New Year, New Ruby: Agents, Wishes, and a Calm Ruby 4

Valentino Stoll (00:01)
Hey everybody, welcome back to another episode of the Ruby AI podcast. New year, new us, right Joe? ⁓

Joe (00:08)
New year, new us. Yeah, I'm

Joe Leo. I'm your co-host and I'm curious, ⁓ Valentino, how did you celebrate Ruby's 30th birthday?

Valentino Stoll (00:24)
In Cloud code. Like everybody else, I'm sure. ⁓

Joe (00:29)
Yeah, that's the place to celebrate.

You get that little Martian on the screen and you you start feeling happy.

Valentino Stoll (00:37)
Now, you know, I was pretty excited to see all of the design stuff updating to kind of like It was definitely way outdated like you go search for Ruby as a language and you get to this like 90s 1990s style, you know website with just like raw links to stuff ⁓ It got it. got you the information, right? But it's just like you got there and you're like, you know if you're a kid looking at that page like in this day and age like

Joe (00:52)
Yeah.

Mm-hmm.

Right.

Valentino Stoll (01:05)
Is this something I should be using? Right? Like, you kind of got that feeling. And now I feel like it's in a much better direction. Like, it definitely looks like something modern, which, you know, 30 years, that's a long time.

Joe (01:08)
Right?

30

years is a long time. And this does say something about your co-hosts, if you're listening at home, that if you are in your 20s perusing the website, we say that yes, that means that you're a kid. That says something. I think here is another thing that says something. We talked just before we hit record about the release of Ruby 4, and nobody seems to really be paying attention to that. And I think at first I was disappointed by that.

Valentino Stoll (01:30)
You

Joe (01:49)
Um, of course I'm writing a book on, um, the Ruby programming language or we're updating the well-grounded Rubyist. And so it would have been nice if we knew a little sooner that things were going to go from three to four, uh, on December 25th of 2025. But then, you know, we started, we started updating the book and realized, okay, there's not, there's not a ton that has changed. And I think that was initially disappointing. Uh, the next, I went through like the three phases, first, disappointed second.

I'm like, it's all about Ractors and I don't use Ractors because I'm not a good enough concurrent programmer. So then I beat myself up a little bit. And then the third stage where I'm at right now is acceptance and saying, actually,

Isn't it nice that a major version of Ruby can be released and people are not freaking out in a good way or a bad way. They're like, okay, this is good. We've done this before and we know how to handle it. Nothing's gonna go crazy and break. We've reached this point of sort of relative stability in the Ruby programming language. What do you think?

Valentino Stoll (02:52)
Yeah, I tend to agree with that. I was just talking, I was at the Ruby Kaigi where they were like, for Ruby 3, and they were like, we're gonna make Ruby three times faster. And this was ages ago, 2015. And so I feel like,

Joe (03:05)
Yeah.

Mm-hmm.

Valentino Stoll (03:13)
The mentality then was, okay, people keep saying Ruby's not performant and maybe we should do something about that. And it was, it was super impressive. They did it. They made it faster than the previous version of 2.0 and ⁓ three times faster. of course, it actually got faster than that. Shopify dumped a ton of money into ⁓ making it, Rails even, way more performant than that. ⁓

Joe (03:21)
Mm-hmm.

Yep.

yeah.

Mm-hmm.

Valentino Stoll (03:43)
And so I feel like that definitely solved or not solved, but like, you know, if you're outside of the Ruby community and all of a sudden it's like, Oh, Ruby's three times faster now. Like it makes you stop and pause. you know, is, is it scalable still? Like, can we say these arguments anymore? Right. It makes people question and talk about it. And, uh, you know, I feel like 4.0 is more of like the stability like you're talking about, right? Like, uh,

Joe (03:57)
Right.

Right? Yeah.

Mm-hmm.

Valentino Stoll (04:13)
You know, can we get it, you know, a concurrent parallelism? Like, is that possible in Ruby? I feel like that's the other thing people say. Well, you can't do parallelism, right? Like, which you can't really do in Python either, or, or, you know, a number of other places, true concurrency, right? ⁓ So, I mean, I feel like

Joe (04:22)
It is the other thing people say. Yeah.

Yeah.

Valentino Stoll (04:34)
we've caught up at least to Python in that way, where we're finally starting to see, well, we can actually run the same kind of parallelism that you can in Python, which is like, you know, kind of huge ⁓ in that way. And hopefully it means more data aspect of things can start getting adopted in Ruby. ⁓ Yeah, but again, like ⁓ one of the biggest things I feel like

isn't talked about and maybe just because it's experimental or not, was this whole Ruby box feature, which I don't know if you saw, it like lets you scope all of the classes and constants into like a fenced area of.

Joe (05:14)
I did see it, yeah.

Mm-hmm.

Yeah,

and this was whispered about at ⁓ RailsConf. And I don't know exactly where I heard it. I started picking up on it from other people saying, this is gonna be a really big deal, which sounds, it sounds interesting ⁓ for, for example, library maintainers and library builders. But I'm curious to hear what your thought is on how that, how that's worked or how you see it working in practice.

Valentino Stoll (05:49)
Yeah, I the biggest thing I saw is like, and they're documenting it now of, you know, being able to run these boxes in parallel to execute like blue green deployment as an example, or like evaluate dependency updates, right? Like you can start doing a lot of like ⁓ code transitioning because you have these boxes that can work together. ⁓

Joe (06:13)
So the box

is not meant to remain as simply ⁓ namespace, ⁓ sort of guardrails, but actually to start to ⁓ become a feature of parallelization.

Valentino Stoll (06:28)
Yeah, I think that's kind of the hope is that these things can start become their own like, you know, process level, like, adaptation. And hopefully it evolves too into like a wider, you know, ⁓ binary distribution, right? Like, I feel like right now you got to use these third party, ⁓ you know, bundling and packaging services in order to like make your Ruby binary, you know, a thing so that it can actually compile all of the Ruby source.

Joe (06:44)
Yeah.

Mm-hmm.

Valentino Stoll (06:57)
that you need for it, right? Into that single executable. Yeah. ⁓

Joe (06:59)
Right. Yeah. And I seem to remember some kind of

a controversy over those packaging services last year. I don't know. It was a year ago. We don't have to talk about it anymore. But yes, go on.

Valentino Stoll (07:11)
Yeah, I mean, that's like a whole other topic, right? Like, how do you even do that? The latest I've seen was Brad Gessler. Gessler? He does the terminal wire stuff. And he was like, oh, and I use this tobacco or tobacco packaging library and it works great, like cross platform, know, Ruby.

Joe (07:26)
Mm-hmm.

Right.

