Large Model Agent: Concept, Development and Practice
DOI: https://doi.org/10.62517/jike.202604401
Author(s)
Shoutong Huang*, Yu Ma*
Affiliation(s)
School of Electronic and Electrical Engineering, Ningxia University, Yinchuan, Ningxia, China
*Corresponding Author
Abstract
A large language model is placed at the center of the reasoning of an LLM-based agent. It coils five things around that center, namely perception, memory, planning, tool calling and action execution. In this paper we review the history of how these agents came to be - in four phases that go through symbolic systems, reactive designs, reinforcement learning, and finally the current state of LLM-driven architectures - and we discuss four trends that are still emerging: multi-agent cooperation, multimodal sensing, long-term memory, and standardization of tools. We also report two systems we built. The former is a ReAct-like question answering agent which queries Wikipedia; the latter, MedGuard, was developed to assist older adults with their medication. The on-time medication rate increased by 60 to 95 percent in a seven-day experiment involving ten elderly participants and satisfaction was 4.6 out of five points. In addition to these findings, we provide a reproducible two-agent safety architecture and field evaluation protocol- two components which, combined together, can help bridge the gap between agent theory and practical implementation in elder care.
Keywords
Large Model Agent; ReAct; Multi-Agent; Intelligent Healthcare; Tool Calling; MCP
References
[1] Xi, Z., Chen, W., Guo, X., He, W., Ding, Y., Hong, B., et al. The Rise and Potential of Large Language Model Based Agents: A Survey. Science China Information Sciences, 2025, 68(2): 121101.
[2] Brooks, R. A. A Robust Layered Control System for a Mobile Robot. IEEE Journal on Robotics and Automation, 1986, 2(1): 14-23.
[3] Anthropic. Introducing the Model Context Protocol. Anthropic Technical Report, 2024. https://www.anthropic.com/news/ model-context-protocol (Accessed: 2025-09-08).
[4] Hong, S., Zhuge, M., Chen, J., Zheng, X., Cheng, Y., Wang, J., et al. MetaGPT: Meta Programming for a Multi-Agent Collaborative Framework. In International Conference on Learning Representations (ICLR 2024). arXiv:2308.00352.
[5] Liu, H., Li, C., Wu, Q., Lee, Y. J. Visual Instruction Tuning. In Advances in Neural Information Processing Systems 36 (NeurIPS 2023). arXiv:2304.08485.
[6] Park, J. S., O’Brien, J., Cai, C. J., Morris, M. R., Liang, P., Bernstein, M. S. Generative Agents: Interactive Simulacra of Human Behavior. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (UIST 2023), 2023: 559:1-559:22.
[7] Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., et al. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. In Advances in Neural Information Processing Systems 35 (NeurIPS 2022): 24824-24837.
[8] Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., et al. ReAct: Synergizing Reasoning and Acting in Language Models. In International Conference on Learning Representations (ICLR 2023). arXiv:2210.03629.
[9] Zhang, S., Yin, M., Zhang, J., Liu, J., Han, Z., Zhang, J., et al. Which Agent Causes Task Failures and When? On Automated Failure Attribution of LLM Multi-Agent Systems. Proceedings of Machine Learning Research, 2025, 267: 76583-76599.
[10] Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., et al. Language Models are Few-Shot Learners. In Advances in Neural Information Processing Systems 33 (NeurIPS 2020): 1877-1901.
