Inherit-MAS: Test-Time Evolution of Multi-Agent Systems through Workflow and Execution Inheritance

2026-10-01

AI Summary

This study addresses the complexity of workflow design and the computational redundancy of test-time evolution in multi-agent systems by proposing a bio-inspired dual inheritance mechanism. This approach implements explicit inheritance at both the workflow and execution levels, integrating meta-model synthesis, critic-based diagnosis, and context matching with large language models for dynamic iteration. By precisely rectifying defects while reusing effective components, it avoids the disruption associated with global reconstruction. Experimental results on the WorkBench and HotpotQA benchmarks demonstrate that the proposed method significantly outperforms baseline approaches, substantially reduces token consumption, and effectively improves both task completion rates and execution efficiency.

Abstract

Multi-agent systems (MAS) built from large language models coordinate specialized agents to tackle complex tasks, but effective workflows are difficult to design in advance. Test-time evolution refines workflows using execution feedback, yet broad revisions can disturb useful components, while re-executing unchanged requests can incur redundant computation. Inspired by the interplay of inheritance and selection in biological evolution, we introduce Inherit-MAS, which makes inheritance explicit at the workflow and execution levels. A meta-model first synthesizes a workflow of worker agents with declared roles, communication inputs, and tool permissions, and a separately prompted judge scores each executed candidate and diagnoses its deficiencies. In ordinary refinement rounds, \emph{workflow inheritance} starts from the latest completed candidate, may discard removable nodes judged unhelpful, and applies a validated edit to address the diagnosed deficiency. When the new candidate executes, \emph{execution inheritance} inherits eligible stored results only if the complete resolved request and execution context match, avoiding redundant model and tool calls. With GPT-4o-mini workers, Inherit-MAS achieves 55.4\% completion on WorkBench and 49.7\% joint F1 on HotpotQA FullWiki, outperforming EvoAgent, EvoMAS, and TacoMAS. With Qwen3-32B workers, it also exceeds these evolving-MAS baselines on both benchmarks. Compared with rerunning the same controller with execution inheritance disabled, execution inheritance reduces worker-token usage by 29.1\% on WorkBench and 34.6\% on HotpotQA, and total token usage by 5.3\% and 18.1\%.

Problem

• Multi-Agent Systems
• Test-Time Evolution
• Workflow Refinement
• Redundant Computation
• Large Language Models

Innovation

• Multi-Agent Systems
• Test-Time Evolution
• Workflow Inheritance
• Execution Inheritance
• Large Language Models