EvoMAS: Heuristics in the Loop—Evolving Smarter Agentic Workflows

Yangbo Wei, Zhen Huang, Ronghao Xu, Hong Wang, Wei W. Xing
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:133736-133763, 2026.

Abstract

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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-wei26m, title = {{E}vo{MAS}: Heuristics in the Loop—Evolving Smarter Agentic Workflows}, author = {Wei, Yangbo and Huang, Zhen and Xu, Ronghao and Wang, Hong and Xing, Wei W.}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {133736--133763}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/wei26m/wei26m.pdf}, url = {https://proceedings.mlr.press/v306/wei26m.html}, abstract = {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.} }
Endnote
%0 Conference Paper %T EvoMAS: Heuristics in the Loop—Evolving Smarter Agentic Workflows %A Yangbo Wei %A Zhen Huang %A Ronghao Xu %A Hong Wang %A Wei W. Xing %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-wei26m %I PMLR %P 133736--133763 %U https://proceedings.mlr.press/v306/wei26m.html %V 306 %X 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.
APA
Wei, Y., Huang, Z., Xu, R., Wang, H. & Xing, W.W.. (2026). EvoMAS: Heuristics in the Loop—Evolving Smarter Agentic Workflows. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:133736-133763 Available from https://proceedings.mlr.press/v306/wei26m.html.

Related Material