my lap for, you know, a set of circumstances that I wish never happened, but it's there and it provides a decent amount of income, right? So like that clears, you know, you know, a big burden right off of my shoulders, right? Like I was always trying to do side hustles or products or whatever because I was like, I, you know, I wanna supplement my income. I wanna supplement my income.
Brendan:Right
Ryan:that, you know, I have a good income situation that I'm comfortable with. I don't need to chase, you know, a $2,000 a month side hustle at the detriment of two businesses that are kicking off, you know, real money, right? Like now I, you know, I've got payroll, right? Like I, I have to pay people, uh, on both businesses, right? So there's stress in that,
Brendan:Yep
Ryan:right? But like, you know, short of a depression level event, you know, we'll be okay, right? So it's like it changes the dynamic and I, I understand that that is a, a fortunate spot for me to be in, so like I, I don't wanna sell it as something that it's not, right? Um,
Brendan:Yep
Ryan:but it certainly frees up a lot of mental space. Uh, you know, 'cause I can still play with AI and figure out how to make it work for those two businesses, and they're two very different businesses that have very different needs, so like it,
Brendan:Yeah
Ryan:keep, you know, keeps me mentally stimulated
Brendan:Mm-hmm. Yeah, that, that makes sense. Um, yeah, I don't have a whole lot else to add beyond just, like, I, I, I feel like that's a lot of what my side projects are, is my research. Like, I do a lot less research types of things with AI. I'll, I'll take things that I've learned doing it a side hustle, which, mean, maybe that's not even the right term since it's not making money. But, you know, I, I will learn things there that I will apply back to the job and less the other way around. I, I would say if anything, like, I get is more of trying to keep up with everything at the job because I'm probably asked to do too many things, and that's not even really AI related. Um, that, that's just the nature of the position I'm in. So AI has helped. I wouldn't say AI has, like, burned me out there, but I'm also… Like, I do rely on AI a lot for pull requests to keep up with the volume, a- and as far as, like, reviewing them and such. And when the
Ryan:I hate pull request review, so I get you
Brendan:Yeah, I never liked it anyways. Um, and this does a better job with the code side of it than I ever did, and it also helps me find edge cases and things. But sometimes it's like, well, did I need to find the edge case there? Like, that probably just wasted extra time that I didn't need to waste. Um-
Ryan:There's a point of diminishing return.
Brendan:Exactly
Ryan:Um, and you know, I try to Where I'm willing to spend the extra time is in making sure that the user experience is a good one
Brendan:Right.
Ryan:But I will still allow a bug to ship.
Brendan:Yep
Ryan:And, and that's sort of hard to say kind of out loud and, you know, transparently, right? Like an AI, you know, as you're, you know… Th- there's always a point in a pro- uh, you know, a, a session that I'll ask Claude to be like, "All right, review what we've done. What have, what have I missed? What are the edge cases? Like, are there, you know, check for bugs." Like, there's always that, you know, style of prompt as I'm getting ready to be like, "Yeah, okay, I think I'm, I'm done with the devel- you know, the feature development or whatever, right? Let's just start to, you know, stress test this a little bit."
Brendan:Yep
Ryan:And it inevitably comes back with like, "Oh, you didn't think about this," or, "You didn't think about that," or, "This is a real gap."
Brendan:Mm-hmm
Ryan:And what I've realized is like it's o- it's still okay to ship with some of that stuff, right? You have to analyze it and be like, all right, like I don't wanna expose personal information like that, that's a gap that I'm unwilling to, you know, compromise on, right? But, um, you know, if this is like an edge case that it's like you look at it, it's like, you know, maybe one person hits that in six months, right? Like
Brendan:Then
Ryan:j- notate it in the follow-up, I'll get to it, right? Like, and the truth is, I'm still shipping with way less bugs than I ever would've before,
Brendan:Right
Ryan:contextually, I have to remind myself of that, that it's like, all right, this is still better than where we were. It's not perfect, but it doesn't need to be. Um, right? But it's easy to look at it and be like, "Oh, well that's a, so it's a bug. Like I, I, I see that it's a bug. I should probably fix that." Well, no, you, you don't actually, right? Like assess it. What's the risk, right? Is the risk, you know, worth delaying the, the push and, and, you know, completing the project? If it is, great, then fix it. If it's not, then notate it and come back to it, right?
Brendan:Yep, exactly
Ryan:and sometimes it's, it's not a bug, but a defect. It's like, ah, that probably shouldn't work quite that way. Like, yeah, I can see there's probably a better way to do that, but like I'm not re-architecting this whole thing at this point,
Brendan:Yeah.
Ryan:right? Like it, it's fixable later. It's not gonna, like, if I push this in… You know, there's some bugs where it's like, or, uh, some defects where you, if you push the code now and then people use it, it becomes hard to change later. And like, I'll do that now, right? Like I-- But if it's like, ah, not a big deal, I can just implement that at another time. Cool. I'll do it later.
Brendan:Yep
Ryan:Let's, let's get some feedback and see how it works. You know, maybe it becomes irrelevant. I, I worked on a feature yesterday. It was like, uh, it was an attribution feature, and the edge case is simply that sometimes, uh, a thing could be attributed to more than one thing, and there's no way for, uh, a program to know, right? Like at the end of the day, the person would have to say, "Yes, it's for this," or, "No, it's for that,"
Brendan:Mm-hmm
Ryan:And you can make some good guesses and probably make it right, but the number of people that actually fall into that camp is like 1% of the entire code base that would have multiple attributions. Like, why am I spending time on this, right? Like I ultimately decided it's like, all right, you know, just give the admin the ability to move it,
Brendan:Right.
Ryan:right? That's an easy thing, right? It's just the belongs to, so it's like a very simple controller, and if they need to move it, they move it, and if they don't care, they don't care. And, you know, I can ship that, and if people are like, "Yeah, this is really not working, like it's always attributed to the wrong thing." Like, well, okay, well then I can do the bigger fix, right?
Brendan:Yep
Ryan:Um, I don't know. I'm rambling. Uh, we've been on long enough.
Brendan:Yeah, I was gonna say, I think we're, we're going on our longest episode
Ryan:Well, it was a grab bag. We covered a lot
Brendan:no, it, it's good. Good discussion. But yeah, may- maybe this is a good place to just wrap it up,
Ryan:Yep, we're just gonna cut it off cold turkey. I'll see you next week
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