Joe Leo  00:00
Hey everybody, welcome to the Ruby AI podcast. I'm your host, Joe Leo. My co-host Valentino is out today; he said he has something to do with Gusto, and he didn't say what it was. I suspect they handed him a shovel, and they told him to start digging for more money for Ruby Central.

Joe Leo  00:17
But I am excited to be joined here by another grizzled veteran of the Ruby world: co-founder and CTO of Cause of a Kind, Mike Raspoli. Mike, welcome to the show.

Mike Rispoli  00:28
Happy to be here. Thanks for having me.

Joe Leo  00:30
Yeah, thanks for coming on. I wanted to start here at the top of the show, because this is— we're recording on May 13, and a post here from May 11 from Ruby Central. Gusto joins the Ruby Alliance, quoting: "The Ruby Alliance is a small coalition of companies that are choosing to make a significant investment in the long-term health,

Joe Leo  00:49
resilience, and sustainability of the Ruby ecosystem and the infrastructure it depends on." We are honored and humbled by how quickly and decisively Gusto moved to join us. They didn't wait to see what other companies would step forward; they recognized the importance of this moment and chose to lead the way. Interesting choice of words.

Joe Leo  01:07
"They didn't wait to see what other companies would step forward." The Ruby Alliance was announced about a month ago, so I think what they meant was: nobody else stepped forward. We're really awfully glad that Gusto did, because they're a big name and they've got lots of money. This is after a series of breathless posts from Ruby Central in the last month, right? Executive Director Sean Kuratin is out in less than a year.

Joe Leo  01:28
Truly unsurprising news there. We're out of money. We can't do the gala at Ruby Conf anymore. Sorry, David Black and everybody else who helped found Ruby Conf. And oh, by the way, we're building a bunch of stuff, like the Ruby Alliance, which isn't building anything, it's just getting more money. Project Dream, which somebody in that meeting was like,

Joe Leo  01:47
"Hey, we just got to say something about AI, so let's do Dream." And then an apprenticeship program where they take a bunch of people from underrepresented groups and have them work, presumably for free, on a bunch of these initiatives. So my first question for you, Michael: is this the end of Ruby Central?

Mike Rispoli  02:04
Yeah, hit me with the hard one right away. Is this the end? You know what, I don't know. There is a conceivable that they could get, like, a bunch of corporate sponsorship. I kind of think that's the future for a lot of stuff. Now, it seems like, I don't know, maybe for most of my career, open source kind of ran through donations, a little bit less corporate sponsorship,

Mike Rispoli  02:26
but it feels like that may be the way that they need to go. And then there's the whole, like, debacle about Ruby Gems. I want to say, like, last year, and early— yeah, that was last year, right? And that was the whole thing. And it was kind of like, wait, what's going on here? And it's really interesting how you see these evolve, because now,

Mike Rispoli  02:45
like, on the other community I'm a part of, is the JavaScript community, we just seem to, like, oh, we'll just build a new thing. Like, oh, I'm just going to go do Deno, and then I'm going to build JSR. Like, forget it. Like, we'll just do a new, like, a totally new thing. And you keep reinventing stuff.

Mike Rispoli  02:59
So, like, one of the interesting things about Ruby, one of the valid things that I liked, was that they didn't seem to spend a lot of time on reinventing stuff. And there wasn't a lot of fracturing for a really long time, and now it feels like there is.

Joe Leo  03:13
Yeah, I think that's insightful. Because we're here on the heels of Last Ever Rails Conf was last year. I am openly wondering if I'm about to attend the Last Ever Ruby Conf. I'm going. All of Deft Method's going. And we're excited about it. You know, I love Ruby Conf and I love what Ruby Central has done over the years. But you're in this world as much as I am,

Joe Leo  03:34
and you've seen things change. And I'm curious to know, is there room, or is there an advantage to some sort of guiding, like, governing body, whatever you might want to, stewards of the Ruby community?

Joe Leo  03:50
And what kind of form should that take in this world of artificial intelligence and extremely fast development and all of this change?

Mike Rispoli  03:58
I think it kind of goes back to that, you know, I've thought about this too. Like, do you do the benevolent dictator for life kind of method around everything? And that seems to have worked well for Python, and it kind of worked, you know, it's worked for Linux too, right? As much as people have problems with Linus, ultimately we ended up in a good spot.

Mike Rispoli  04:17
I kind of like when there is sort of like a guiding principle like that versus the committee-based. The languages that I've been a part of that didn't do so well, like, Elm had this with Evan, but then he lost interest. Or it seems like he lost interest. The Elm community is probably going to scream at me right now and be like, "No, we're still here." Like,

Mike Rispoli  04:36
but it does seem like he did lose a bit of interest in continuing to develop Elm as a UI. So you run that risk. I don't think Matt's is going to, like, lose interest in Ruby. I don't think DHH is losing interest in Rhea. Like, it doesn't feel like any of that is happening in the Ruby community. And there have been, you know, those efforts of, like, pushing certain individuals out and for various reasons.

Mike Rispoli  04:56
And I feel like the language itself has always ended up in a good place, and the framework itself has always ended up in a good place. I worry if there's more committee or other corporate sponsorship, what ends up happening there. It's going to be interesting to see. But like, Golang has done well under Google stewardship, even though I feel like Google doesn't use it very much.

Mike Rispoli  05:16
Like, from what I've been told by Google people, is Golang isn't actually a very big internal use for them. Like, as big as C and C++ and other things. But people love it, so they've done a good job with it.

Joe Leo  05:30
I think there's something we said for what you're putting out there, and I agree with you. There's like a thin line here between corporate sponsorship and corporate investment. Gusto has, Shopify has, I'm sure several other companies have, departments that are focused on just working on Ruby and on just working on Rails,

Joe Leo  05:51
because it benefits them. They use the language so much that some gain in performance, some kind of efficiency, some kind of fix to memory, right? This is why you got these great compilers coming out of Shopify. It benefits their company, and so they put that investment in. In a similar but different way, Google did that for a long time, right?

