Language model orchestration improves task handling with flexible worker pooling
Decoupled Multi-Agent Orchestration
Artificial Intelligence
Summary
Coordinating multiple AI agents to work together can be tricky when each agent is fixed and tied to specific tasks. The authors introduce DeOrch, a new way to plan and assign tasks separately from choosing which AI workers do them. This approach allows the system to better learn who is good at what and lets new workers join in without retraining everything. Tests show DeOrch is more efficient and adaptable than earlier methods, working well even when faced with new types of workers or tasks.
What this means in practice
- •For ai system designers: Build multi-agent AI systems that can integrate new agents on the fly without retraining the whole system.
- •For automation engineers: Improve task planning and agent collaboration efficiency in automated workflows with varying worker pools.
Authors
Xinle Wu, Yao Lu
Abstract
Learned orchestration can automatically construct effective language-model multi-agent systems, but existing approaches couple planning to fixed worker pools and train decomposition and collaboration from the same terminal outcome, limiting transfer and obscuring credit assignment. We introduce DeOrch, which separates worker-agnostic planning from concrete worker selection. Its two-stage planner first decomposes the task without worker information, then chooses collaboration operations using compact, worker-identity-free matchability feedback from the pool, enabling conditional credit assignment to decomposition and collaboration decisions. A lightweight matcher estimates worker suitability from behavior on a fixed probe set and adapts online with a contextual bandit, allowing new workers to be incorporated without retraining the planner or matcher. Across diverse in- and out-of-distribution tasks, DeOrch outperforms prior automatic MAS orchestration methods with fewer worker calls than competing learned orchestrators, remains effective when transferred to an entirely unseen worker pool without retraining, and shows consistent gains from both components.