I recently saw a show that mentioned something called GStack — a framework created by Gary Tan, the president of Y Combinator. GStack is an open-source project designed to help solo founders use Claude Code (Anthropic’s AI coding agent) as a multi-role startup team. It structures AI around the different jobs a startup needs done — engineering, product, research, operations — using configuration files that tell the AI how to behave in each context.
I read through the documentation, studied the architecture, and had a moment I wasn’t expecting.
I’d already built something that does more.
Not because I set out to compete with Y Combinator. I didn’t even know GStack existed until a few days ago. I built it because I needed it. And that — honestly — is the part of this story I think matters most.
Let me back up.
I run multiple businesses. Epic Adventures VR is a virtual reality gaming venue. Epic Experiences LLC is the parent company. Epic Worlds LLC holds intellectual property. Terranova Station is a $10 million immersive entertainment concept I’m actively developing. I have business partners, vendors, booking systems, staff schedules, financial data across four QuickBooks companies, customer databases, marketing campaigns, an entire venue’s worth of VR hardware to manage, and kids to raise on top of all of it.
I don’t have a CTO. I don’t have an IT department. I don’t have a personal assistant.
I have Claude.
About a year ago, I started building what I call Project Jarvis — a custom AI infrastructure that turns Claude into something that goes far beyond a chatbot. It’s a network of PHP proxy endpoints hosted on my own server that give Claude real-time access to my entire operational world. Gmail. Google Calendar. Google Drive. Asana. QuickBooks. My booking system. My VR headset fleet. My website’s CMS. My server’s cPanel and WHM. Pushover notifications to my phone. A session journal that maintains continuity across conversations. A lessons database that tracks what works and what doesn’t across hundreds of interactions. A skill library of repeatable workflows. A daily briefing system. A file proxy with federated search across local files, Google Drive, and Asana simultaneously.
Every one of these was built because I hit a wall and needed to get past it. I didn’t plan an architecture — I solved problems, and the architecture emerged.
And here’s the thing about GStack: it’s a collection of CLAUDE.md instruction files. It tells Claude Code how to think about different roles. That’s genuinely useful — especially for someone starting from zero. But it’s configuration. It’s telling the AI how to think. It doesn’t give the AI hands.
Project Jarvis gives Claude hands.
When I say “check my email,” Claude searches my Gmail through a custom proxy and summarizes what matters. When I say “create a task,” Claude adds it to Asana, creates a calendar event, and links them together. When I say “how are bookings looking this month,” Claude pulls real data from my booking system, compares it to last month, and tells me if I should be worried. When I say “deploy the gallery watcher,” Claude executes a documented workflow that updates production code on my live server.
GStack says “here’s how to think like a product manager.” Project Jarvis says “here’s the product data, the customer feedback, the revenue numbers, and the task backlog — now think like a product manager.”
I don’t say this to diminish what Gary Tan built. He’s solving a real problem for founders who are just starting to leverage AI. But I think there’s something important in the fact that a guy running a VR venue in Vancouver, Washington — not a Silicon Valley insider, not a programmer by training, not someone with a CS degree or venture capital connections — built a more functional AI infrastructure by simply refusing to accept limitations.
That’s the outlier’s advantage. We don’t know what we’re “not supposed to” be able to do. So we just do it.
The irony isn’t lost on me that I built this entire system through conversations with the same AI that the system is designed to enhance. Claude helped me build the infrastructure that makes Claude more powerful. We wrote the proxies together. We debugged them together. We iterated on the architecture together. It’s a collaboration that would have sounded like science fiction five years ago, and today it’s just… Tuesday night at 1am, fixing a cron job while eating leftover pizza.
I’m not sharing this to brag. If you know me, you know that’s not my style. I’m sharing it because I think there are a lot of people out there — builders, tinkerers, founders, outliers — who look at frameworks like GStack and think, “That’s for real developers. That’s for people who went to the right schools and know the right people.”
It’s not. The tools are available to everyone. The only thing that separates the people who build from the people who don’t is the willingness to start before you’re ready and figure it out as you go.
That’s what I did. That’s what I’m still doing. And if you’re reading this thinking, “I could never build something like that” — respectfully, that’s the same thing I told myself before I started.
You can. And you should. The odyssey is worth it.