Joe Leo  06:13
They invested in the language because the language was benefiting them. Beyond that, when you talk about sponsorship, then there has to be an organization that collects that sponsorship, those sponsorship dollars, and then there has to be a committee that decides what to do with those dollars. And so it does become a different thing. And for a long time, we got a lot of benefit out of Ruby Central.

Joe Leo  06:33
And that's not to say that we won't in years to come, but it's looking kind of shaky. And I'm just kind of openly curious what the future holds. I guess everybody is to a degree, and we'll learn more when we get to Vegas in a couple of months. Are you going? Are you going to Ruby?

Mike Rispoli  06:48
No, I was thinking of going to Rails world, but I wasn't going to go to Ruby Conf this year, yeah. Because we had just gotten back from AI Engineer in Miami. For us, we have our families and stuff, so it's like hard to get away. But I was like, all right, maybe it'll be Rails world. Admittedly, the Rails world, they do a great job with social media. It always looks like fun. I'm like, just like makes me want to be there. So I'm like, we got to go to that.

Mike Rispoli  07:07
It's going to be in the United States. I was like, flying to Amsterdam was not possible for us last year, so I was like, we got to go to that. I didn't even really think about Ruby Conf, to be honest. I mean, maybe that answers your question about what's going on with Ruby Central. I was like, I didn't even really think about it, you know? Like, I was like, oh, yeah, there's a conference they're putting on. That's what you're doing.

Mike Rispoli  07:25
No, it feels very much like, I guess when I think of Ruby, it's like so synonymous with Rails. And you talk about Shopify, DHH, that whole ecosystem. And it gives me a lot of confidence in investing in it, because I know that, you know, Shopify is doing well. They're at the forefront of AI.

Mike Rispoli  07:42
So I think about things like that too when I'm choosing to continue to use Rails, especially kind of our return to Rails after years away. You know, it was not an AI-driven decision, because AI would tell you to build everything in TypeScript and Python. It was really like, I was like, hey, I kind of have this hunch this is going to work better. And then we were at Omicon this year,

Mike Rispoli  08:03
so seeing Toby talk about Shopify's investment in AI and kind of the future there, I said, you know, I think the future is really bright for Ruby and Rails in particular, and we should be advocating for people to be on this platform.

Mike Rispoli  08:15
Now, all these governance issues, I kind of like this idea, and this is maybe there's a piece of me that likes sort of a decentralization, so to speak, but there's a part of me that's like, eh, maybe it should all be owned in one place. You know, that was how I felt even about the Gems issue. I was like, maybe this should live with the Ruby language.

Mike Rispoli  08:36
This should live together. It shouldn't be this issue. There shouldn't be this takeover of people's path. Like, it shouldn't even be possible to do things like this. It's interesting. Then again, now I'm with the recent attacks on NPM, and I think Gems is going through that now also. I kind of come back to like, eh, maybe we should be doing the Go model,

Mike Rispoli  08:55
like, or what Ryan Dahl was advocating for Deno originally too, which is Git repo, URL, those are your packages. And like, you own them. And like, get away from the centralized managers. I don't know. You know, we're going to see what's going to start to happen as the surface area of all these things takes, like, gets the attack surface area gets bigger anyway.

Joe Leo  09:14
Yeah, I agree with you. And there's a lot to say about that as well. But I want to, the attack surfaces specifically, but I definitely want to turn now to Cause of a Kind. And especially because you've got some all new branding out there, right? New website. New website, modernize your software. I went there looking for one thing and I saw another. So tell me about that. Tell me about the word modernize.

Mike Rispoli  09:36
Justin and I, every year, we're always evaluating. And as an agency, there's these ups and downs. And I tell people all the time, we've, through last year, we were having, we were like riding high, signing tons of deals. We were doing amazing. And then January, we were still doing great, but then we've kind of, now February, March, April,

Mike Rispoli  09:55
there's been a bit of a lull, I want to say, in like our pipeline. And this happens every year. I'm prepared for it as an agency, but every time that sort of happens, I have to ask myself, is this just normal or is there like something going on? And because it doesn't match the typical seasonal lull of our books,

Mike Rispoli  10:15
I said, I think that there are shifts happening and things that we were previously offering are becoming less appealing. And then we also kind of had to reevaluate our price points. There are things, for example, landing pages, websites. We used to do a lot of this type of stuff, WordPress, even the Shopify theme market.

Mike Rispoli  10:36
I don't know with AI if there's going to be much of a competitive place that we can sit in that space. And we sort of reevaluated and I said, well, what are some of the things that we've been doing that go really well? What are some of the verticals?

Mike Rispoli  10:51
It seems like healthcare tech is big because there's a security requirement there that means that they're not just going to have the CFO vibe code, their accounting thing, into existence. We hope, we should hope, right? So I was like, great, we know healthcare. Education, we're getting contacted, like some of our best leads coming in now. Education tech is big.

Mike Rispoli  11:10
People are looking to do like big migrations of things to new platforms. And AI is there. It helps me do things like that, but the people, they don't know how to do it. We're talking like huge databases. So I was like, okay, these are good places to go. And then the other thing that started to happen was these modernization initiatives. We had been working with a couple of like PE funds,

Mike Rispoli  11:30
and we're contacted by another partner of ours that says, I get these things all the time, which is basically a company wants to get acquired. They do like a tech audit, and there's problems. And they want that remediated before any kind of purchase happens. And I look at that as another big spot where you're not just going to vibe the remediation into existence.

Mike Rispoli  11:51
And that started with this idea. Now, this product, we added a third product of really top-tier offering called War Room, which we have two levels of subscriptions that we've always offered, kind of like this half-work stream and a continuous work stream. Added War Room, which is on-site, specifically like we're going to come in.