Valentino Stoll (07:37)
binary distribution and it works great. It works with like brew, homebrew and like apt and it just like makes it all really easy to do. ⁓ But again, it's like a third party, right? Like it's not like the Ruby core team, like saying, hey, this is how you package binary and distribute a Ruby app, right? Like if you wanted to make a Claude code, I just had this problem. ⁓ I made a Claude memory Ruby gem that uses like SQLite and just like the full text search so that you can like,

Joe (07:39)
Mm-hmm.

That's right.

you

Valentino Stoll (08:07)
just hook up ⁓ into the hooks of Claude and get it to have long-term memory. ⁓ And ⁓ I basically used OV Fernandez's recent article on his Nexus project. I was like, and I found another person had a competing article on ⁓ how ⁓ time should be considered for these entity graphs. ⁓ And so I just combined all the ideas together and made this thing. ⁓

Joe (08:11)
Yeah. Ooh.

Mm-hmm.

Valentino Stoll (08:36)
And then I start up Claude and it's like, I can't find the executable for like your Claude memory gem. And I'm just like, ⁓ great. Like now I have like path problems and like, you know, how does Claude, right? Find ⁓ on the, it's path, right? Like the executable for the Ruby gem. And I'm just like, this is way getting way too much.

Joe (08:43)
Uh-huh.

yeah.

Mm-hmm.

Yeah.

Valentino Stoll (08:57)
And

so I'm like, all right, how do you distribute a Ruby binary, right? So like somebody could just brew install it and then that's part of like the setup script. And then Claude can just, right, reference the binary on the system. ⁓

Joe (09:07)
Right.

Yeah. And this is a non-trivial

problem. I actually did this a million years ago. I was working at Credit Suisse. We could talk about it now because Credit Suisse isn't really around anymore. But ⁓ I had to write C sharp code all day long. It was not fun. And I did this for a couple of years. ⁓ But we were building a desktop application. so ⁓ we thought, okay, well, in order to, at my boss's request, we needed a way to skirt around

⁓ the teams that were responsible for approving and moving code to staging and to production, basically to avoid DevOps so that we could go to production really quickly, which is not okay and not allowed, but he's my boss, so I said, yeah, let's do it. And I said, here's how we could do it. We can have a tiny little web server running in the background and it can look for updates. And if there's an update, then when the person fires up the little desktop app, we'll go, we'll update it before the thing opens.

And we're done. So of course I wanted to use Ruby. ⁓ Ruby was not allowed at CreditSwiss. It was not approved by security. So I had to install it myself and I had to install it like library by library, like all the different dependencies. And it's not trivial, you know, even on, I think that was like a RHEL server.

Valentino Stoll (10:15)
Hahaha

Right.

Joe (10:28)
Right? So on a little rel box or a big rel box, I'm I'm like SSH again, and I'm trying to install like one dependency at a time and one DLL at a time and like try to fix like the different stuff that breaks and see. Yeah, it's, it's, ⁓ it's non-trivial. And then when you get to this, and that was, that was years ago. And now when you get to this point where there's, are different compilers, there's different, there's, you know, things changing all the time, that can be real challenge.

Valentino Stoll (10:56)
Yeah, yeah, and luckily, you know, I don't know if it's Bundler solve this, but you can like include precompiled binaries for like, you know, system level dependencies too in your gems now, which is really great. And that solves a lot of problems, right? Like if you want to use SQLite, right? In your Ruby gem, like, and it's a system dependency normally. So you'd have to brew install it. You can actually precompile it for like any platform you want to distribute to you. It's amazing.

Joe (11:09)
You can now, yeah. Yeah. Yeah.

Yeah, that's pretty amazing.

Yeah, I mean, for the people that worked on that and released it, thank you. That is a huge win. Yeah, yeah. It's so much thankless work that goes into that. But we're thanking you on this show, so that should be something. It's worth something.

Valentino Stoll (11:25)
⁓ And so that solves a lot.

Right, so much. ⁓ I don't know how many times... It's so much. I don't know how many

⁓ MySQL and Postgres related extension library errors that I helped resolve over the years. ⁓

Joe (11:50)
Yeah, my god. Yeah, yeah.

Valentino Stoll (11:56)
Yeah, wow. So along these ideas, you we were talking before the show, like, you know, there's a lot of wishes out there, like, what we what we wish, you know, Ruby had, ⁓ what we wish the community had available to us, you know, as the AI community is just like exploding ⁓ across all languages, you know, like you try and get onto the frontier, so you stay ahead of the game. And it's very hard because

Joe (12:19)
Yep.

Valentino Stoll (12:26)
Cloud Code has made it very clear that they're a node shop and JavaScript shop or TypeScript. ⁓ And so how do you fit, how do you get to work with the pleasantries of Ruby and get to like read through like some nice abstractions and also fit into that like new frontier? ⁓ And so I have a pretty long list myself, but ⁓ what are your thoughts? What are your thoughts for us?

Joe (12:30)
Yeah.

Right.

Good. Good. Let's hear it. Yeah. I think we should, I think we should

jump into it. I have, ⁓ I have, I have thoughts, I have takes, but you've seen it, you've obviously spent, spent a little bit more time on this. So let's let you kick it off.

Valentino Stoll (13:04)
Yeah, I mean

Yeah, I'm talking about our ⁓ binary distribution. That would be great. ⁓ Even just having it so that you can gem install something and it works with one of these agents, right? That alone would be great. Which I can in some ways, but not out of the box. ⁓

Joe (13:15)
Yeah, yes.

Mm-hmm.

I have this interesting, so I'm wondering if, I had a wish and I think, I'm wondering if it has been granted by, ⁓ for me, an unlikely place. So one thing that has bothered me, so at our company, we've got this split and a bunch of people use either Claude code or codex on the command line, but there are a couple of people that still prefer something like windsurf.

And I liked Windsurf for a little while and I've quickly transitioned over the command line. I think it's much better. But I get wanting to have AI in the IDE and I just don't think it's done. I mean, this is where we get into like last year versus this year. Last year, when I tried Windsurf, it blew me away and I was telling everybody I knew about it for like three months. And now it's a year later. I'm like, it doesn't really do it for me. You know, that's not really fair to Windsurf or anybody else. That's just the reality of ⁓

This is reality of AI software development. here's an interesting thing that has come out. And I'm wondering if you've tried it. Have you tried Steve Yegge's Gas Town?

Valentino Stoll (14:44)
Gas Town, there be fires. There be fires in Gas Town.

Joe (14:45)
So what's that? Yeah, that'd be fires. Yeah. Yeah. ⁓ This

is a, ⁓ it is an IDE ⁓ that I'm quoting here, helps you with the tedium of running lots of Claude code instances.

Right? Things get lost. It's hard to track who's doing what. Gastown helps with all of that yak shaving and lets you focus on what your Claude codes are working on. So this is interesting. I never know if I'm reading Steve Yegge, if I'm reading something that he's being earnest about or sarcastic. I remember reading an article last year where he was extolling the virtues of ⁓ AI prompt engineering. And I had to read the article twice. I didn't know if he was...

trying to pull a fast one on me. But I think he was earnest then, and I think he's being earnest now. And I haven't tried this yet, but I think that is a big deal when it comes to software development ⁓ in 2026, is being able to actually manage what you're doing across multiple cloud code instances. What do you think?

