One Human, One Company: The Organization-as-Code Thesis
For fifteen years, I've been obsessed with one question: How do organizations actually work?
Not the org chart version. Not the management consulting version. The real version — the messy, political, human, beautiful, broken reality of how groups of people coordinate to create value.
And now I'm asking a follow-up question that would have sounded insane five years ago:
What if one person could encode all of that into a system?
The Thesis
Organization as Code (OAC) is the idea that every function of a traditional organization — sales, operations, strategy, quality assurance, customer service — can be mapped into executable, testable, improvable structures.
Not automated. Not replaced. Encoded.
The difference matters:
- Automation says: "Remove the human."
- OAC says: "Make the human's decisions programmable, auditable, and scalable."
Why Now
Three things had to converge:
- LLMs that can reason — Not just pattern-match, but actually navigate ambiguous situations with contextual intelligence
- Cheap, fast inference — Running an "organization" of AI agents is now economically viable
- A builder stubborn enough to try — That's me, apparently
What I'm Building
The AI insurance agent is the first test case. It's not just a chatbot — it's an entire sales organization encoded into agent architecture:
- A conversation agent that handles customer interactions
- An evaluation agent that scores conversation quality
- A strategy layer that decides which approach to use
- An observation system that tracks everything for learning
Each of these maps to a role in a traditional organization. But instead of hiring, training, and managing humans, I'm writing code, running experiments, and iterating daily.
The Honest Part
It's harder than I expected. Not technically — technically it's fascinating. But emotionally? Building a company of AI agents while sitting alone in a room in Cluj-Napoca is... isolating.
There's no team to celebrate wins with. No colleague to rubber-duck a problem. Just me, the code, and an uncomfortable amount of coffee.
But that isolation is also clarity. There's nowhere to hide from bad results. No politics to navigate. No consensus to build. Just: does it work, or doesn't it?
The Road Ahead
I don't know if OAC will work at scale. I don't know if one human can truly run a company this way. But I know the experiment is worth running.
And I'm going to document every step of it here. The wins, the failures, the moments of doubt, and the occasional breakthrough.
Welcome to the build.