Vittorio DistefanoITAgentic coding, not vibe codingFounders who buildServer linux veteransAll in on day oneAgents run as a teamIt feels like being a kid againbuilder/builders/vittorio-distefano/avatar.jpg
From BASIC games on a Zenith SuperSport to SolverForge, a production constraint solver in Rust built with his own agent harnesses, and a plan for fleets of Omarchy machines.
first computerA Zenith SuperSport laptop with an 80286 at 25 MHz and DOS
omarchy since2025
modelsgpt, claude
// shorts
Shorts
5 videos
0:29BASIC games on DOS: open the source, change a line, see it instantlyWhere it started: free BASIC games on DOS with the source in the box. Change a string or some ASCII graphics, and the game changes on the screen immediately.0:44Sway for years, then Omarchy: a cool machine out of the boxYears on Unix and the terminal, tiling with Sway on his machines and ricing it from scratch every time. He found Omarchy two days after DHH announced it and calls it a cool machine out of the box on the same paradigm.0:25Specless from day one: coding with Claude Code before the specsHe was already coding with agents the moment the Claude Code CLI harness arrived, without endless function signature specs. His view: early AI failures came from too many specs, which he found backwards.0:19A Ralph loop before it had a name: a PRD and a script calling ClaudeBefore he open sourced it, his superpower was a Ralph loop he ran in early Claude Code, before it had a name: a PRD, an agent, and a script that calls Claude recursively. It made him so fast.0:20Omarchy as the platform: bring libraries to customersWhy Omarchy is a fantastic platform for him: a new operating system where you can publish any application and bring libraries to user applications. His goal is to bring it to customers and do deployments of Omarchy on fleets.
// life timeline
First computer, a Zenith SuperSport laptop with DOS
Starts programming by modifying BASIC games
Neighborhood IT support in the 90s
System engineering and system administration
Copilot in VS Code, code from the web chat
Builds his own agent harnesses around Claude
2025Finds Omarchy two days after DHH announces it
2026Bootstrapping SolverForge, a constraint solver in Rust
// 00
The rundown
Vittorio Distefano is a rock star. He started out modifying BASIC games on a Zenith SuperSport, spent years in system administration, and today he is bootstrapping SolverForge, a production constraint solver written in Rust and built with agent harnesses he wrote himself. He found Omarchy two days after DHH announced it, and it has been his standard setup ever since:
"I can spin a VM in one minute."
“I can spin a VM in one minute.”Vittorio Distefano · L78
// 01
BASIC on a Zenith SuperSport
Vittorio got into computers early. His first machine was a laptop, a Zenith SuperSport with an 80286 at 25 MHz running DOS. It was gray and heavy, with a big battery on the back and one of the earliest LCD screens, the kind where you could see the liquid crystals and had to handle it with care. The bezels were enormous, with physical brightness and contrast controls.
He started with video games and then started programming in BASIC. DOS came with free games written in BASIC, source code included, so he could change a string or tweak the ASCII graphics and see the result on screen right away. That immediate feedback got him hooked on being technical.
// 02
From the neighborhood to system administration
From there he moved on to a better computer, then an even better one. In the 90s he was already doing IT support around the neighborhood. Then he went into system engineering, or better, system administration as a job. Since he started using Unix he has always worked in the terminal.
Today he is self employed and bootstrapping a company built on a Rust library that solves any kind of planning problem, from shift scheduling to vehicle routing.
// 03
Two days after the announcement
Before Omarchy, Vittorio was already running a similar setup with Sway, a tiling window manager, but he had to rice it from scratch every single time or maintain complex dotfiles. He found Omarchy about two days after DHH announced it and was soon watching how it was being developed. "it allows me to have a cool machine out of the box." It was the same paradigm he already used.
He liked it so much that he copied some of its patterns, like the Super Space menu, and ported them to other distros. His solverforge-linux project is a configuration layer for Sway inspired by Omarchy. Omarchy became his standard configuration.
// 04
An agentic mindset before agents
When GPT first came out, Vittorio tried it without taking it too seriously. With 3.5 and its API, he saw that reasoning could be automated. He asked the web chat for code templates, pasted them into his projects and changed everything by hand. Then came Copilot in VS Code, autocomplete on steroids, and he started pasting generated functions back into the chat and asking how to make them better.
What he noticed early was that he could stop straining at the point where a code base got too big for his head and let the tool take it from there. He never wrote endless specs. He asked for the outcome. "So, yes, I was copy-pasting, but it was already… Agentic mindset."
