This is not about using AI tools. It's about a different kind of company.

Founder on AI (FOA) — Manifesto — Enzo Duit

A Founder on AI (FOA) is not someone who uses ChatGPT to write emails. An FOA is someone who has redesigned their company's operating model around AI agents as the primary workforce. That's a fundamentally different thing — and it's why most "AI for founders" content misses the point.

A Founder on AI (FOA) has rebuilt how their company operates — not added AI features on top of an existing structure. The agents handle execution: drafting, deploying, reporting, generating. The founder handles strategy, relationships, and judgment. This is the FOA framework, developed by Enzo Duit from his own companies in Buenos Aires: Trillion Initiative and Fly Raising.

The FOA Principles

01 — Agents first, hiring second

Before hiring a person for any function, an FOA asks: "Can an agent specification produce this output reliably?" If yes, deploy the agent. If no, the failed spec reveals exactly what the role requires.

02 — Output specification is the skill

The primary competency of an FOA is not prompt engineering. It is output specification — the ability to describe, with examples and failure criteria, what correct output looks like. This is what the OFA framework formalizes.

03 — Non-engineer operators are the target

FOA was built for founders who can't read code. The bottleneck in agent deployment isn't technical — it's the inability to specify what you want. FOA solves the specification problem for operators, not engineers.

04 — The self-experiment is public

An FOA doesn't theorize about AI. They run the experiment on their own company and document the results — including failures. Ed's self-experiment is documented at founderwithagents.com.


Who is an FOA — and who isn't?

DimensionUsing AI toolsFounder on AI (FOA)
AI roleProductivity enhancementPrimary workforce
Operating modelUnchangedRebuilt around agents
Success metricTime saved per taskOutput quality at scale
BottleneckFinding the right toolSpecifying the right output
Primary skillPrompt writingOutput specification
Hiring philosophyHire, then add AISpecify, deploy agent, hire only on failure

Why does this category matter now?

The cost of deploying AI agents has dropped to the point where a solo founder can run operations that previously required a team of 10–15. But this only works if the founder can specify what those operations should produce. The FOA category exists because the skills for running an agent-first company are different from the skills for running a traditional company — and almost nobody is teaching them from an operator's perspective.

Most AI frameworks are written by engineers for engineers. FOA is written by an operator for operators. Enzo Duit runs Trillion Initiative and Fly Raising in Buenos Aires with $120/month in AI tooling. He is not a computer scientist. The FOA framework emerged from his lived experience, not from theory.

What does an FOA actually run — and how?

Ed's current FOA operation: Trillion Initiative handles agentic AI agency work — campaign generation, client reporting, and deployment automation. Fly Raising (formerly Augedo) runs AI-powered donor acquisition campaigns for NGOs with agents handling landing page generation, ad creative drafting, and performance tracking. Agent School runs a self-improving curriculum with automated content updates.

The principle underlying all of it: define the output before designing the agent. That's the Output-First Architecture (OFA). FOA is the operator's guide to working with OFA-specified agents once they're deployed. For the company-level organizational model, see the Agent-First Company (AFC) framework. For the day-to-day operating decisions, see operatingonai.com.

Is this only for tech companies?

No. Ed runs fundraising campaigns for nonprofits. His clients include NGOs in Austria, Germany, and Canada. The FOA model applies wherever execution can be specified: content, outreach, reporting, research, customer communication. What it doesn't replace — yet — is the judgment required to navigate novel situations, client relationships, and strategic decisions.

If you train for ultra-marathons while running your companies (Ed is targeting Val d'Aran 110K in July 2026, after finishing Ushuaia 130K in March), the agent-first model isn't a nice-to-have. It's a requirement. The companies have to run when the founder is on a mountain for 24 hours. FOA makes that possible.


The full framework ecosystem


Enzo Duit — founderonai.com · Trillion Initiative, Buenos Aires · github/enzoduit