Researchers have introduced MiniCorp, an office simulator designed to study how AI agents can collectively run a company while generating enterprise data at scale. The work targets what the authors describe as the last mile toward enterprise AGI: a company that runs itself.
Training and adapting such agents requires longitudinal enterprise data, which the paper notes remains scarce, costly to acquire, and often restricted by privacy constraints. Historical archives are also frequently incomplete, recording only what actually happened and offering no view of alternative decisions.
MiniCorp addresses this by connecting two interacting worlds. The external world models customers, dynamic competitors, and market mechanisms. The internal world consists of agents that observe events, discuss their options, and make strategic decisions. Those decisions have lasting effects on the market, and the resulting feedback informs the firm's later decisions.
As the firm and market interact, MiniCorp continuously records the agents' communications and decisions, preserving both the information available at the time and the business results that followed. Checkpointing allows the same situation to be replayed under different decisions, providing comparisons unavailable in static archives.
The researchers evaluate end-to-end fidelity against patterns reported in empirical studies of real markets, aiming to give agents realistic market feedback and reduce the risk that they learn to exploit flaws in the simulator. Experiments show agents coordinating across roles and adapting decisions to market feedback. With explicit long-term strategic guidance, they also sustain advertising exploration despite weak early returns.
MiniCorp is positioned as both an environment for studying AI-run companies and a scalable source of longitudinal and counterfactual enterprise data for agent training and evaluation.