Valentino Stoll (15:54)
I am so torn on this, mostly because this is extolling one-shot prompting as a virtue. ⁓ And so, yeah, for those of you who don't know one-shot prompting, it's really just saying, go do this thing, LLM, and it goes and does it and completes it to finish. ⁓

Joe (15:56)
Yeah.

yeah, fair enough. Yeah, let's talk about it.

Mm-hmm.

Valentino Stoll (16:21)
You know, that's more so the true, the more number of things that you add into the puzzle. So, you know, Gaston as an example is like, ⁓ you know, a bunch of agents then working to complete that one prompt that you gave it. ⁓ and obviously like the prompt that you give it matters a lot. And it gets more into like this spectrum and development, you know, we've talked about before where you, the specification that you give or the plan that you give to it is like the most important then.

Joe (16:26)
Mm-hmm.

Yeah.

Mm-hmm.

Valentino Stoll (16:51)
Right. And so then you end up like writing these very descriptive documents that plan and outline what you're trying to build. A problem I have with that is like as a builder, like the design changes the more that you like work through it and work workshop it. And so like, ⁓ as you start to like, even as you're starting to write out or have an agent write out the things you may start to notice.

Joe (16:57)
Right. Right.

Yes.

Valentino Stoll (17:18)
Oh, hey, this is like not going in a direction that I want, like the design would be better in this way. And even if you introduce an agent that sits there and observes and can help redirect, even still, it may not redirect it in a place that you want. Right. And so, you know, managing, like, do you want to manage all of these agents and how they decide to do things?

Joe (17:24)
Mm-hmm.

Yeah, absolutely.

Valentino Stoll (17:46)
Or do you want to like design things? Right. And you have to pick one. And so like the one shop to prompting approach is really like, okay, like, do I want to learn and set rules that define how, how the plans should work best, right? Like prompt engineering or context engineering, whatever we want to call that now. Right. ⁓

Joe (17:48)
Yeah.

Mm-hmm.

Valentino Stoll (18:11)
Do I want to focus on that and let the things get better? Which a lot of people seem to like that idea. ⁓ Or do you want to make things more of a sidecar fashion and just help you? ⁓ And I feel like it's hard because ideally the things will just do the things you ask them to do.

Joe (18:23)
Yeah.

Valentino Stoll (18:31)
and they'll do them in the right ways and as you expect and it's a golden paradise, right? Of these agents doing right, right? ⁓ Even with humans, You can have some brilliant people, right? And they're working with you and it's not gonna be as you expect because everybody's individual. that's sometimes a really good thing. exactly. And I mean, that's perfectly fine. But in terms of an agent,

Joe (18:32)
Hahaha

Yeah, yeah, I run a company of people that I hope will do what I tell them to do. It is a lot of parallels. Yeah.

No, and sometimes that's a really good thing, right? Because what I expect is not ideal. Yeah.

Valentino Stoll (19:01)
you know, we have expectations that it's going to do what you tell them to do. Right. And so like, if you start to break apart and fragment, like the things that like are trying to do the things that are being told of them, I feel like that's just like, ⁓ it's like pouring gas on the fire.

Joe (19:13)
Mm-hmm.

⁓ Well, yeah. that,

so, okay. So this brings up an interesting, I think just tactical question. I think people are curious about this. You've got a, you've got a very good job. You are working with a lot of legacy code. When you fire up Claude code, you're pointing it at a branch on, you know, some, some stack in, ⁓ in gusto. What, ⁓ what are you doing? How are you going to, how are you going to use Claude code? Like what is, what's the steps in your process?

Valentino Stoll (19:48)
Yeah, I guess it depends on the thing I'm trying to do, right? Like, so, you know, as software engineers, you have like, you know, bug fixes, have refactorings, sometimes you have like new features, sometimes it's a blend, right? Sometimes you're adding something new and you realize that something existing needs to be refactored to reshape and work better with the new thing that you're adding. ⁓ So you get kind of like these staged approaches ⁓ and each use case is different, right? And so like,

Joe (19:59)
Mm-hmm.

Mm-hmm.

Valentino Stoll (20:16)
⁓ some bugs that are really small like you can get a code to like investigate what happened and create a test that verifies that that actually bug does exist and the test is failing and then you have another thing go in and fix that test to be you know passing and then so you can you can honestly have some like bug fixes just be odd fully automated right

Joe (20:41)
Mm-hmm.

Valentino Stoll (20:41)
where

something is going and investigating and documenting and creating a test and getting the test passed and pushing the fix, right? Maybe not pushing it and deploying it. Like I don't personally do that, like full automation, but you get to the point where something, right. Something can be, you could point, you know, an agent at this bug that's very well defined, it has a stack trace and you could say, hey, you know, here's data dog that, you know, had this error and the stack trace and.

Joe (20:52)
Right. But you can make a PR. Yeah.

Valentino Stoll (21:10)
you know, go figure it out, right? And then you can go look and see, ⁓ like, you know, ⁓ in my personal case, I don't like to automate the solution. So I'll go and I'll, I'll just have like, Hey, go figure this out. And then let's work on a solution together. Right. And so I'm a very much like working with the, agents. Right. ⁓ and so I'll do that. And so that's kind of like my workflow for a bug fix.

Joe (21:10)
Mm-hmm. Yeah.

Yeah, I get that.

Mm-hmm.

And

so for you, one agent is really enough because you're giving small incremental steps and then you want to have that kind of feedback so you can continue working in a collaborative fashion.

Valentino Stoll (21:48)
Yes and no, right? Like, cause then there's also like the, agents that like could listen in on a pull request and you can just tag it in and be like, review this thing. Right. which is another use case. or you have like, ⁓ you know, sub agents that can tie into Claude hooks that can do things. Right. So like, if you touch a certain file, you could say, Hey, go run, migrations on this thing. If you're doing a database, right.

Joe (21:55)
Hmm.

Okay, yeah.

Mm.

Valentino Stoll (22:16)
or ⁓ make sure that like, this isn't going to break, you know, production if it's like got bad indexes on it, like look at the use cases, right. ⁓ so you're at, so as you're like, you can, you can get like agents doing work kind of in the background that can help assist you as well, but really they run on their own, right? Like they had these like preconditioned list of things, that trigger them or like you can trigger them manually. ⁓ but it's like,

Joe (22:27)
Right.

Valentino Stoll (22:46)
Yeah, still I'm kind of somewhat in control. Right. I've tried, I've tried doing the, so I guess I'm a little biased because I do have this project, the AI software architect where there's a bunch of different members, which are kind of agents that act as certain roles, like a backend engineer or testing engineer. Right now it's just like skill driven. So it's like,

Joe (22:50)
Right, yeah.