Mike Rispoli  12:10
And it was a top-tier offering. We talked about it, I want to say, a year and a half ago before ever putting it on the website where I said, we should figure out. I just had this hunch that this was going to be the thing. It's sort of like forward deployed engineering as a service, because we had a number of clients where I felt like if we had gone on-site, things would have even gone smoother. Not for a long period of time,

Mike Rispoli  12:30
but like trying to get this engineer in a meeting was impossible. Trying to get security, figure out where things were. And I was like, man, if we were just in the building, we could have said, okay, we're going to be in the building this week. Everyone's got to be here. We got to get all this figured out, because we run into these .NET code bases that have lots of legacy software, things that people don't know what's going on.

Mike Rispoli  12:52
There's sometimes a hodgepodge of different languages. Someone built this thing once upon a time, that thing. So we were looking at that in the mid-tier kind of enterprises, and we said, hey, we should go after this type of stuff.

Mike Rispoli  13:04
And then just AI enablement in general, one of the other pieces in building agents for people was I felt like I needed to go and watch people work sometimes. I couldn't automate what I didn't understand. And I think that was something I keep seeing on Twitter. And I was like, this doesn't actually work, where you're just like,

Mike Rispoli  13:23
I'm going to go in and put in like automated answering service, and all of your problems are going to go away. And I was like, I would love to see someone try and sell that to a small business owner. And I already know what they would tell you here on Long Island. I was like, let's revamp this. Let's talk about modernization, upgrades, because this is where I think the puck is going and where like we're now getting,

Mike Rispoli  13:45
like I want to say in the last like month, this is where all the lead volume is coming from, is this modernization. We have migrations, modernizations happening, new apps we want to build or old apps we want to upgrade. So I don't know if that's a long-term trend, but I feel like it is. And I just want to get away from what could be vibe coded.

Joe Leo  14:05
That last statement, you know, trying to get away from what could be vibe coded, that's what a lot of people of your and my ilk are trying to do. There's so much that can be automated, which is good. You know, we use the tools, you use the tools. We can get our agents to write a bunch of code, and then we can be responsible about it and make sure that the code works.

Joe Leo  14:27
But we're going to move really quick on that. And there is a lot of downward pressure on anything that can be built without too much human intervention. And so what I hear you saying, and you can correct me if I'm wrong, is, hey, let's look for the places where the humans really got to be there. Like you really have to have somebody who knows what they're doing, even with the tools,

Joe Leo  14:48
or else something could go wrong, right? Something could break. You could have a security violation. You could have a bunch of patient information let loose into the public, right? Or something like that.

Mike Rispoli  14:59
Yeah, exactly. That's what I'm thinking, because risk is the thing people are willing to pay to mitigate. We pay for insurance for this reason. You know, nobody would have insurance if we only needed to worry.

Joe Leo  15:10
The bills are coming up, you just reminded me. Yeah.

Mike Rispoli  15:13
Dude, it's just been brutal. It's one of the things like I tell people, like the one of the hardest things about being a small business owner is health insurance. We could do like a whole episode on that, because I'm like, for a place that likes entrepreneurs, they really don't like entrepreneurs. So, but yeah, this is kind of what I've been racking my brain around.

Mike Rispoli  15:32
And now we talked about this on our podcast, whereas we were always after the short sale cycle type stuff, we're starting to go after longer sale cycle type stuff, because we're realizing that the short sale cycles type stuff, a lot of folks are like, oh, we're going to try and do it ourselves first. And I actually think that stuff is going to come back around.

Mike Rispoli  15:52
And if you look at kind of where I'm positioning our podcast and some of our other marketing material is, I actually want to teach you to do it yourself as far as you can get it. And then I want everybody who gets stuck, which will probably be like if a hundred people do it, they'll probably be like 40% that gets stuck or something.

Mike Rispoli  16:09
And this is like the goal of trying to build this lead funnel of let's teach everybody to do everything they can with AI. Like let's blow the lid off this thing. But what remains is going to be really hard work. And that's kind of where I want to sit. And then it goes back to hiring engineers. If you take low-value work, it's very hard to keep people employed. The stuff turns around too fast.

Mike Rispoli  16:29
Like a landing page takes me a couple of days to stand up. I can never hire someone around a landing page pipeline unless hundreds of people are bringing their landing pages to me. I can't do it.

Joe Leo  16:41
Yeah, no, that makes a lot of sense. You know, we went through this. I know Cause of a Kind is, how long have you guys been in business?

Mike Rispoli  16:48
We've been in business nine years, but since we left our jobs, it's been four. And so that's why I always, yeah.

Joe Leo  16:54
That's a big cutover. Yeah. I had an advisor when I was first starting out with Def Method, and I was working full-time, and I had Def Method or whatever I was calling it at the time, you know, on the side. And he was just like, it's never going to succeed until you depend on it for your, for dinner. And you've got to learn that.

Mike Rispoli  17:10
That's true.

Joe Leo  17:10
So, okay, so four years, congratulations on that. I bet nine years in general. What I've noticed is that things have changed a lot in the entrepreneurial space in that time, and certainly over the last four years, because what used to happen is that people would try to build things.

Joe Leo  17:30
Scrappy is a generous term for it. Really, it's poorly. Just make it work for one path through the code, right? And the idea was that's going to cost a lot of money, even doing that. It's going to cost you 150, 200,000 dollars. And if you want to get done anyway, right?

Joe Leo  17:48
And then you have to go out and you have to raise money so that you can build more. And it was always like, well, we'll fix it later. We'll fix it later. Which, of course, never happens. And that's when you inevitably would call Def Method, because you would eventually get yourself in a place where you were kind of screwed. And we love that work, but now you don't have to do that. It's still an option.

