To address the problems of insufficient adaptability of fixed processes, failure cascading, and uncontrolled resource consumption in long-chain tasks for large language model agents, this paper proposes a state-aware adaptive orchestration mechanism based on dynamic task graphs and budget constraints. This mechanism combines task difficulty, execution history, remaining budget, quality confidence and risk exposure into a single state vector, and then selects a workflow based on the combined utility of quality, cost, latency and risk. Runtime re-orchestration is performed by online reliability updates, partial rollbacks and verification gates. Secondary statistics from publicly available data on WebArena and automated agent design research show that the overall success rate of fixed agents in real-world long-chain environments is significantly lower than that of manual operations. Workflows obtained through automated search also outperform representative manual processes in the four task categories of reading, mathematics, multi-task and science. This study shows that the orchestration object should change from a fixed role sequence to an observable, evaluable and rollback-capable dynamic task graph, and provides a way to build stable and cost-effective agent systems.

Shefeng Li. Research on Adaptive Workflow Orchestration Mechanism for Large Language Model Agents with Dynamic Task Graphs and Budget Constraints. International Journal of Big Data Intelligent Technology (2026), Vol. 7, Issue 3: 1-8. https://doi.org/10.38007/IJBDIT.2026.070301.

[1] Wang L, Ma C, Feng X, et al. A survey on large language model based autonomous agents. Frontiers of Computer Science, 2024, 18: 186345. DOI: 10.1007/s11704-024-40231-1. URL: https://link.springer.com/article/10.1007/s11704-024-40231-1

[2] Li X, Wang S, Zeng S, et al. A survey on LLM-based multi-agent systems: workflow, infrastructure, and challenges. Vicinagearth, 2024, 1: 9. DOI: 10.1007/s44336-024-00009-2. URL: https://link.springer.com/article/10.1007/s44336-024-00009-2

[3] Gao C, Lan X, Li N, et al. Large language models empowered agent-based modeling and simulation: a survey and perspectives. Humanities and Social Sciences Communications, 2024, 11: 1259. DOI: 10.1057/s41599-024-03611-3. URL: https://www.nature.com/articles/s41599-024-03611-3

[4] He J, Treude C, Lo D. LLM-Based Multi-Agent Systems for Software Engineering: Literature Review, Vision, and the Road Ahead. ACM Transactions on Software Engineering and Methodology, 2025, 34(5): Article 124, 1-30. DOI: 10.1145/3712003. URL: https://dl.acm.org/doi/10.1145/3712003

[5] Jimenez-Romero C, Yegenoglu A, Blum C. Multi-agent systems powered by large language models: applications in swarm intelligence. Frontiers in Artificial Intelligence, 2025, 8: 1593017. DOI: 10.3389/frai.2025.1593017. 

[6] Chowa SS, Alvi R, Rahman SS, et al. From language to action: a review of large language models as autonomous agents and tool users. Artificial Intelligence Review, 2026, 59: 71. DOI: 10.1007/s10462-025-11471-9. URL: https://link.springer.com/article/10.1007/s10462-025-11471-9

[7] Abou Ali M, Dornaika F, Charafeddine J. Agentic AI: a comprehensive survey of architectures, applications, and future directions. Artificial Intelligence Review, 2026, 59: 11. DOI: 10.1007/s10462-025-11422-4. URL: https://link.springer.com/article/10.1007/s10462-025-11422-4

[8] Webb T, Mondal SS, Momennejad I. A brain-inspired agentic architecture to improve planning with LLMs. Nature Communications, 2025, 16: 8633. DOI: 10.1038/s41467-025-63804-5. URL: https://www.nature.com/articles/s41467-025-63804-5

[9] Kim S, Yu Y, Seo H. Artificial intelligence orchestration for text-based ultrasonic simulation via self-review by multi-large language model agents. Scientific Reports, 2025, 15: 12474. DOI: 10.1038/s41598-025-97498-y. URL: https://www.nature.com/articles/s41598-025-97498-y

[10] Pozi MSM, Sato Y. A data-augmented model routing framework for efficient LLM deployment in edge-cloud environments. The Journal of Supercomputing, 2025, 81: 1573. DOI: 10.1007/s11227-025-08034-8. URL: https://link.springer.com/article/10.1007/s11227-025-08034-8

[11] Han, W. (2025, November). An Empirical Study on Building a High-precision Forecasting System for Financial Market Trends Based on the PyTorch Deep Learning Framework. In International Conference on Computational Technologies for Research in Data Analytics (pp. 518-528). Cham: Springer Nature Switzerland.

[12] Wang, Z. (2026). Construction of a Data-driven Decision-making Framework for Bulk Commodity Supply Chains. Engineering Advances, 6(3).

[13] Wang, Z. (2026). Research on Operation Data Analysis and Efficiency Improvement of the New Energy Industry Based on Machine Learning. Procedia Computer Science, 285, 605-612.