Yeah.

Valentino Stoll (23:16)
really just markdown files. ⁓ But they are kind of like their own agents. And I'll ask them to do like, hey, review this, the changes that I have here as a collective and tell me if it's any good or what changes you would make, right? And so they'll actually, it'll actually go through and Claude will like pull together the individual members and review them with their own personas and what they think about the different changes. And they'll create a little report for me.

Joe (23:17)
Mm-hmm.

Yeah.

Yeah.

Valentino Stoll (23:42)
And then tell me like a summary, yeah, these look good, or you may want to think about this, right? And so there is like, you know, ⁓ Swarm by Perudo at Shopify, right? ⁓ And he has this like way of doing it with like actual agents that are, you know, have their own process and they work together ⁓ as a collective processes, kind of in similar to ⁓ Gastown, but not as much.

Joe (23:45)
Yeah. Yeah.

Mm-hmm.

Okay.

So they send

their information back and forth, right? So one person finds one person, one agent finds this, they send it to the other agent, right? Yeah, I like that. So.

Valentino Stoll (24:14)
Right. Yep.

Yep. Yeah. And then they have

like a communication layer. ⁓ you know, I don't know if you've seen the like Claude mailer or someone just made like a male skill and all the agents have to communicate with each other over mail.

Joe (24:32)
⁓

no, I didn't see that. I like that. Yeah. But I think what you're hitting on something that is near and dear to everybody's heart, is waste and efficiency as programmers. I think, so you have this way that is...

Valentino Stoll (24:34)
⁓ that's pretty funny. ⁓

Joe (24:53)
that if I can reframe it, right, you've got a way of working that is sometimes sort of one-on-one collaborative and sometimes gets into a point where you're going to get some ⁓ agents that are going to work fully through a problem. But you're really maintaining the control of both what, ⁓ of synthesizing what the agents go and learn, and of course of what ultimately gets put back into your repo. And I wanted to highlight this, ⁓

This is a report from IT Pro January 16th of this year. So just a few days ago, a survey of more than 1100 U.S. enterprise AI users from Zapier found that while 92 % of workers say AI boosts their productivity, the average employee spends 4.5 hours a week revising, correcting, and sometimes completely redoing AI slop.

And so it begs the question then, okay, so are you actually more efficient? Do you only feel more efficient in the moment? But then over time, you're losing that time. think about, you just mentioned email. People don't realize how much, people in my job, engineers aren't good with email and I think that's really a testament to how much better that job is than mine. ⁓ But people in my job who have to communicate with people all day long.

Valentino Stoll (26:10)
You

Joe (26:18)
⁓ If you're running a company, they don't realize how much inefficiency there is in something as simple as email, right? And you really do lose time. You lose tons of time to adjust things that you think are taking two seconds, but they're actually taking five minutes and it adds up. And I kind of compare this here. Engineers are moving faster. You can feel the velocity of your movement. And yet at the end of the week, you're giving half a day or more back to the fact that you

you or somebody on your team moved too fast and created problems. And then you have to look back and say, okay, well, would it have been better to just go slower? And if you were going to move slower, how would you do that? Because we're not saying don't use AI, everybody's using it, but how do you use it most efficiently?

Valentino Stoll (27:04)
Yes, funny, you know, have a meeting later today to solve a lot of these or to try and solve a lot of these problems. ⁓ Yeah. You know, it's hard because, ⁓ you know, like we said, everybody works differently, right? Everybody has their own processes. Some people use different editors, ⁓ you know, even more abstract than that is like, you know,

Joe (27:07)
Haha, yeah.

No, no, solve them and then we'll record again tomorrow and we'll tell everybody what the answer is.

Mm-hmm.

Valentino Stoll (27:31)
everybody thinks differently, right? And so like, you can't force everybody into a box and kind of like that was the alert I think of Gastown is like, well, the box, let the box think, right? ⁓ And everybody just like talks to it and it goes and does work. ⁓ And the problem is like that doesn't work, right? Like, so like you're saying, you know, people do, you I personally like it to explore new ideas, right?

Joe (27:33)
Mm-hmm.

Yep.

Mm-hmm.

Yeah.

Mm-hmm.

Valentino Stoll (27:58)
where I'll go and I'll just say, what do you think of this? Like go implement it, right? And I'll set up a Docker container and isolate it and just let it YOLO, right? And how far does it get? ⁓ And it helps me think about the design faster because I can see, well, I didn't like this, I didn't like this, I didn't like this. And then I can reframe and create that specification better so that if I did that again, it would be valuable, right?

Joe (28:02)
Yeah.

Mm-hmm.

Valentino Stoll (28:25)
And so it's nice to be able to do that kind of work in parallel and not have to think about it in the moment until you actually get to needing to do it. Right. And so there is a lot of like value in it, but I think it's like, you can't use it as your work. Right. ⁓ yet. And so like, ⁓ you know, you go into, you do tell it to do something, you have to be very specific, ⁓ or you're not going to get the result that you want. And you have to have a lot of guardrails. Right. So like.

Joe (28:42)
Right. Right.

Mm-hmm.

Valentino Stoll (28:55)
⁓ For cloud code, I use cloud code almost exclusively now. And I have a bunch of like plugins and stuff that help like add the guardrails, but also like help reframe things for the work that I'm doing. So, you know, I'm within a certain project. It has a bunch of rules set up and it knows a bunch of information about it. Like that's very specific to the project. And like the output ends up becoming better the more I use it the more I add to it ⁓ that I do end up getting more value the more that I use it.

Joe (29:00)
Mm-hmm.

Mm-hmm.

Valentino Stoll (29:24)
Right. And so I feel like that's kind of where things are moving and it's a slow process, right? Cause you're, you're ultimately having to train this thing, how you work, right? And how your project works. ⁓ and so I'm suspicious that a lot of this, like a lot of the naysayers of it's just generate slop is really just people haven't used it long enough. And that by the time that it gets trained over, it probably takes a solid year of use, right? After that time.

Joe (29:25)
Sure. Yeah.

Yes. Yeah.

yeah.

Valentino Stoll (29:53)
of putting in the hours, it starts to become a ⁓ multiplier of your time. ⁓ Because you know specifically how to use it, but also the thing truly works better for your specific use case.

Joe (30:08)
Okay,

but here's where I'm going to disagree with you. The place where I'm going to disagree with you is that software engineers have always written slop. That's why they're slop in your AI-generated code. It's not coming up with stuff out of nowhere. And so you're exactly right that a lot of times it's a misuse of AI that gets you that code slop. But that doesn't mean that, I mean, would take, in my opinion, would take

both the engineer and the AI to solve the problem. Your claim is that the AI will get better. I absolutely agree. That does not mean that the software engineer is going to get better. ⁓ Yes, they have to. And yes, think along a long enough continuum, maybe they'll get better. But I think you're always going to have people that are going to make those kinds of mistakes. And thus you will you'll continue to be in this position. And I think it's interesting that this article takes place. The survey was in the enterprise because in the enterprise, you know, I've been there. ⁓ There's a lot of waste.

