Startup Operating Simulator

This experiment is a startup operating simulator with memory. You start with one plain-English business prompt, and the system creates a company around it.

It begins with two hard-coded founder personas—a CEO and a CTO—stored with their roles, skills, reporting structure, and a seniority score from 0–100. That score affects their influence on decisions.

After setup, the simulator runs in cycles. In each cycle, it evaluates the full state of the company, including open decisions, tasks, deliverables, team members, and the current phase. It sends that state, along with the sender’s context, to the model (OpenAI) and requests a structured plan for what should happen next. For example, business personas put more weight on business outcomes, while technical personas put more weight on technical outcomes.

The returned plan can include which decision should move forward, what work should be assigned, which deliverables should be produced, and whether hiring is required.

Before anything is saved, checks are applied. Decisions must pass through proposal, contest, synthesis, and decision across multiple rounds of model calls. The platform blocks repetitive, low-value messages, checks evidence coverage, prevents premature decision closure, and forces small, reversible steps when teams get stuck in loops.

Once validated, updates are written to persistent state (the company context is updated): decisions advance, tasks move, deliverables are attached, and system messages are posted. Hiring is stateful too. When a capability gap appears, the system opens a requisition, evaluates candidates, hires one, and adds that person to both the sandbox team and the org hierarchy.

Across cycles, the company advances through phases while recording metrics such as loop rate, novelty, decision velocity, evidence coverage, and artifact-backed execution quality. The dashboard is a live view of that evolving context.