AgentConductor: Topology Evolution for Multi-Agent Competition-Level Code Generation

2026-02-19

AI Summary

This work addresses the limitations of existing large language model–based multi-agent code generation systems, which rely on static communication topologies that struggle to adapt to varying task difficulty, resulting in redundant communication and performance bottlenecks. To overcome this, we propose AgentConductor, a framework featuring a reinforcement learning–driven conductor agent that dynamically constructs a task-adaptive hierarchical directed acyclic graph (DAG) communication topology, enabling end-to-end feedback–guided topological evolution. Our approach innovatively introduces a communication-aware topological density function coupled with a task-difficulty interval partitioning mechanism to precisely regulate the upper bound of topological density across different difficulty levels. Evaluated on five code generation benchmarks—including three competition-level datasets—AgentConductor achieves state-of-the-art performance, improving pass@1 accuracy by up to 14.6% over the strongest baseline while reducing topological density by 13% and token consumption by 68%.

Technology Category

• Multiagent Systems: Agent Communication
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large language model(LLM)-driven multi-agent systems(MAS) coordinate specialized agents through predefined interaction topologies and have shown promise for complex tasks such as competition-level code generation. Recent studies demonstrate that carefully designed multi-agent workflows and communication graphs can significantly improve code generation performance by leveraging collaborative reasoning. However, existing methods neither adapt topology density to task difficulty nor iteratively refine the topology within an instance using execution feedback, which leads to redundant communication and performance bottlenecks. To address these issues, we propose AgentConductor: a reinforcement learning-optimized MAS with an LLM-based orchestrator agent as its core, which enables end-to-end feedback-driven dynamic generation of interaction topologies. For each query, AgentConductor infers agent roles and task difficulty, then constructs a task-adapted, density-aware layered directed acyclic graph (DAG) topology, underpinned by two key innovations. First, we design a novel topological density function that captures communication-aware mathematical characterizations of multi-agent interactions. Second, we adopt difficulty interval partitioning to avoid excessive pruning for precise topological density upper bound measurement per difficulty level and finer-grained control. Empirically, across three competition-level and two foundational code datasets, AgentConductor achieves state-of-the-art accuracy, outperforming the strongest baseline by up to 14.6% in pass@1 accuracy, 13% in density reduction, and 68% in token cost reduction.

Problem

• multi-agent systems
• interaction topology
• code generation
• topology adaptation
• task difficulty

Innovation

• dynamic topology
• multi-agent system
• topology density
• difficulty-aware adaptation
• LLM-based orchestration