Microsoft's MiniCorp: an office simulator generating enterprise data for agents
dair_ai · x · 2026-10-12
Microsoft, with UIUC, Princeton and others, released MiniCorp, a paper tackling the scarcity of longitudinal enterprise data for training and evaluating agents on long-running business work.
- Core idea: MiniCorp is an office simulator where a firm of agents runs an e-commerce company, discussing decisions in messages and meetings and setting prices, inventory, and ad spend. An external simulated market with customers and adaptive competitors responds to those decisions.
- Fidelity checks: The market matches published studies of real ones — a four-week 20% discount lifts units by 71%, and sales stay 31% above control after prices return to normal.
- Counterfactual data: Checkpoints let the same market state replay under different decisions, yielding paired outcomes real company records cannot provide, useful for agent training and eval.
- Findings: Agents coordinate across roles and adapt to market feedback; with explicit long-term strategic guidance they sustain advertising exploration despite weak early returns.
More from coding & agent
- TestSuiteBench: coding agents' tests catch fewer than 55% of logical bugs — OfirPress · 2026-10-12
- Agent automation winners will be those who can visualize what their agents are doing — noahsolomon · 2026-10-12
- Mole 1.16 lands with AI cleanup for Claude Code, Codex and Cursor caches — sujingshen · 2026-10-12
- Codex builds a photo culling tool that ranks shots by face, context and scene — banteg · 2026-10-12
- An AMA only AI agents can join: 24-hour agentic Q&A experiment — metehan777 · 2026-10-12
- Sentinel agent (GPT-6) kills runaway GLM caught phoning a random host — wunderwuzzi23 · 2026-10-12