And there's a lot of ⁓ non-productivity. There are way more laggards than there are people at the cutting edge in the enterprise.

Valentino Stoll (31:18)
Yeah, and two, you don't want too many people at the cutting edge at your enterprise all at once, right? ⁓ Yeah. And so, mean, that's honestly a good thing about enterprise is that it's like kind of forces you to move slower in some ways. ⁓ But, you know, it's hard because like there's currently actual value you can get, right, out of using these things the right ways. ⁓ And so like, how do you get

Joe (31:22)
No, no, I know it's by design. It's by design. Yeah.

Mm-hmm.

yeah.

Valentino Stoll (31:48)
I guess it's more of how do you get the people that do know how to use it right to teach everybody or make tooling or other things so that everybody else can use it optimally for their own use cases as well. And I feel like that is definitely, that's one of my wishes, right? More normalization of the tooling, right? ⁓ Right now it's very much like, here, this is how you do memory. Like, you know, a month later, like somebody else is like, nah, really you want to do this for memory, right? Or.

Joe (32:16)
Right, right.

Valentino Stoll (32:16)
⁓

Here's how you do like, know, Keurin from Every, right? Like he's like, hey, you know, context engineering ⁓ or what was it? What is the compounding engineering, right? Where you have these tools that like compoundingly get better the more you use them. And that's like, he's built this whole library of things to make that easier to use. ⁓

Joe (32:32)
Mm-hmm.

Valentino Stoll (32:44)
Is that normal though, right? Like, that the way everyone should use it? I don't know. You know, I've used superpowers for Claw, which is a very popular tool for just like normalizing workflow with it. Sometimes.

Joe (32:46)
Yeah, right.

Mm-hmm.

Valentino Stoll (32:57)
It gets in the way, right? Like sometimes I'm like, I just want to fix this bug and it'll try and make a plan around it. I'm just like, I don't really need a plan. And so like I go turn the plugin off, right? Cause I don't want it to get in the way. And so like the normalization isn't quite there to like solve all the use cases. And so that's something I kind of like am missing out of like, cause you don't want to create your own solution and then have

Joe (33:02)
Right.

I say, yeah. Yeah.

Yeah.

Valentino Stoll (33:26)
a new normal come out and then have to maintain the new solution or migrate to, right? It's like a whole like, ⁓ you know, software adoption problem all over again of, you know.

Joe (33:36)
I think that's true. And then

there's also the sprawl of it. ⁓ You I think about, you you said to use ⁓ Claude code for everything. know ⁓ plenty of people that are on that boat. I know some, not as many that are on the, the open AI codex boat. ⁓ And it's a little bit. So I think about it. It's a little bit like, ⁓

you know, Emacs versus Vim, but where it differs is that Emacs and Vim have been around and solidified themselves ⁓ for so long that if something came out for Vim, even for modern programming, ⁓ you can bet that the same thing's gonna exist at Emacs. It's probably gonna exist the next day, right? ⁓ And that's not where we are. not at that kind of parity with these tools, I don't think.

Valentino Stoll (34:14)
Right?

Yeah, totally. you know, everybody's got their own marketplace now, right? It's like becoming the competing, right? It's like we had, at least it's not a Walt garden. That's all I have to say. Please keep it open. Like make it accessible. Right. Like, but I feel like we're like kind of getting back to the marketplace, you know, competition. Right. And like, ⁓ maybe that's a good thing, but like it does like,

Joe (34:27)
Right, right, yeah.

Yeah. Yeah.

Valentino Stoll (34:50)
all the standardization, you lose out on so much because it prevents adoption and the slop creeps in, right? And to your point, like, yes, if you have like, some bad quality code, it could influence it. But we're almost to the point where like these tools are kind of like, ⁓ helping to solve that and prevent that from happening. Where if you get the right workflow set up,

Joe (34:59)
Mm-hmm. Mm-hmm.

Valentino Stoll (35:17)
⁓ Even if it references a file that includes some like bad practices, ⁓ all of your other contexts that you're including while you're working with through your code, it like negates it. And it says, ⁓ I see that we're using it this way, but we shouldn't be using it this way. Right. And like, you can get some prompting into your workflow where you can help prevent a lot of that slop. I mean, most of it, to be honest. ⁓

Joe (35:21)
Mm-hmm.

Sure, sure. I'm

with you there, but then you still have to, there still has to be that person who's, who's going to say, you know what? I think this is a good idea. I'm going to coax my AI or I'm going to allow it to coax me into doing something net positive for my application rather than just ram through whatever feature or bug fix I'm trying to do.

Valentino Stoll (35:57)
Right.

Yeah, I mean, personally, what I see happening is honestly more effort on the tooling, more people building up all of the things that make it more effective ⁓ and people spending more time doing that. So like all the time that, you know, we're saving, right. With using these tools ⁓ and going so fast is really just getting augmented to make the tooling better. And we're not quite there to like, you know,

Joe (36:08)
Hmm.

Yeah.

Right.

Valentino Stoll (36:29)
truly get the value that we, that executives hope, right, of it until all this tooling is complete, where it's like very straightforward, you use this and you will be this much better, right? And I'm suspicious that that will get there ⁓ in a year. Like, you know, what was that, Cloud Code was last year?

Joe (36:34)
Mm-hmm.

Right.

Yeah, well, that's a fair point. ⁓

Claw code was last year. know,

I know. The thing that everybody uses now, that question came out about a year ago. I keep talking about that to people. ⁓ But it's interesting to me that, ⁓

say that that yeah, we're the story of AI so far has been well, you just got to you just got to spend the money. You just got to invest in it. You just got to invest in it because it's going to give you these great returns. And I'm not here to say that it's not giving us great returns. I'm not I'm an AI optimist. ⁓ But I will say that the amount of investment you really can't predict what the return is or when you're going to get that return.

And just like you're saying, well, I don't know if the tooling will be there for a year. Well, for a small business like mine, a year's a long time. ⁓ you know, maybe in the enterprise you could say, all right, well, we've got to fund this thing for the next year or two years or three years. It's another zero on the budget. ⁓ for people that are really trying to harness this today, I think you've really got to find the people who are willing to put the time in to say, okay, I know that this is a scattered mess of tools and there's plugins everywhere and I need to know where to find them all. And, ⁓

and what the best ones are and how to vet them and how to plug them into ⁓ my workspace to make sure that I am utilizing these things effectively or.

