Organizational agents coordinate tasks with user constraints and memory
Org-Agent: Beyond Personal Assistants Towards Organizational Agents
Artificial Intelligence
Summary
When language AI agents help groups, they need to understand different users’ roles and share information securely. The authors create Org-Agent, a system that breaks complex jobs into smaller steps, orders them by what depends on what, and manages rules about who can do what and when. This helps the agent remember shared information and handle conflicting requests while working towards group decisions. Their tests show that Org-Agent performs well in managing multiple users and their knowledge during tasks.
What this means in practice
- •For enterprise software teams: Create AI assistants that coordinate tasks among multiple employees with different authority and access rights in complex workflows.$Commercial implications: Enables enterprise software vendors to build smarter AI agents managing user roles and shared knowledge for organizational efficiency.
- •For collaborative workspace developers: Build tools that maintain shared memory and resolve conflicting user inputs when automating group decisions and task execution.
Authors
Luyao Zhuang, Yujing Zhang, Zijin Hong, Yilin Xiao, Xiao Huang
Abstract
Language model agents serving organizations must coordinate requests from multiple users while using knowledge distributed across their interactions. We identify two complementary capabilities for this setting, namely cross-user interaction and decision-making, as well as cross-user memory and knowledge use. Both capabilities are governed by organizational constraints across three aspects: user identity, authority, and access permissions; the attribution and temporal validity of information; and rules for resolving conflicting requirements across users and completion requirements for joint decisions. These constraints shape what information or decisions must be obtained before an action can proceed and what conditions must be satisfied during its execution. Motivated by this, we introduce Org-Agent, a unified constraint-centric reasoning framework that organizes task execution in three stages. Specifically, Org-Agent decomposes a task into atomic subtasks and constructs a task dependency graph whose edges encode the dependencies among them. Building on this graph, it schedules the subtasks in dependency order through topological sorting. It then executes each subtask while accounting for the task's constraints, supported by evidence-acquisition and memory-management tools. Experiments on MUSES-Bench and GroupMemBench demonstrate the effectiveness of Org-Agent on both capabilities, and ablations further support the contributions of dependency modeling and tool use.