Joe Leo  18:06
There's always an option to do it poorly. But we don't usually go after bootstrapped startups anymore for that reason, because the turnaround is so quick. The amount of time you have to spend sort of educating the entrepreneur who has never done this before is intense. And so my time would get sucked into it. But now, like I've had this recent experience where,

Joe Leo  18:28
hey, we can get something off the ground in a couple of weeks. I used to tell people, we're not going to ship anything that you're going to want to show anybody for like three months. And now, you know, in a couple of weeks, you can get something out there. And so now you can build it really quickly and you can build it well.

Joe Leo  18:42
And so you're not forced to make that horrible choice of like, do I have to spend a lot of money or wait six months just so that I can get something that is well constructed?

Mike Rispoli  18:52
And we get a lot more people doing like the lovable thing first. I have one right now where they built it in Lovable. They validated it internally, but they were like, we definitely can't use this prototype. Things don't work. Like we don't want to be hitched to Lovable's wagon. This is a big enterprise, but it allowed them to put it in front of a team, figure it all out.

Mike Rispoli  19:12
Now there's a person working on kind of the classical design with Figma came to me and I basically take them both and I just go with Codex. And it's so funny. It's clear that they haven't worked a ton with an AI-driven engineer. They gave both to me. They outlined the first sprint.

Mike Rispoli  19:32
I was like, okay. It was like one of those moments where I was like, easy, like definitely going to be able to, but like almost got the whole product done in two weeks and like very, very close. And then just rebuilt totally on rails. Like I did like a rails rebuild. And I was like, listen, we don't need to worry about reusing any of that code. I'm glad you have it. I'm glad it's here. I'm glad you can pull specs out of Lovable, because that's what helps Codex.

Mike Rispoli  19:53
Still a million decisions to make in between though. Like as I'm going through it, I'm like, well, you have, you have duplication here and there's all sorts of stuff that we still talk about, but the speed of it is incredible. And so now what's happening is that same venture goes, oh, we had this other project that we thought we were going to do after that,

Mike Rispoli  20:12
but now we want to just keep going because there's all this software that we've been using SaaS for or we always wished we had, and now we can afford to just do it at this pace. So I'm like, great, but this is through a startup that is still building a SaaS. It's in the film industry. So once again, very secure.

Mike Rispoli  20:32
They're very, very worried about especially like the creative getting stolen and things of that nature, where they're all asking for like bespoke custom deployments. And so what was supposed to be the vision of a SaaS company venture funded is sort of turning into a, maybe we do custom deployments of this and we sell licenses of this stuff.

Mike Rispoli  20:53
And it's so easy to do bespoke now, because I just am like, okay, we're going to take exactly this thing and we're going to make it this way with these three additions. And the AI makes that really, really possible in a way that just wasn't before. And so I'm like, this is a whole new business model, because the former business model was always, we want to make multi-tenant SaaS.

Mike Rispoli  21:12
And now it's like, well, we'll do an on-prem deployment. That's the idea behind War Room too. I'm like, listen, you pay for it. I'm flying in. Like no problem. So this is what people are wanting. They're like, do we do an on-prem? Do we do a backup, something like this?

Mike Rispoli  21:28
Because the security, again, going after these like high security areas is definitely like piquing my interest, because it seems like this is going to be a very big thing. And there's like a new security problem like every week.

Joe Leo  21:40
Well, yeah. And you had mentioned this at the top of the show, and it's something I've been writing a lot about lately, which is that some of these security issues that have been happening, you know, famously one with GitHub, one with Vercel a couple of weeks before that, it's the greatest time to be a nefarious actor in the low, right?

Joe Leo  22:03
It's also the greatest time to be like a security engineer, because there's so much, right? There's just so much. And one thing that I had hit on that I'd like to get your reaction to is it seems to me like just being able to detect your security flaws is really not enough anymore, right?

Joe Leo  22:22
We need something where we are actively and continuously resolving security flaws, because look, I've been detecting security flaws for my whole career. And when I see them, I'm like, all right, you know, I fire up Docker. Docker gives me a whole bunch of CVEs that exist on my machine, and I get to them when I get to them in order of severity, and I don't worry about it too much.

Joe Leo  22:41
Well, there was never like an ever-evolving, continuously built mega monster of an AI searching for every security flaw to expose. And now there is, or at least there can be.

Mike Rispoli  22:54
It feels like someone put the Ralph loop on and just said, just keep trying to break into these services and stop. Yeah. And don't stop. And so there's like two prongs. One is like the penetration testing, which I had been through them, but I never like learned to do it myself. And now suddenly I was like, oh, this, I'm kind of interested in this.

Mike Rispoli  23:14
Like, you know, very Mr. Robot. I'm like, okay, I have, and again, AI is providing a lot more time in my schedule to learn some of this stuff that I would have previously not had time to do. And then, so we want to offer that. And then I think about that too, like continuously trying to penetrate your own stuff all the time. Like as a regular,

Mike Rispoli  23:33
like part of, not just the Snick or something that runs statically, but an actual like fire up the headless browser and like somebody tries to break in once a month kind of part of like an.

Joe Leo  23:42
Like the Chaos Monkey, like Netflix's Chaos Monkey, but for security. Yeah.

Mike Rispoli  23:46
Exactly.

Joe Leo  23:47
I totally agree with you.

Mike Rispoli  23:48
And you just kind of attack it. Now at some point, as all things, I think that's going to be huge, but I also think we're going to resolve a lot of these. Like I think this continuous attack is going to make all the software more resilient. And at some point, you'll hit a place where we were maybe a year ago where it was like, it felt like everything was pretty secure and there was a zero day bug.

Mike Rispoli  24:07
It wasn't very common. There's going to be like an uptick in this, and then we're going to gradually get over that hump. But I think that we have a long way to go, because the other thing that opens your surface area was you talked about the Vercel attack. I was suddenly worried because I go, all right, yes, I'm built on Next.js, but I'm also on Vercel. Had I not been on Vercel,

Mike Rispoli  24:27
I could have patched it and been like, well, no one was coming after that tiny little app or whatever. But now that you're on Vercel, it's like, well, they stole everybody's stuff. And so now I'm like, man, you look at Vercel, you look at Cloudflare, you look at AWS, and I'm like, should I be going with like a smaller cloud provider? You know, where I've been looking at.