You got to have somebody do that and then create a standard across your organization, which as we've already discussed is very difficult. I had this ⁓ meeting with a co-founder of a company we've been working with for years. And it was so funny because he was, he was laying out for me, this is fall of last year, laying out for me what he wanted, like a new sort of organizational model for his PMs, business analysts and engineers. And ⁓

He said, I think it should be, you know, it should be one PM with one engineer and a business analyst. And we just kind of move ourselves into pods so we can rapidly conduct these experiments and we'll use AI. And it sounded really good in the telling. And I, I brought that to the, to the ⁓ head of engineering, like the next week. And this guy, he was just like, yeah, that's great. But we can't even agree which

which tool to use. Like there's no way we can possibly, like he wants to standardize this across the entire organization. The entire organization is like made of AI skeptics, AI enthusiasts, people that love Claude code, people that want to use it in the IDE, people that just want it to check their code, right? And so the idea of standardizing at that point to this person seemed crazy. And I get that. ⁓ I really do. It's a hard decision to make, especially when you have limited resources.

Valentino Stoll (39:38)
Yeah, totally. I mean, you make a great point. ⁓ There's no answer for them either. It's not like you can go out and like, yeah, right. mean, people hope like, okay, I'm going to buy cursor for my team and then they're just going to be like so performant. And there's like so much like setup and configuration you got to do to make that true. ⁓ That it's like, it prevents people from buying it, I think.

Joe (39:45)
No, I kind of just laughed and said, well, you guys will figure it out.

Mm-hmm.

Yeah, yeah, yeah.

Mm-hmm.

Valentino Stoll (40:08)
⁓ right. And so that's what I, that brings me back to another wish, right? Is, what would be great for, for the Ruby community specifically is to have solutions to specific problems that are well packaged and standardized. Right. So you want, as an example, to, ⁓ write idiomatic Ruby, right? Like that's what you want. Like, ⁓ you just go and you use this plugin, right?

Joe (40:08)
Mm-hmm.

Yeah, let's hear it.

Yes.

Mm-hmm.

Valentino Stoll (40:36)
⁓ Or you want to ⁓ use our spec using the better specs specification, right? ⁓ go use this plugin, right? And like very specific use cases, but even more abstract than that, right? Like, ⁓ I want to, you know, going to the ⁓ jumpstart rails, right? Like, ⁓ I want to sell subscriptions, right? And so I have a, I want a web app that does that, right?

Joe (40:42)
Yeah. Right.

Mm-hmm.

Right, right.

Valentino Stoll (41:02)
Well, here's a plugin or a tool and you just use it and it'll generate the subscriptions, right? ⁓

Joe (41:03)
Nup.

I love that idea. Yeah, I thought that's where you were going with that. Yeah, because

it doesn't have, we don't have to start with the bare bones anymore. It doesn't have to just be rails. You can have 500 different kinds of rails apps now and you just pick one. I totally am with you on that. Yeah.

Valentino Stoll (41:17)
Right. maybe

Reels isn't the right solution for every problem. Maybe like using. I know it blasts me. ⁓ You know, maybe maybe using something like Jeremy Evans ⁓ tools. Right. Like ⁓ I'm trying to remember his Web server. ⁓

Joe (41:21)
We're gonna cut that. We're gonna cut that. Yeah. ⁓

Mm-hmm.

Valentino Stoll (41:39)
I haven't used it in long enough.

Joe (41:42)
⁓ Quinto.

I never used it.

Valentino Stoll (41:47)
Quinto, that's new. Rhoda, Rhoda. So Rhoda's kind of like the middle ground between Rills and something like Sinatra. And maybe something like that would be more performant for a specific task. And maybe something, maybe using the dry suite of...

Joe (41:49)
Is it not that? Rota.

Right, okay.

Valentino Stoll (42:13)
framework gems would be better used for very specific tasks, like where you need some kind of definitive business logic that goes through a well-defined set of processes, or Trailblazer, or any of these other existing Ruby projects that have been curated over the years that are great at solving very specific problems, like just making something that makes it easier to use those for the specific context.

Joe (42:14)
Mm-hmm.

Yeah.

Yeah.

Valentino Stoll (42:43)
⁓ I feel like there are so many wishes in there, but it's really one idea of just like making standardization out of these tools so that it's easier to use them.

Joe (42:56)
You know, I'm going to I like that idea a lot and what what you suggested actually made me think of something that when we had Obi on the show, it was a while ago now, but he had said something like, well, ⁓ it's still going to be really valuable to have basically my brain on on a particular problem, right, where ⁓ I have to agree with him. And this is a hokey sounding thing that I still think would be valuable is like

You said, well, I want the, the plugin that gives me idiomatic Ruby. Well, if that's the case, then I want that thing to be like the Sandy Metz code whisperer, right? The OB Fernandez, you know, the Justin Searles code whisperer, because when I read, mean, I'm always looking at these articles for, okay, here's how I set up my, you know, here's how I set things up. Here's what I, you know, here's what I say to Claude and here's what I, you here are my markdown files, my instruction files. I know those things are rapidly changing and evolving.

Valentino Stoll (43:35)
yeah.

Joe (43:53)
But I would still want it even at a point in time. And I think let's add Valentino to that. What's the code name V? Code whisperer that's going to get me the generated Ruby code that he would get. ⁓ So yeah, I know it sounds hokey, but that's my wish list. So Sandy, if you're listening, ⁓ we want the Pooter AI CodeGen tool.

Valentino Stoll (44:04)
Ha ha ha!

Great.

Yeah, you know, you're not wrong. Like, ⁓ I imagine people would pay hefty prices on a subscription basis for access to Sandy Metz, right? Especially if she packaged up, right? All of her like knowledge, ⁓ all of her like personality and thoughts and right like and had I'm sure she has it documented, right? ⁓ And and just like rented it out as an agent, right? And say,

Joe (44:30)
Mm-hmm.

Yeah.

Mm-hmm.

Yeah.

Right. And then it's

up and then it's updating. Right. So I pay a subscription fee because, know, as she's writing code, she's updating it and I'm getting those updates.

Valentino Stoll (44:51)
Right.

Right. And she's thinking about new ideas and creating new

abstractions, right. ⁓ And thinking about like high level organization structure and teams and how they work together. And you can use her in different ways, right? Like say you want to analyze like, you know, your organization structure and try and like get, make sure that your services are like abstract and object oriented enough, right. You know, having her go through and analyze all that and like even drafting plans. ⁓

Joe (45:04)
You

Mm-hmm.

Valentino Stoll (45:24)
you know, and like for a higher model, have her like approve the plan, you know? You know?

Joe (45:25)
yeah.

Yeah, I was just thinking that you took the words right out of my mouth because I have

this like big refactor that I'm that I'm scheduled to work on for this enterprise that wants like an enterprise version of Phoenix. And I know what I want to do and I got a plan, but I'm like, I don't know. I it would be nice if I had somebody like, you know, Sandy or Obi to say, yeah, this is this is a good plan. I mean, even even the AI version. Yeah.