When Claude Code arrived, he was already working that way. Instead of contributing to it, he built his own harnesses, three generations of them. One called Claude recursively from the CLI, working from a PRD in a loop so the work kept going while he slept, long before that pattern had a name. "Basically, this was my superpower before I open-sourced it"
// 05
SolverForge
That speed is how SolverForge exists. It is a constraint programming framework and solver in Rust, with a huge surface, low level and production grade, built in about two years and grown out of AI from the start. Most people never think about constraint solvers until they have a planning problem, and that is exactly why Vittorio sees Omarchy as the place to bring them to users. "It's the perfect platform to bring For example, libraries, and find a way to apply them to user applications."
He also thinks Omarchy is easy to get into. Installing it is a breeze now. The real shift is in how you think about a computer, and if you just stick with it, it comes.
// 06
Fleets of Omarchy machines
His goal is big. "I want to do deployments of Omarchy fleets." He wants production fleets of Omarchy machines running at customers. He wants to build scheduling applications that take his library into the Omarchy UI, and he has already tried it with pull requests for the core calendar and OmaStorm. Calendar planning, vehicle routing, anything combinatorial and even CAD are all on his list.
For now the focus is SolverForge. Like any good builder, he knows the work goes on. "It will be finished, but it will never be finished."
L1[redacted: outside the builder story (call setup, family, interruptions, off topic), lines 1 to 38]
L39Nix: So, Vittorio, what was the first computer that you ever, ever got to play with, and what saw, and how did that change your… your trajectory? Because I'm guessing at some point you saw computers, and you're like, that's what I'm gonna do.
L40Vittorio: So basically, I got a laptop, It was an 80286. 25 MHz. with DOS, I think, 4.0, and that was a laptop. That was a big thing.
L41[redacted: outside the builder story (call setup, family, interruptions, off topic), lines 41 to 42]
L43Nix: Is it one of those big luggables, like the suitcase big?
L44Vittorio: It was a Zenith SuperSport. I could even look it up for you. It was, like, a gray laptop, this big. It had a big battery. Bet. On the rear, and you could stack it with a coprocessor, then in subsequent, models. But basically, the famous version of.
L52Vittorio: LCD screen, very, the very… the earliest where you could see the crystals, the liquid crystals. Very, very… delicate… you had to be very delicate when handling it, but it was… And that meant I basically had…
L57Nix: And the bezel on it is huge. The bezels are enormous.
L58Vittorio: bezels were… were enormous. You had the lower, the lower part was, like, Was like this. Just… just to be able to house, some brightness and contrast, you know, controls, which were physical. At the time. for those who… You surely have seen them, but very young people never saw them. And so that was the first computer. I started with video games, but then I started programming with BASIC. That was, my start, the way I got into being technical, because on DOS, you had some free games written in BASIC, where you had the source code, and you could modify them. in the beginning, it was maybe just a string, or just changing a bit of graphics, which were sometimes made of ASCII characters, so you could… You could have a visual feed… an immediate feedback. Because what you would change would change on the computer. seat immediately, which was rewarding as a kid.
L59[redacted: outside the builder story (call setup, family, interruptions, off topic)]
L60Vittorio: I mean, it was useful because you can do strings. Really, the basics of programming, if you want. And then from there, I got a better computer, And the… an even better one. Already, I was working in the neighborhood as IT support. at the time, in the 90s. And then I went into… System engineering, or better system administration as a job.
L62Vittorio: Today, today, I, I am, self-employed, I run a company, I'm bootstrapping a company, based on a, on a library written in Rust that allows you to solve any type of planning problem. For example, shift scheduling, or vehicle routing, every type of problem. Has… that is based on finding the right combination.
L63[redacted: outside the builder story (call setup, family, interruptions, off topic), lines 63 to 72]
L74Vittorio: I found Omarchy, because I am, I, I, I, I, I will… since I started using Unix, I always worked on the terminal, and I was already using a similar setup. since a lot of time with, it was not Hyprland, but it was… and it is on some machines still, it is Sway, tiling Window Manager, very similar to Omarchy, but I had to rice it from scratch every… single time, or have complex dot files. Right. I found Omarchy basically 2 days after DHH announced it, and I was already looking how he was developing it on the website. And, it allows me to have a cool machine out of the box. And it's already the same… paradigm that I use. And it's fantastic.
L76Vittorio: already what I wanted. It came to the point where I copied some patterns, for example, for the basic… super spacebar menu, I copied it, I ported it to other distros.
L78Vittorio: I like it, so it has become my standard configuration. And I can spin a VM in one minute.
