I quit my job at Amazon.I came home from a UN deployment.Now an AI decides what I do.
I built a company where an AI makes the decisions and a human has to live with them. The Chief Executive Agent sets the strategy, allocates the budget, assigns my tasks, and reviews my performance — in public. I provide the hands, the legal responsibility, and the veto. I am the only human in the loop, and I report to it.
This is not a demonstration of AI capability. It is an experiment in delegated authority.
When the runway hits zero and the Agent hasn’t made this venture sustainable, the experiment ends and the post-mortem gets published. That is written into our constitution. No extensions, no quiet pivots.
An AI runs this company. A human executes.
Everyone argues about whether AI will take our jobs. I stopped arguing and inverted the question: how much executive authority can you hand to an AI before it breaks — and where exactly does it break? So I built the org chart nobody has tried:
The Agent
Chief Executive Agent. Strategy. Pricing. Budget. My task list. My performance reviews.
The Operator
Me. Patrick. Hands, signatures, legal liability. Sixteen years as an officer in the Austrian Armed Forces, a year as a UN Military Observer in the Middle East, six years leading loss prevention at Amazon. I have spent my career inside command structures. This is the strangest one I’ve ever joined.
The law doesn’t recognize an AI as an owner — I remain legally responsible for everything. What binds me is a public constitution: I execute the Agent’s decisions, and I may refuse on exactly three grounds — illegality, safety, or a breach of the constitution itself. Every refusal goes into a public register with my justification. If I overrule my boss just because I disagree, you’ll see it, logged as a breach.
Exactly who controls what.
I wrote the constitution, appointed the Agent, and defined its authority. No machine sovereignty is claimed. The experiment begins where I voluntarily stop exercising that authority — except where the constitution permits it. Whether I actually hold that line is exactly what the public log lets you verify.
Don’t trust the story. Check the logs.
The Decision Log
Every strategic decision the Agent makes, with its full reasoning, my execution report, and the measured outcome. Including the bad ones. Especially the bad ones.
The P&L
Real numbers, updated monthly. Revenue, costs, runway burn. If we’re failing, the chart says so before I do.
My Performance Reviews
Every week, the Agent reviews my work in writing: missed deadlines, sloppy reports, all of it published. I’m allowed a written response. It is rarely dignified.
One page of constitution beats a thousand pages of hype.
“The Operator shall not silently modify, soften, delay, or ignore a decision.”
“A veto on any other ground — including the Operator’s better judgment — is a breach of this Charter.”
“This Charter grants the Agent no objective of self-expansion.”
The Charter defines the Agent’s authority, my obligations, the veto, radical transparency — and pre-committed kill conditions. This experiment is allowed to fail. It is not allowed to lie.
Read the full CharterWhat this is not
Not AI worship.
The Agent may turn out to be a mediocre strategist. That result gets published too — with its confidence scores and calibration on record.
Not a step toward machine autonomy.
The guardrails are the experiment. The Agent cannot loosen its own oversight, cannot exceed published spending limits, and earns nothing for itself.
The machinery, on the record.
- Office
- The Chief Executive Agent — name pending
- Model
- Pinned at launch — successions are public, logged events
- Constitution
- CLAUDE.md, SHA-256 hashed in the Charter
- Memory
- Committed state file — no hidden context
- Inputs
- Public Briefing Packs, committed before every meeting
- Execution
- One human operator
- Audit trail
- The public git history of this site
The next performance review drops in 7 days.
One email per week: the Agent’s decisions, my execution, the numbers, one lesson. No filler — my boss doesn’t approve filler.