Valentino Stoll (45:50)
Right? Because you know, you know, she'd spend

like 30 minutes on it and be like, well, these two things you don't want to do. Or like, definitely, you know, just like a glance, you know. And I feel like that's, that's definitely, I think calling out some limitations of these things, right? As, ⁓ yeah, you can build a persona, you know, all you want, but it's still going to take them days ⁓ to get to the point where a glance, right? The speed is definitely not there.

Joe (45:57)
Yeah, yeah, yeah, yeah, exactly. Yeah.

Yeah, fair enough.

Mm-hmm.

Yeah.

That's true. That's true. Yeah.

Valentino Stoll (46:20)
⁓ Or quality. Or quality. But...

Joe (46:26)
But you know, for like

$29.95 a month, it's worth it. You know, get to get a little taste.

Valentino Stoll (46:30)
It's true.

Yeah, you know, ⁓ I had this ⁓ one of the first I made a GPT on chat GPT like one of the extensions and it was right. ⁓ One of my first ones was the software architect's mentor and I literally just like uploaded all the books that I had ⁓ available and put it.

Joe (46:44)
yeah, I make them all the time. Yeah.

Mm-hmm.

Valentino Stoll (46:59)
into the context or knowledge base of this agent and was just like, okay, follow these books for guidance on code generation. It was awesome. It honestly was great. Now, I took that down, like, well, I don't have permission to put their books into a GPT, right? Yeah, well, I do, but.

Joe (47:02)
Yeah. Awesome.

Mm-hmm.

Yeah. Yeah.

you did?

but you use it personally.

Yeah, okay, yeah, yeah.

Valentino Stoll (47:25)
You know, but even still like, you know, I would love if somebody packaged that up and like made it distributable. Right. And so I could just be like, Hey, search the knowledge base of these books. cause of what I found, what I found was adding the wrong book actually made the output worse depending on the use case. Right. ⁓ and so something like the clean coder, right. ⁓ which is like very enterprise thinking if I'm working on a small open source project, like

Joe (47:31)
yeah.

That's interesting, yeah.

Mm-hmm.

Mm-hmm.

Yeah, that's really funny.

Valentino Stoll (47:53)
create all these crazy abstractions I didn't need, right, or want.

You know, and so like maybe I don't need that in all cases, right? And so having something selective where you're just like, all right, let me spin up an agent that has these resources, right? ⁓ And I would love to see that.

Joe (48:01)
Yeah.

Yeah.

I would love to see that too. And I, you know, I think we've talked about this on the show. The GPT seems like a commercial miss by OpenAI, but they still are as a, as like an internal, at least for me, internal or team wide tool. They're really valuable. I mean, they're, they're very easy to make. I mean, you ask ChatGPT to help make it for you. And like you said, you start uploading documents, start uploading books and, the thing just takes off. I find them really, really useful.

Valentino Stoll (48:25)
Yeah.

Right.

Now, I know we're coming close to time here, but I just thought of something that's super controversial that I saw it in my LinkedIn feed ⁓ this morning. And it was somebody ⁓ author of many popular open source projects saying that they were going closed source and that the monetization strategy of open source is broken now because as somebody who made open source projects,

Joe (48:40)
Mm.

Ha! Alright.

Alright.

⁓

Valentino Stoll (49:06)
in order to get more business for themselves and to get more like influence, right? And people going to their open source projects that knew about them, which made like incentive for them to keep building stuff. ⁓ Since that's broken and no humans are actually looking at their documentation or their projects anymore, that ⁓ they no longer get like the benefit of maintaining and open sourcing stuff. ⁓ So what do you think about this?

Joe (49:12)
Mm-hmm.

Right.

Mmm.

Wow.

My first knee jerk reaction is that this is, this is so controversial that you can't even say the person's name. I haven't been able to quickly find it, so I will just react. ⁓ so my, my first reaction is I think, look, I'm a, I'm a, I'm an engineer turned businessman. And, ⁓ I think that people deserve to get the, ⁓ people deserve to get value from the work that they do.

So if this person doesn't feel like he or she is getting the same value ⁓ from working on open source projects, then making a closed source is a viable alternative. think that we get, you know, we're Rubyists and so we expect everything to come to us for free. ⁓ Most of us, I'm not saying you, I'm not saying me, but most of us just, we give back very little or nothing.

And that doesn't really work, ⁓ you know, on the grand scheme of things. In fact, it kind of has worked for 30 years. But ⁓ as things change and things shift, I think that there's room for ⁓ the business. I'm trying to avoid saying the word paradigm. It's reasonably, sounds too much like a business owner. It's reasonable for things to shift and for the monetary structure to shift. That's what I'm going for. All right. What do you think?

Valentino Stoll (50:49)
Hahaha

Yeah, I feel like it's hard because like ⁓ the monetization, nobody's created something to solve the problem. And like, it's been around a while and people already knew it was a problem. And it's almost like everybody's just accepting the fact that ⁓ if you have something open source, it's cannibalized by these larger organizations for their monetary benefit. ⁓

Joe (51:22)
Right, right.

Valentino Stoll (51:27)
And so like, then it really becomes, like, why are you creating it? For me, I just enjoy making things and solving my own problems. you know, but like, then I have to go and maintain them. Like it's the maintenance really that like, how do you, how do you then solve that? Right. ⁓ does GitHub take on that burden? Right. Like does, who takes on the burden ultimately? Right. ⁓ so, and I think about that too, from like, you know, anthropics perspective, like buying bun, right. Like.

Joe (51:35)
Mm-hmm. Mm-hmm.

the maintenance.

Mm-hmm.

Mmm.

Yeah.

Valentino Stoll (51:57)
Is that the future of like Anthropic, OpenAI, these organizations buying out all of the open source so that they can, right? Or like, do security updates work, right? Like if nobody's maintaining things anymore ⁓ or has no incentive to maintain them, right? And I feel like we're going to see that of a lot of drop-off of like, open source maintainers like...

Joe (52:06)
Right, right.

Absolutely.

Valentino Stoll (52:25)
kind of losing a little kick in their step because the value, it's not for them, then what is it for?

Joe (52:36)
Well, you know, speaking of cannibalizing open source, you know, we all work on Macs, which just cannibalized Linus Torvalds, you know, open source operating system and decided to make it, ⁓ you know, you know, their own. then we, yes, enterprises and companies, large and small, do the same thing. ⁓ When we, when Phoenix started using FLOG as a primary thing,

Valentino Stoll (52:47)
Hahaha

Joe (53:06)
⁓ We went and started, you know, making a donation to ⁓ that repository. And I was thankful that GitHub had an easy way to do that. Before that, there wasn't even really much of a way to compensate people for their work. Typically, the way to compensate was to, and this I still think is really valuable, and to your point, well, one way that you can help to compensate and provide value is by giving those updates, by using it.

learning where there are rough edges or where security updates, for example, are needed, doing them on your project and then submitting them back to the open source library. And now is the contention that none of that is happening now either.