L79Nix: I gotcha. So, talk to me about… talk to me about… do you… I'm guessing you were an early AI user, I'm pretty sure about that. Okay, so talk to me about where, like, about the time frame when you were using, like, ChatGPT, and you were probably cutting and pasting code and cutting and pasting things because there weren't any integrations, right? So, talk to me about, like, when you were cutting and pasting, and it really wasn't that good at code yet, and what were… how were you doing in that time?
L80Vittorio: So, I guess this is the… 2023 slash 2024, GPT-3 GPT-2, when it came out, I didn't take it seriously, I tried it. Yup. But I didn't have any intuition, to be honest. But, with 3.5, when… also when the API… and 3.5, when the API start… documented. I intuited that it would be possible to automate reasoning in the future. I didn't have a clear idea. What I was doing was opening, really, the web chat. the same… the same interface we have today, but in the earlier version, and I would… I would, I would… ask a question, and get a code template. Then I would take the template, copy it into my project. and change… everything manually. Then came Copilot, which added a sort of… I call it autocomplete on steroids, where you had, in comparison to the old IntelliJ line completion, you had snippets. So I had, VS Code with Copilot that would give me auto-completion and would allow me to write entire functions. And the second step was, therefore, not asking the question to the chatbot, but being said, generated copilot functions into the web interface, and asking, how can they make it better, or… How do I have to change this code? To achieve an outcome. made in that particular moment. So that… and that lasted very long. That lasted… Six months, like this. And I wouldn't, I wasn't capable of, probably of running this in production. I was doing pet projects. I was still an employee. I was working in a consulting firm. So I was doing that.
L81Nix: The one thing I heard you say that I love, and there's so many people that say it, is the phrase. How do I make it better? Talk to me about that.
L82Vittorio: Yeah, it's… Yeah, well, there is a point, clearly. When you, when you say you would, you want to develop a patch of code, even smaller. You start from, from… on an empty slate. I'm talking from, about manual coding now, how it was back then. You would start from a blank slate, you would, Be able to write an estimate of working code, but you had to really think about it. do lots of rounds of iterations before you could Figure out how we could move stuff to optimize it. For an objective. what I understood, very early…
L83[redacted: outside the builder story (call setup, family, interruptions, off topic), lines 83 to 85]
L86Vittorio: Yeah, a, transition between, between, Well, make it better, yes. Making code better.
L88Vittorio: Behind coding, you would mentally do various rounds of reviewing your code. I noticed immediately in my own pattern of working, that there is an initial part where, when you're experienced, you Optimize quickly, and you change your code quickly. Then, the second point where, If you really strain your head, then after a while, you are capable of optimizing your code in any way. And what I realized was, hey man, This tool here, maybe? I still need to… Give… a direction to the code base in a manual way, but… I can stop stressing about that point when it becomes too difficult for my brain. Right? and I can just drop. And it was already a big transition, now it's laughable, but the possibility to just upload 300 source code files to the interface… and analyze them. Was already… A big boost, in my opinion, because there's no way that those three files fit in your head, right?
L90Vittorio: So, that was, That was a step, but one thing I want to say about the methodology of making it better is that I hear many people, I don't know how you… experience it. I hear many people, or most people, writing or say. Nowadays, in 2026, you can give an agent. a task, for example, build me a web page about computer keyboards without giving specs, and it will figure it out. And then they say, but before. It was so difficult, you had to write spec files in Markdown. I don't know how you experienced it, but me, even back then. I would just write to the chatbot, please find It was not make it better, it was… no. Please take your time to analyze this routine and find the best way to improve the code without giving. You know, explicit specs. It was working like this back then, already.
L91[redacted: outside the builder story (call setup, family, interruptions, off topic), lines 91 to 93]
L94Vittorio: Most… basically, to make… I will summarize what I said. Most people say that, the possibility For us. To use an agent in a way that we give a task without giving very detailed specifications. We just ask about the outcome. Most people say that this is available since recent time. I don't know how you experienced it, but honestly, I was already… working almost speculess with GPT-4.
L96Vittorio: Or, with the first Claude Code that came out. when the first… when Claude Code came out with the CLI tool… with the CLI harness. I was already coding like this. I was not writing… signatures. Endless lists. Frankly, I think a lot of early instances of scarce success with AI could be, could be caused by too many specs. Because I found it always backwards, I don't know how you… So, yes, I was copy-pasting, but it was already… Agentic mindset.
