AI agents are increasingly leveraging files and software tools to produce professional deliverables. Such work draws on materials, decisions, and feedback accumulated across assignments. We introduce OfficeTown, a generative simulation framework for professional organizations where tasks and their context emerge from ongoing organizational work.
Given a natural‑language description of an organization’s purpose and activities, OfficeTown initializes an organization whose members coordinate work, delegate bounded assignments to assistant agents, and review their outputs. Artifacts, decisions, and unresolved issues persist and shape subsequent work, while responsibilities and permissions determine each member’s visibility and actions.
We curate work requests, their context, and review feedback into executable resources for evaluation and learning. From this we build OfficeTown Bench, containing 406 tasks and 4,836 evaluation criteria across 23 simulated organizations in eight professional domains, incorporating domain‑expert feedback.
Frontier‑model evaluations reveal substantial headroom. Supervised fine‑tuning and on‑policy distillation on OfficeTown data markedly improve Qwen3.5‑9B on external professional‑work benchmarks, including GDPval. Supplementary experiments show that a principal‑assistant harness achieves higher scores than single‑ and multi‑agent baselines.
These findings highlight the value of organizational simulation for generating work‑grounded data and environments, supporting the continued development of generalist agents.
Review