JADE: Bridging the Strategic-Operational Gap in Dynamic Agentic RAG

Yiqun Chen, Erhan Zhang, Tianyi Hu, Shijie Wang, Zixuan Yang, Meizhi Zhong, Xiaochi Wei, Yan Gao, Yi Wu, Yao Hu, Jiaxin Mao
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:18188-18206, 2026.

Abstract

The evolution of Retrieval-Augmented Generation (RAG) has shifted from static retrieval pipelines to dynamic, agentic workflows where a central planner orchestrates multi-turn reasoning. However, existing paradigms face a critical dichotomy: they either jointly optimize modules within rigid, fixed-graph architectures, or enable dynamic planning while treating executors as frozen, black-box tools. We identify that this decoupled optimization creates a “strategic-operational mismatch,” where sophisticated planning strategies fail to materialize due to unadapted local executors, often causing negative gains despite increased system complexity. In this paper, we propose JADE (Joint Agentic Dynamic Execution), a unified framework for joint optimization of planning and execution within dynamic, multi-turn workflows. By modeling the system as a cooperative multi-agent team with a shared backbone, JADE enables end-to-end learning driven by outcome-based rewards. This approach facilitates co-adaptation: the planner learns to operate within executor capability boundaries, while executors evolve to align with strategic intent. Empirical results demonstrate that JADE transforms disjoint modules into a synergistic system, yielding strong performance improvements via joint optimization and enabling a flexible balance between efficiency and effectiveness through dynamic workflow orchestration.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-chen26gh, title = {{JADE}: Bridging the Strategic-Operational Gap in Dynamic Agentic {RAG}}, author = {Chen, Yiqun and Zhang, Erhan and Hu, Tianyi and Wang, Shijie and Yang, Zixuan and Zhong, Meizhi and Wei, Xiaochi and Gao, Yan and Wu, Yi and Hu, Yao and Mao, Jiaxin}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {18188--18206}, 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/chen26gh/chen26gh.pdf}, url = {https://proceedings.mlr.press/v306/chen26gh.html}, abstract = {The evolution of Retrieval-Augmented Generation (RAG) has shifted from static retrieval pipelines to dynamic, agentic workflows where a central planner orchestrates multi-turn reasoning. However, existing paradigms face a critical dichotomy: they either jointly optimize modules within rigid, fixed-graph architectures, or enable dynamic planning while treating executors as frozen, black-box tools. We identify that this decoupled optimization creates a “strategic-operational mismatch,” where sophisticated planning strategies fail to materialize due to unadapted local executors, often causing negative gains despite increased system complexity. In this paper, we propose JADE (Joint Agentic Dynamic Execution), a unified framework for joint optimization of planning and execution within dynamic, multi-turn workflows. By modeling the system as a cooperative multi-agent team with a shared backbone, JADE enables end-to-end learning driven by outcome-based rewards. This approach facilitates co-adaptation: the planner learns to operate within executor capability boundaries, while executors evolve to align with strategic intent. Empirical results demonstrate that JADE transforms disjoint modules into a synergistic system, yielding strong performance improvements via joint optimization and enabling a flexible balance between efficiency and effectiveness through dynamic workflow orchestration.} }
Endnote
%0 Conference Paper %T JADE: Bridging the Strategic-Operational Gap in Dynamic Agentic RAG %A Yiqun Chen %A Erhan Zhang %A Tianyi Hu %A Shijie Wang %A Zixuan Yang %A Meizhi Zhong %A Xiaochi Wei %A Yan Gao %A Yi Wu %A Yao Hu %A Jiaxin Mao %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-chen26gh %I PMLR %P 18188--18206 %U https://proceedings.mlr.press/v306/chen26gh.html %V 306 %X The evolution of Retrieval-Augmented Generation (RAG) has shifted from static retrieval pipelines to dynamic, agentic workflows where a central planner orchestrates multi-turn reasoning. However, existing paradigms face a critical dichotomy: they either jointly optimize modules within rigid, fixed-graph architectures, or enable dynamic planning while treating executors as frozen, black-box tools. We identify that this decoupled optimization creates a “strategic-operational mismatch,” where sophisticated planning strategies fail to materialize due to unadapted local executors, often causing negative gains despite increased system complexity. In this paper, we propose JADE (Joint Agentic Dynamic Execution), a unified framework for joint optimization of planning and execution within dynamic, multi-turn workflows. By modeling the system as a cooperative multi-agent team with a shared backbone, JADE enables end-to-end learning driven by outcome-based rewards. This approach facilitates co-adaptation: the planner learns to operate within executor capability boundaries, while executors evolve to align with strategic intent. Empirical results demonstrate that JADE transforms disjoint modules into a synergistic system, yielding strong performance improvements via joint optimization and enabling a flexible balance between efficiency and effectiveness through dynamic workflow orchestration.
APA
Chen, Y., Zhang, E., Hu, T., Wang, S., Yang, Z., Zhong, M., Wei, X., Gao, Y., Wu, Y., Hu, Y. & Mao, J.. (2026). JADE: Bridging the Strategic-Operational Gap in Dynamic Agentic RAG. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:18188-18206 Available from https://proceedings.mlr.press/v306/chen26gh.html.

Related Material