Recently, a research team from Tsinghua University has addressed the issue of enhancing the capabilities of LLM (Large Language Model) agents by proposing the AgentTuning method. This approach involves constructing the AgentInstruct agent dataset and employing a mixed instruction fine-tuning strategy. The authors utilized AgentTuning to fine-tune the Llama 2 series, resulting in AgentLM. The results indicate that AgentLM significantly outperforms Llama 2 on various agent tasks, with the 70B version even surpassing GPT-4 on many tasks, offering a powerful open-source alternative. This research provides new insights for the development of LLMs in the field of agent tasks and lays the foundation for future implementations of more intelligent agent systems.