PolyU proposes Org-Agent, a constraint-centric framework for organizational AI agents

PolyUHK · hf · 2026-10-01

Researchers at PolyU HK introduce Org-Agent, moving beyond personal assistants toward agents that serve organizations. The paper identifies two complementary capabilities—cross-user interaction/decision-making and cross-user memory/knowledge use—both governed by organizational constraints around identity, authority, information attribution, and conflict resolution.

Org-Agent is a constraint-centric framework that decomposes tasks into atomic subtasks, builds a task dependency graph, schedules subtasks via topological sorting, and executes them with evidence-acquisition and memory-management tools. Experiments on MUSES-Bench and GroupMemBench show effectiveness, with ablations confirming the value of dependency modeling and tool use.

Original post →

More from coding & agent

coding & agent channel →