Can LLM Agents Sustain Long-Horizon Organizational Dynamics?

2026-05-31

AI Summary

This work addresses the challenge that large language model (LLM) agents struggle to maintain goal consistency and execution coherence in long-horizon, structured organizational tasks—particularly under conditions involving task dependencies and artifact accumulation. To tackle this, the authors propose TaskWeave, a novel framework that introduces, for the first time, a structured simulation memory mechanism into LLM-based organizational collaboration. TaskWeave employs a Formulate-Partition-Diagnose-Align cycle to sustain planning state integrity and integrates dependency-aware trajectory memory to ground execution effectively. Built upon a hierarchical agent architecture, TaskWeave demonstrates substantial improvements over existing approaches in a year-long simulated IT company environment, achieving notable gains in organizational coherence, execution grounding, and utility on downstream enterprise NLP tasks.

Abstract

Large language agents are increasingly used for social simulation, yet it remains unclear whether they can sustain coherent behavior in structured organizations, where goals must propagate through hierarchy, tasks depend on prior execution, and artifacts accumulate over long horizons. We formulate long-horizon organizational simulation as a memory-centered coordination problem and introduce TaskWeave, a hierarchical agentic framework that maintains planning states through a Formulate-Partition-Diagnose-Align cycle and grounds execution through dependency-aware trace memory. We evaluate TaskWeave in a year-long IT company simulation and compare it with other multi-agent frameworks on organizational coherence, execution grounding, and downstream enterprise NLP utility. Experiments show that TaskWeave supports coherent and long-horizon organizational dynamics while producing grounded artifacts and adapting to external environments. These findings suggest that structured simulation memory is a key mechanism for building reliable LLM-based organizational simulators.

Problem

• long-horizon simulation
• organizational dynamics
• LLM agents
• coherent behavior
• structured organizations

Innovation

• TaskWeave
• long-horizon simulation
• dependency-aware trace memory
• hierarchical agentic framework
• organizational dynamics