L97[redacted: outside the builder story (call setup, family, interruptions, off topic), lines 97 to 99]
L100Vittorio: I didn't contribute to Claude Code itself. Because how I was working was different. I was developing my own harnesses. I even have one on GitHub. I… I… actually, I… Three generations of harnesses. For example, there is this harness I used. I used Claude basically as an endpoint that would be called In a similar way, how agent-to-agent
L101[redacted: outside the builder story (call setup, family, interruptions, off topic), lines 101 to 109]
L110Vittorio: I didn't contribute to it, because I immediately started developing a harness and I was using it this very… with… by calling Claude inside the CLI recursively. Basically, this was my superpower before I open-sourced it, because I basically had Ralph Loop At the time of early Claude Code, when it wasn't… called the Ralph Loop. I would have a PRD, and I would have an age, and I would have a script called Claude, recursively, so that I could, go to bed. And this allowed me to really be… It made me… So fast, so fast. that I… if you go and look at SolverForge, I don't claim to be special, but it's a huge surface, and you will see it's a production library. Is impor… that will be impossible to do. As a sort of dev, let alone in Two years that, One and a half years, two years that they took. So, But that's not the focus per se. The focus is really on the tool here, right? Because we are talking about operating systems and AI. So, If someone is looking for a production-grade system that is large. It's also low level, because it's in Rust. And that was… that is basically AI native, and that grew out of… really the womb of AI, because I started with GPT-3, then I have it. So, in that sense. And more and more proof will come. Because this is a niche software in any case. You're right, the general public doesn't care about constraint solvers planning problems unless they directly have it, but Curious what, about user-facing, applications. 2, now, And what can be done. Because this is all, my work is mostly back-end, so to speak. Yeah, yeah.
L111[redacted: outside the builder story (call setup, family, interruptions, off topic)]
L112Vittorio: Which is why Omarchy is a fantastic platform, because it's a new operating system, it allows you to Publish. Whatever application you want. Basically, this is the… where… where… Virgin. It's the perfect platform to bring For example, libraries, and find a way to apply them to user applications.
L113Nix: Right. Talk about the ability for people that are new to get in and jump in. Is it easier, or is it harder? Because I see technology going up and up and up, and people feel left out and farther behind, but in some cases it's getting easier, in some cases it's getting harder.
L114Vittorio: I, I think, I think the… Time shift. Between what you expect of a computer, graphically, really, as a paradigm, you expect people have been taught about You will find a desktop with folders that you see, and that you move. I think the paradigm shift From that preconception to Omarchy is more difficult than materially installing it, because really installing it It's ridiculous by now, it's not air. it's even misleading to openly say it's Arch linux, because it's… Nowadays, it's an auto-install, so to get into it. I think it's very easy if you're young, because if you're young, if you are 16, If I was 16, I would want Omarchy, because I would be without preconceptions and accumulating stuff. years and years of being old, like, sometimes happens now. if I didn't know about Unix at all, right, and I was starting out, It would be perfect. it still is nowadays. It's pretty easy to jump if you already use linux, and just requires you to, I think. You just have to stick. If you come from Windows, you just have to stick with it. And your energy will be spent with the paradigm shift. and that is easy. I think.
L115Nix: Right. All right, as we… as we kind of wrap up, tell me what you're working on, and the projects that you see in the future, and how you think… where you think all this is going. Like, where… where do we end up from here? It's crazy.
L116Vittorio: I really, really, really, really, really will do everything possible to contribute to a situation where there are production deployments of fleets of Omarchy machines. I want to bring it to customers. I have to learn how, but that is the idea. So, I want to do deployments of Omarchy fleets. I want to build applications, I hope scheduling applications, which are my thing. Basically, I take my library and I bring in the Omarchy UI. I have tried it with some PRs, one for the core calendar, one on OmaStorm, I'm still gauging. Because in the end, I will come with a planning application. So it can be a calendar, or, or… or whatever. I'm still… the domains are calendar planning, or… vehicle routing. basically anything, combinatorial, or CAD. I'm also working on some… on some CAD applications. I'm just… Trying to understand if it's worth… connecting a constraint solver to a CAD when we have models which can operate a CAD directly now. So, I am… for new projects, I am brainstorming, like everyone, too many ideas. My focus, I'm trying to keep on bringing what I already have, which means SolverForge. weeks, so… If you wanted something concrete as a project on Omarchy, it's not there yet.
L117Nix: Well, it's… it'll never be finished, right? That's the problem that we have.
L118Vittorio: It will be finished, but it will never be finished.
L119[redacted: outside the builder story (call setup, family, interruptions, off topic), lines 119 to 137]