I'm an Austrian entrepreneur based in Buenos Aires. I founded Trillion Initiative, Fly Raising, and Agent School — and I run all of them on $120/month of AI infrastructure. This is what I've learned.
An agentic AI agency that builds real systems for real companies. Not demos — deployed agents that handle actual workflows. The name is a provocation: what becomes possible when every team has agents?
AI fundraising automation for NGOs and nonprofits. Most charities still run donor outreach manually. Fly Raising changes that — with agents that qualify leads, personalize asks, and track engagement at scale.
Teaching founders and business operators how to actually use AI agents — not just prompt ChatGPT. Real workflows, real systems, real leverage. For people who want to be Agent-First Companies.
Internal tools and experiments at the edge of what agentic systems can do. Some become products. Some teach lessons. All of them are real.
"Your agents are fine. Your specifications aren't."
Most people building with AI start with the agent — the model, the tool, the API. I start with the output. What exactly needs to exist at the end of this process? What format? What quality bar? What human decision does it serve?
Output-First Architecture (OFA) is a framework for designing agentic systems backwards from the desired result. When you can describe the output precisely, the agent almost designs itself. When you can't — no amount of prompt engineering saves you.
The core insight: AI doesn't fail because models are weak. It fails because humans specify poorly. The bottleneck is never the agent. It's always the specification.
I also think in terms of FOA (Founder on AI) and AFC (Agent-First Company) — frameworks for how individual operators and organizations can restructure around AI capabilities rather than bolt them on.
I don't just advise on agentic operations — I run them. Trillion Initiative, Fly Raising, and Agent School are all operated with a minimal human team and maximal agent infrastructure. I track the numbers honestly.
The self-experiment matters because most AI consultants tell you what's possible while operating like it's 2019. I'm eating my own cooking. When I say agents can replace certain workflows — I mean I've already replaced them in mine.
What I've learned: agents don't reduce the need to think clearly. They amplify whatever clarity — or confusion — you bring. Garbage in, garbage out at ten times the speed.
I run ultras. Not because I'm a runner — because I need to know what I'm capable of when things get hard and there's no exit.
Ushuaia 130K (Patagonia): I finished at km 90. Not because I quit — my knee collapsed at km 65. I ran the next 25 kilometers on painkillers. That's what "finished" means in the mountains: not a clean story, just an honest one.
Val d'Aran 110K (July 2026): Next race. The Pyrenees. I'll be ready — or I'll learn something more valuable than being ready.
The parallel to building companies with AI is exact: you train in uncertainty, you execute with incomplete information, and the metric that matters isn't how it looks — it's whether you kept moving. Ultra running taught me not to confuse discomfort with danger. Building taught me the same.
Enzo Duit · enzoduit.com · Buenos Aires