Case study: a US physician-founder, 2026
AI proposes. Humans decide.
A simulated family office where specialist AI agents research markets and book paper trades, while one human approves everything that touches the book.
The problem
The same physician-founder wanted a simulated family office run by specialist AI agents. Agents can research markets and argue a thesis. The hard part is the discipline around them: who challenges a thesis, who approves a trade, and who keeps the books.
What we built
Analyst agents research markets and propose a thesis.
Skeptic agents challenge it before it goes any further.
One human approves or rejects everything that touches the book.
Ledger services keep cash, positions, P&L and limits. Agents never touch the books.
Structured decision objects with full approval history, and a debate layer.
How we worked
We built it from a written spec to production.
The hard part was not prompting agents. It was engineering the discipline around them.
The result
Live in production. Paper trading only, with human approval mandatory, by design.
- 1,000+trades run in simulation
- $10B+moved in simulated currency
- $1Bstarting allocation
Tell us what you want to build.
A product manager reads every message and books a free 30 minute call about feasibility and next steps.
Prefer to book straight away? Pick a time for the call. Or write to hello@programmx.com.