Joe Leo  24:43
Some obscurity, some security by obscurity. Yeah.

Mike Rispoli  24:46
Yeah. Yeah. And again, that's not true security, but at least it gives you time where it's like, oh, they stole all of Vercel stuff and you're like, great, I'm going to go patch everything up on my side because I wasn't a part of that. Now there's some aspect of in the agency world is you don't get fired for choosing IBM. You don't get fired for choosing Vercel kind of situation where it's like,

Mike Rispoli  25:06
well, all of Vercel got stolen, so your stuff got stolen by default. Don't worry, you're in the same boat as most of your competitors. And so like sometimes there's like a comfort in that. Like when AWS is down, everyone goes, oh, cool, AWS is down. Don't worry. But if my bespoke under my desk thing was down, you'd be like, oh, Mike's a moron. Like there's much.

Joe Leo  25:25
Yeah. That's true.

Mike Rispoli  25:26
So you got to like balance those two things.

Joe Leo  25:31
Let's talk about the tools for a minute. I was really interested in this. I think, Mike, you had this right. You've been talking for a long time about, at least since last year, about not just relying on one model, not just relying on one kind of piece of software, one harness, and kind of extolling the virtues of open source,

Joe Leo  25:52
whether that's in the models, whether that's in the hosting. I know you did at one point do a lot of self-hosting on your own framework desktop. I don't know if that's true anymore. I'll be interested to hear about it. So the thing that I thought, I never disagreed with you, but I was always like, okay, well, that's interesting, but maybe there's no urgency behind it. Well, then Opus 4.7 comes out a few weeks ago,

Joe Leo  26:12
and then all of a sudden people were started paying 10 times the amount that they were paying for their token usage because they just absolutely had to have Claude Code's latest and greatest. And when you get past the marketing, whether you believe it or not, about how wonderful and how great Opus 4.7 is, you realize no matter how great it is,

Joe Leo  26:32
all the other models are going to be just as great in probably two weeks to six weeks, except there were a whole bunch of people that were not prepared for that and just ended up either spending tons of money or grinding to a halt on their development. I would imagine that that statement does not describe you because you were kind of well diversified. So tell me a little bit about that.

Mike Rispoli  26:50
So I had kind of felt the degradation of Claude Code and Opus 4.6 for some time, and it was myself and a number of our other engineers were complaining about it. And so I immediately was like, well, not a problem. I got a Llama. I'm going to go try. I'm like, every time something like this happens, I go and like, let me fire up and see what the open source world is doing.

Mike Rispoli  27:09
I tried Gemini 4 or the Gemma 4. That was it. I wasn't happy with that performance at all, and it was slow and it wasn't very good at coding. Qwen Coder was pretty good, and that I was able to run locally. But again, you run into some of the speed issues. And then I tried Kimi 2.5, which was something that DHH had mentioned on a post.

Mike Rispoli  27:30
I said, let me go with the Llama, try that. That I couldn't run locally. I don't have enough RAM for it, but on Ollama Cloud for like 20 bucks or something like that, that was actually good. So suddenly I was like, oh, well, I don't really need Opus. Like I could get by with this. And I go back to, it requires me to use a little bit more of my engineering skills,

Mike Rispoli  27:50
which I kind of like anyway, because it doesn't do as much of the thinking for you. And I was like, I'm not going to pay you a thousand dollars a month for like this. You know what I mean? I'm just going to do a little more work on my side, and that's fine. Then of course, GPT 5.5 comes out. I was using OpenCode, but then I was like, let me try Codex. And that's been the thing that I've been using.

Mike Rispoli  28:09
So I was like, oh, this is great. So it's so funny because then they did that whole Mythos stunt, you know, with the whole thing. Yeah. It's like, this one's too good for you. Like you can't touch the, yeah. I'm like, okay. Like I was like, and so I was like, okay, fine. I guess I'm not worthy. Like I'll take my money elsewhere.

Mike Rispoli  28:28
I took it right to OpenAI and Ollama, and like, and that was it. Right now I'm enjoying Codex and GPT 5.5, and I do switch to Kimi just like every now and then, but I actually find GPT 5.5 on low thinking is getting like the best performance for me right now. And I was really surprised.

Mike Rispoli  28:48
Again, one of these things where you kind of have to be in the bubble watching this stuff, someone had mentioned it. I said, let me try that. And it was funny because I had this situation that I want to say it was maybe like a Monday, Monday night. I switched to GPT 5.5. It does a solution, and it created in this Rails code base. It built a bunch of these service modules around this one feature,

Mike Rispoli  29:08
and it kept me up because I didn't want to commit it. I was like, this is not the way I would do it, but it worked. So I was like, kind of went through and it was working, and I was like, do I go back? And I go to bed. I woke up the next day. I said, I can't let it go. So I was like, listen, I want you to do it this way. And so it made all my corrections. I was like, this should be like three functions, not modules. Like, I was like, this should be like this.

Mike Rispoli  29:27
This is how I would do it. It fixes it. But then I switch it to low thinking, and it actually got it the way I would have wanted it in that mode. So I was like, ooh, I'm using this model a little wrong because in the Claude world, everybody's on like high or extra high, but with GPT, everybody's like go low with the thinking.

Mike Rispoli  29:48
And this is where I say you have to try the different models because they all behave differently. They're all good at different things. And I think this is going to be a future. Like I'm really excited if Qwen Coder, if I could run Qwen Coder locally and get very similar performance, because I don't need the history of the world in my model. You know, I just need it to know.