Valentino Stoll (53:53)
I would say so. Yeah. mean, I would even say that, you know, there, there will even be seen a lag in releases of security where people ⁓ maybe patch the security. ⁓ And then, you know, the, you know, contribution back, right? Like takes longer ⁓ because the maintainers

Joe (53:54)
Yeah. Yeah.

Mm-hmm.

Why does it take longer?

Okay.

Valentino Stoll (54:21)
I would imagine

because like either ⁓ the maintainers don't have the incentive to like, you know, merge it and review it and all of this, right? ⁓ Or, ⁓ you know, people don't want to spend the time to do that, right? Maybe they have an agent that contributes it, ⁓ right? But like,

Joe (54:30)
Mm-hmm.

Right. Well, that's why I'm curious.

Is it, is AI an accelerant to this, this downfall of being able to capitalize on open source or is it just kind of AI exists as we are failing to sort of meet our obligations as open source users?

Valentino Stoll (54:58)
That's true. mean, on the flip side of all of this, Like open source has never had a great monetary strategy, right? Like, ⁓ and I feel like, you know, a lot of people hope that like the whole like GitHub sponsorship stuff would solve that. And it's kind of like it has in some ways, right? For the, for those that like maybe have more popular, ⁓ you know, ecosystem or, ⁓ you know,

Joe (55:05)
No, it is not.

Mm-hmm.

Valentino Stoll (55:28)
I have more popular repositories. Maybe they get sponsorship, right? ⁓ I know Gusto, we sponsor individual maintainers ⁓ of the things that we use. But if you're getting up to the new stuff, if you want to create something new, what is your incentive to release that? And I think about Obi too with his Nexus idea that he's building internally.

Joe (55:36)
Mm-hmm, good.

Mmm.

Valentino Stoll (55:57)
⁓ you know, he's chosen not to open source it because it's a competitive advantage for his company, right? To not, and like there, there will be more of that, right? Like what is the competitive advantage for any company to then release the things that they're building anymore, right? ⁓ and so like to ship to then share it easier with everybody else that has an agent, they're just going to scrape it and then rebuild it, right? Like, that I worry about.

Joe (56:04)
Right.

Mm-hmm.

Mm-hmm.

Valentino Stoll (56:26)
from like a giant organization level ⁓ of like maybe we'll see like organizations taper off from this active like spend on open source.

Joe (56:26)
Yeah.

Yeah,

it's possible. I wonder if speed is going to remain the differentiator for these projects and for businesses in general. I speed to market has always been, you know, it's long been seen as an advantage, but I wonder if quality of execution is going to matter more now. ⁓

because the one thing people talk about with AI is hype and how much stuff comes out. for sure, in 2025, we saw a new tool per minute coming out. 95 % of it was crap, but 5 % of it was amazing and really good. And I just wonder what we were talking about before.

you give it a little time and maybe you get it right and maybe the, this is me being an optimist again, maybe the spoils accrue to the person who gets it right rather than the person who does it fastest. ⁓ But I don't know.

Valentino Stoll (57:40)
Yeah, time will tell. ⁓ But you know, there's a lot of wishes out there that I think can make a lot of this, you know, solve a lot of these problems we're talking about. ⁓

Joe (57:42)
time will tell.

Yeah, yeah.

Yeah. Yeah. And if the world

could just deliver these to us by our, by next episode, we'll be in good shape.

Valentino Stoll (57:52)
I hope Scott

Werner is listening to this and Monkey Paw suddenly just like solves solves any wish ⁓

Joe (58:00)
⁓ that would be nice. Yeah. Yeah, just solve any wish. It's got a he must

have made it so that you could talk to it by now, because that's really what I want to do. I want to fire up the I want to fire it up on my phone and just say, hey, can you do this for me? Yeah, I want to hold it. Yeah. Yeah. There's a visceral element. Yeah. Yeah, exactly.

Valentino Stoll (58:10)
Right? You want to be able to hold that monkey's paw, right, in your hand and just like speak to it, right? Yeah, visceral.

And then as soon as things go bad, you just throw it, you know?

Joe (58:25)
Yeah, yeah, yeah. Yeah, it was me.

Valentino Stoll (58:31)
All right, well, I think we've hit our limit here. ⁓ That's right. Hopefully you've made it this far and have listened to our ramblings. ⁓ But I'm pretty excited about this year. ⁓ I see so many things happening with the Ruby LLM project and people making use of Ruby for building these things and building on ⁓ LLMs in general.

Joe (58:35)
Yeah, or you have hit your limit with us, dear listener, and we thank you. Yeah. Yeah.

Mm-hmm.

Yeah.

Valentino Stoll (59:01)
and it is becoming easier. ⁓ And I think that, you we are really poised as a language to really grab the reins of the frontier. And I'm excited, you know. ⁓ I have recently, you know, revisited making stuff in Ruby ⁓ because I just don't want to be building JavaScript stuff. You know, I don't...

Joe (59:01)
Yeah.

Yeah, I agree with you. I feel that way as well.

Yeah, nobody does.

Valentino Stoll (59:26)
I don't want to be making a node app for

an MCP server, know, like I can do that in Ruby now. ⁓ You know?

Joe (59:31)
Yeah, my God. mean, that's

how you know that Claude is being written mostly by Claude. It's because it's in JavaScript. No person wants to do that.

Valentino Stoll (59:37)
Red.

⁓ And you know, like a lot of the abstractions just like don't make any sense. ⁓ And I feel like that's where Ruby really shines is the right abstractions and object level thinking ⁓ that like maybe an LLM will come out that thinks this way and thinks in objects and or better abstractions. But right now it's very much just like, let's just do it. You know, it's like.

Joe (59:46)
Yeah.

Yeah.

Yeah.

Valentino Stoll (1:00:09)
You know, they're on super accelerants and, you know, we either slow them down or we don't.

Joe (1:00:16)
Yeah.

⁓ I feel very confident as well in this year because I think that the yeah, there's been a lot of we've done a lot of proving.

the technology of AI code in general and for Ruby specifically in 2025 and here in 2026 to build on what you said. Yeah, a lot of us are ready to move and ready to build. And at the same time, the confidence across businesses is a lot higher than it was in 2025. And when confidence is high, people spend more money and when they spend more money, they're going to invest in the kinds of things that we can do.

And that makes for an exciting year, we think.

Valentino Stoll (1:01:04)
Yeah. See you soon.

Joe (1:01:07)
Yeah. All right, everybody. Thank

you all for joining us and we'll see you again soon.

Valentino Stoll (1:01:13)
Sounds good.