EvoMAS: Heuristics in the Loop—Evolving Smarter Agentic Workflows

The rapid development of Large Language Models has driven Multi-Agent Systems (MAS) growth, but constructing efficient MAS requires labor-intensive manual design. Current automation methods generate templated agents, use monolithic optimization, and ignore task complexity gradients. This paper presents Evolutionary MAS (EvoMAS), a biologically-inspired framework whose core is a dynamic and diverse repertoire of seven evolutionary strategies---six biologically-inspired operators (3 exploration, 3 exploitation) together with a custom operator for domain-specific transformations---driven by adaptive strategy selection. These are complemented by role-level evolution that refines agent specialization and collaboration patterns, and a curriculum-guided schedule that evolves workflows from simple to complex tasks with cross-stage stability. Additionally, to resolve the contradiction between the inefficiency of pure evolution and the rigidity of manual design, we introduce the Cyber Creator, a meta-controller that conducts heuristics-in-the-loop learning by formulating and reflectively updating evolutionary rules and strategies. Evaluations demonstrate that EvoMAS consistently outperforms existing methods across multiple domains while maintaining cost efficiency, with roles evolving from homogeneous actors to specialized reasoning ensembles.

Many AI applications now use multiple AI agents that work together, rather than relying on a single model to solve every problem. However, designing such multi-agent systems is still difficult and often requires experts to manually decide each agent’s role, how agents communicate, and how the whole workflow should be organized. This manual process can be slow, costly, and hard to adapt to new tasks. This paper introduces EvoMAS, a framework that automatically improves multi-agent workflows over time. Inspired by biological evolution, EvoMAS explores different ways to organize agents, selects better designs, and gradually refines them. It can adjust agent roles, choose different improvement strategies, and train the system from easier tasks to harder ones. A key part of EvoMAS is the “Cyber Creator,” which guides the search process using rules and feedback, helping the system avoid inefficient trial-and-error. Experiments on reasoning, coding, question answering, embodied interaction, and tool-use tasks show that EvoMAS produces stronger and more cost-efficient multi-agent systems than existing manual or automated methods. Overall, EvoMAS aims to make multi-agent AI systems easier to build, more adaptable, and more practical for real-world use.