Mike Rispoli  30:08
Like I can't wait till we get like the Ruby on Rails only model, that kind of thing where you can actually say, hey, you can run this locally. So if the clouds are down and everything's out, you're still kind of working with something. It's open source, so you own it. But it's very clear from the business owner's perspective that we're very dependent on this because everybody was like,

Mike Rispoli  30:29
one thing I don't want to do is go back to hand typing the code. Like that's the, I absolutely don't want to do that. So it's very clear that this is the future for writing code, and we were able to get by when things are down, but it was really nice to say like, hey, some of the open source stuff is getting there. It's getting close anyway.

Joe Leo  30:48
Now, when you're switching between models, what is it like? Is it a big lift? Like how are you switching?

Mike Rispoli  30:54
It's not a huge, huge lift. With Ollama and OpenCode, it's pretty easy. You can just change it right in the tool. Yeah. I've been thinking about trying py.dev too. That's the other one that I have not gotten a chance to try, but someone recommended to me. If you go with like the harness like Codex, it's a little different. I think you can also switch there,

Mike Rispoli  31:13
but most of the time I just go to OpenCode and I'll run Ollama and I'll run with the model I'm running locally, or I'll run with a cloud Ollama model. That's been kind of my flow to test these things out. And yeah, it's worth having the subscription to it because sometimes I'm like, I'm just going to use Kimi for a day. Kimi asks you a lot more questions,

Mike Rispoli  31:32
so it definitely, it reads between the lines less, which can be a good thing. It'll ask you like quite a few questions. You don't need to use that skill that some people use where it's like, interrogate me until I'm tired. Kimi will interrogate you till you're tired. I found that.

Joe Leo  31:50
You just touched on this a second ago, but is there a threshold? Is there some tasks where you're just like, okay, I'm not, I'm not reaching for an LLM right now. There's something that just needs to be done. I need to either figure this out or I need to type this out or I'm going to test this out. Are there things where you're just like, I'm going it alone?

Mike Rispoli  32:09
You know, unless I'm intentionally doing it. So I do intentionally try to code a little bit each day or at least every week because I don't want to lose kind of the skill. What I will say happened, it happened with Codex when it created that PR with too many modules.

Mike Rispoli  32:26
I knew there was something wrong, but I want to say that I was working on two or three different things. I was in some meetings. I wasn't in the headspace to think about it. And there was a big part of me that said, I just want to throw this thing over the fence. And while I tested it, it works. I actually had to like go on a walk and think, okay, forget that Codex did anything.

Mike Rispoli  32:46
You're in Ruby on Rails, you're sitting in front of your ID. How would you have done it? It took like completely clearing my head and like not being polluted with what it had done. And I said, I would have done this. It would have been three functions called this. I went back to Codex and said, hey, this is what I would do. And I let it write the code and it did it. So I didn't like go back to hand typing it myself,

Mike Rispoli  33:05
but I did have to take that time and say, I'm going to go on a walk and just forget everything that was done because I need to get back into the thinking mode. I think it is easy to get lulled into this like, uh, da, da, da, da. Like, you know, you got all these things, everything's going smoothly, and then you hit one where you're, the pull request just has a very clear code smell and you're like,

Mike Rispoli  33:26
I can't let this go. And I find I have to step away to bring my brain back to focus mode on that one problem.

Joe Leo  33:33
I like that solution quite a bit. Stepping away has helped me many, many times in the past. I mean, that's why the Pomodoro was invented, right? It was so that we could step away at frequent intervals and let the background processors in our brains operate.

Mike Rispoli  33:47
Exactly. Exactly. It's especially true now because you get so stuck in that agent loop and they perform so well for so many tasks that when you finally hit a task that you can't, I think even like our website redesign was a good example of that where I knew what I wanted. I didn't know how to describe it to the LLM the right way.

Mike Rispoli  34:07
I had to really kind of think it out, draw it out, and then work with it to see what was going to work, what wasn't going to work. Again, I still use the LLM for writing a lot of the code, but I did know, like I will feed it, like this is what the CSS should be. This is how you should be structuring the HTML because I knew that it would kind of not get what I wanted and it would do something weird.

Joe Leo  34:28
Well, so on that topic, I'm interested in this because you've had a lot of experience in this. You're coming in at a point where a lot of people have tried to vibe code something or they've tried to build something on their own with AI. They've either failed or they've come up against a real blocker. I'm curious to know if there are some anti-patterns that you see or there's some common things that you're like,

Joe Leo  34:49
well, yeah, this is where you went off the rails. A lot of people go off the rails here. Is there anything that kind of stands out to you?

Mike Rispoli  34:55
Yeah, for me, it's mainly, aside from the tech stacks all being picked exactly the same. It's just like Supabase, Next.js, Vercel, Resend. Like there's a classic stack that gets picked no matter what. And so sometimes people, and I'll get these PRDs where they reason their way into using that stack.

Mike Rispoli  35:14
And I'm like, you know, there's like a way simpler way to do that than this. So there's the over complexity, but it's because the LLM or the vibe tool is centered around this one way of doing things. It can't step into a different ecosystem. So you get stuck with that. But the other piece of it is building way too much.

Mike Rispoli  35:34
Like every single thought that, oh, you know, it'll be nice, you know, it'll be nice, you know, it'll be nice. And suddenly you end up with these really crazy dashboards. They look beautiful. And it reminds me of like what Dribbble was back in the day for design, where it's just like amazing. And you're like, this is amazing. Completely impossible. It's a bill, but really, really cool.

Mike Rispoli  35:55
It would be all this stuff. And I go, where are you getting all this data from? How does that, you'd point to like one little chart and they're like, how does that work? And they're like, oh, well, it's going to do this thing, but I haven't really figured that part out. And so they build all this, they overbuild all this stuff, and then they kind of want your help making it real. But really what I need to help with is I need to rein this thing in into a much, much smaller product.

Mike Rispoli  36:16
Because the other thing that people forget is marketing is so freaking hard. It's truly the hardest thing. And if you build a very complicated product, how are you going to market that? What feature are you going to show people? You have to have a way for people to enter.

