arXiv:2611.16526v1 Announce Type: new
Abstract: LLMs have quickly evolved from models used in isolation to agentic systems that delegate tasks to external tools and other LLMs to solve complex assignments. This agentic augmentation has led to impressive capability gains. However, when and how delegation improves performance remains poorly understood. For example, it is unclear whether delegation helps when tasks involve post-cutoff information that the orchestrating LLM cannot reliably resolve from its own knowledge. In this work, we use purpose-built datasets with pre- and post-cutoff splits to compare various delegation modes. We find that delegation provides no consistent accuracy improvement over non-delegated setups, with effects varying across models and tasks. When tasks involve knowledge unavailable to the orchestrator, delegation can substantially degrade accuracy, while direct tool use with the same LLM performs much better. Our analysis identifies distorted delegated queries, inefficient tool use, and failures to preserve and use information between agents as recurring failure modes. These findings suggest that effective orchestration requires both well-calibrated delegation decisions and effective exchange of information between agents.
arxiv
Lost in Delegation: Studying the Effectiveness of Task Orchestration in Generative Search
This arXiv paper investigates when and how delegation improves LLM performance in agentic systems. Using purpose-built datasets with pre- and post-cutoff splits, the authors compare delegation modes and find no consistent accuracy improvement over non-delegated setups. Delegation can substantially degrade accuracy on tasks involving knowledge unavailable to the orchestrator, while direct tool use performs much better. The study identifies distorted delegated queries, inefficient tool use, and failures to preserve information between agents as recurring failure modes.
Provides concrete, recent evidence on a core open problem in agentic AI: when delegation helps or hurts, with actionable failure modes for system design.
Comments & Questions (0)
No comments yet
Be the first to comment!