Mike Rispoli  36:31
And if there's like seven different potential entry points, how are you going to begin to carve out a place in the marketing world? You know, how will you build social, podcast, all of the landing page? You know, you have one hero section to capture everybody's attention. Your product does 20 different things. What's the thing that's going to bring people in?

Mike Rispoli  36:49
So people sort of forget and they think that the beauty of the lovable of their vision is going to make its way out into the world. And I think you're seeing this a lot is just overbuilding. And so I really try to take into account, and I think this is the future for agencies too, is you really can't just be a dev shop anymore.

Mike Rispoli  37:09
You really need to think about, of all the things you want, this is a really cool end goal when we get to like the Salesforce level, but where do we start? What's the one thing that we're going to build that gets people into the platform that you can actually market? And when people look at it, they go, I need that. Versus I don't know if this does, like if you go,

Mike Rispoli  37:28
I tell people, go to things like ClickUp, it does everything. And so some people, that's very appealing. They're like one-stop shop. I was like, try to market that. We do everything. You know, I was like, that's a very hard stance to take. And those products typically didn't start with doing everything. They captured their market in something,

Mike Rispoli  37:48
in doing one thing well and moving forward from that. So I think that from a messaging perspective, these people give themselves incredible challenges when their products are that big also.

Joe Leo  37:58
You're 100% right about all of this. It strikes me that you are uniquely positioned to talk about this because you do spend so much time, way more than most in our industry, focused on the marketing piece for cause of a kind, right? Like how are you positioning yourself? What do your business models look like?

Joe Leo  38:18
What do your products look like? And I see it, you know, I see it with the week-to-week and the Disco Discovery, right? Or the Disco, what is that? Disco.

Mike Rispoli  38:26
Week-to-week, we don't even really sell anymore, but that was our week-to-week and month-to-month. That was a product. And so that's the other thing you learn is there are products that do well for a period of time. Like we were selling week-to-week like crazy three years ago. We were like, this is going to be the thing. And then just like that, no more week-to-weeks. Couldn't sell, couldn't get anyone to buy it anymore.

Mike Rispoli  38:45
Honestly, I was exhausted. You do a year of week-to-weeks. It was like building a new, it was like building 52 products in a year. Like it was, I don't know what we were thinking sometimes, but I was like, man, this is exhausting because I may be the only person in the company that can actually execute on the week-to-week, which was design and develop. Like, yeah.

Joe Leo  39:01
Yeah. Well, that's a good day off. That actually dovetails nicely. It brings me to my next point, which is that, so you know all this, how do you teach it? You've got other developers that are not going to be so well-versed to go in and say, okay, great, customer X, glad that you built this vibe-coded thing and it does 10 different things, but that's not going to sell.

Joe Leo  39:19
It's going to really make your life hard when you try to productize these things that are right now just in sort of demo mode.

Mike Rispoli  39:25
This dovetails nicely into a yet a new series that I'm working on, which is on forward deployed engineering, specifically training, because I think this is the piece that's missing. We have about three engineers now that actually, one of them used to be an FDE with me a number of years ago. We were probably the, maybe the second company after Palantir to have forward deployed engineering.

Mike Rispoli  39:47
This is going back like eight years, maybe 10. It's actually 10 years now. Company just got acquired. So it's like 10 years ago at this point. We were forward deployed engineers. I was like, let me take all the lessons. Most of the lessons of that job also came from completely outside of software. And I'm realizing that a ton of engineers don't have any work experience outside of software.

Mike Rispoli  40:08
So there's these weird situations when you work with a customer, non-technical or technical, that come up. So I've been doing this series where I tell a story from like working at rent-a-car or going rock climbing that pairs with this other situation that happened in software that something like the developer hates me.

Mike Rispoli  40:28
They're resisting everything I want to do. Things that are very real about forward deployed engineering that you don't deal with if you were always a product engineer at a SaaS company. And that's the piece I have to teach.

Mike Rispoli  40:39
And I don't think everybody's going to like it, the job itself, I mean, because it's definitely the series I'm calling behind enemy lines because you definitely feel like you step into hostile territory a lot.

Mike Rispoli  40:51
And it can be on both sides, whether it's the engineering team or it's upper management, or you have to deploy something because they want to start getting the ROI on it and you're the guy who's supposed to be integrating this. All of these things require all these weird soft skills. And then there's the upsells too. Like a lot of engineers miss those opportunities.

Mike Rispoli  41:12
When I was a forward deployed engineer, you had to kind of look for these upsell opportunities too, where you were like, oh, actually, you know what? There's another product we have that might suit this use case, but you can't lose your credibility. That's the trickiness. Like a salesperson can blow their credibility sometimes, but as the FDE, you have to be able to gently talk about a potential upsell,

Mike Rispoli  41:33
but also not lose your kind of impartiality and credibility. How do you tow that line? So I keep thinking about this. And so I've been recording these like 10-minute episodes on these little pieces of forward deployed engineering, mainly to release, but also to train my team, because I think this is going to be something we have to really gear up for,

Mike Rispoli  41:53
because I can't be the linchpin of like all of the product. How do you do product and design and sales and marketing? How do you take all these things into account? And it really is having a business sense. And some of it is running a business will teach you this. And some of it is just, if you don't have it, hopefully the stories kind of help back up some of this stuff.

Mike Rispoli  42:13
Like, hey, this is kind of how you need to think about the accounting or the calculus of development from a business person's perspective.

Joe Leo  42:21
I really like that answer. And I like that you're building that out for your team as well as for the public. And I think if you're listening to this and you're interested in that, you should definitely check out Behind Enemy Lines. It reminds me a lot of my, I mean, I'm still a consultant, but my consulting days as an employee when I was going from project to project.

Joe Leo  42:39
And yeah, there were all of these different things that you needed to take into account. And of course, we were always on site back then. And so it was being enmeshed in the company's politics, the company's initiatives, what the company was trying to do. When the company had an all-hands, we were there. It wasn't Zoom where it was like, well, we could just invite who we want, right?

Joe Leo  42:58
Like we knew everything that was going on. We went to lunch with them. There were all these things. And so getting that experience is totally invaluable. You're right. For conversations like these, I also think that customers are getting smarter and savvier in their selection. And I can remember clearly a customer that we landed this year.

Joe Leo  43:18
They just said to me, I think it was the first or second sales call. And they said, well, thank you for pushing back on some of this stuff. Because we were just like, we want to challenge assumptions. We want to make sure that they're thinking clearly about the product development, the direction of the business, you know, all of these things. And basically what this person said to me is everybody we've talked to so far,

Joe Leo  43:38
all these other dev shops are just like, yeah, okay, we'll do it. We'll build it. We'll build it. We'll build it. That used to be a good thing, right? It was just, well, do you have the people that can do X and Y where X is a framework and Y is like a business vertical? And now it's like, hey, if you're just yes-ing us to death, we know you're just going to take what we're saying and stick it in AI and we could do that, right?

Joe Leo  43:58
So we need somebody who's going to think critically about what we're trying to build and be a partner with us.

Mike Rispoli  44:02
And you get that too. Like we've had customers where they'll actually try to, they want Git repo access and they like want to try and start building it. I'm not opposed to that sometimes, but sometimes that to me signals there's like an issue with the development because why are they feeling like they need to do anything themselves? Are we giving good enough?

Mike Rispoli  44:22
Like this goes back to those people skills. Because I'll get people, they're like, the meeting went great. I don't know. And I was like, but why are they saying they want to do this now? Sometimes the perspective on how the thing went isn't the same. And that's where I have to step in, sort of say, what's going on? And a lot of the times the product maybe was going in a direction they wanted something else or they wanted to try something.

Mike Rispoli  44:44
And of course, there's pushback, but I said, you have to really make sure they're bought into the pushback. It's not enough to just push back. You have to also sell them on it. So this goes back to the sales skills. And sometimes engineers hate that. They're like, yeah, but I'm the engineer. Like my expertise should be valuable. I go, trust me, but doctors deal with this too. Like, I was like,

Mike Rispoli  45:03
I was like, there's a genetic quad to the whole thing where you have to just sort of sell them on the idea, make sure they're bought into it. People can, if you tell someone's going to take three weeks, you can't do that anymore because they realize they're like, but I just vibe-coded it in a day. How could that be possible? So I also caution people on this too,

Mike Rispoli  45:25
is if you're not sure, just be not sure. Say, I'm going to take a step. It doesn't cost you anything anymore to take a stab at something. For all you know, it takes 30 minutes.

Joe Leo  45:34
Yeah. Yeah, yeah.

Mike Rispoli  45:35
Right. And you're like, great. Because sometimes there's a third-party service that's there. You'll run into that a lot too. You know, it is, we want to integrate with this thing. And just simply say, it might be quick. I just need to look into it. Because I don't know every third-party tool on the planet and we'll see what happens. But again, most of that stuff is all possible. You just have to be aware.

Mike Rispoli  45:54
And this is all from our engineers now really are working directly with customers. We were always at the forefront of that, but now we're especially so where we don't put any kind of project management or like layer in between. Because to me, that was a very traditional agency style where you would go to a person and then it would get like shipped off to a dev team somewhere.

Mike Rispoli  46:14
And I never wanted to be that kind of shop, but now it feels like that's a huge disadvantage.

Joe Leo  46:19
Yeah, I think you're right about that. I like what you say to your team too. I mean, really, every call that you have with your customer is a sales call, right? You think you're doing a stand-up, but you're not. You are selling them on why what you have done in the last 24 hours is okay. And you can keep doing it, right? Like every single time. So embracing that is a lot better than trying to hide behind,

Joe Leo  46:40
well, I'm just a developer and I stand behind my work. It's great to stand behind your work, but you still have to justify it because somebody else is paying the bill.

Mike Rispoli  46:47
Exactly right. And selling to engineers is different than selling to non-engineers. And that's a part of this, the thing I would say I learned earlier in my career is you have to find out there's features of all the things you're building, but what are the benefits to an engineering team versus to a stakeholder who's maybe on the marketing team? They're going to be two very different things.

Mike Rispoli  47:08
And sometimes what benefits one doesn't benefit the other. And you see, that's where the tension happens, where you get caught in the middle, where let's say the marketing team wants to migrate to Shopify, let's say, but the engineering team likes their homegrown solution.

Mike Rispoli  47:22
But the marketing team is winning because they're the ones who are saying they can raise the revenue and the CEO must kind of side with that side of things. It's all like this. And so now you're stuck smack in the middle because you need the trust of the engineering team, but they hate the marketing team because they're pushing forward this initiative. So in a lot of ways, you're like the peacemaker and the negotiator of both scenarios too.

Joe Leo  47:43
That's true. Mike, it's been great having you on the show. We'd love to have you on again. I'm sure Valentina would love to talk to you about especially some of the work you've done with different models. And we'd love to hear more about cause of a kind and how it's growing and changing and developing. I know I'm very interested. Also looking forward to seeing you tonight at Artificial Ruby.

Joe Leo  48:03
So if you're around, if you're in New York City, come on down.

Mike Rispoli  48:06
Very excited. Yeah, we're very excited. I don't get to the city often enough, so I'm excited. You know, it's one of those things I look forward to every month is getting to Artificial Ruby.

Joe Leo  48:14
Yeah. I love having another Long Islander in the room.

Mike Rispoli  48:17
Absolutely.

Joe Leo  48:17
Makes me feel like I'm back at home.

Mike Rispoli  48:19
Yeah, yeah. Same, same.

Joe Leo  48:22
All right, Mike, thanks very much. And we'll talk to you again soon.