A Survey of the Training Process of LLM-Based Agents

ACL ARR 2026 January Submission5191 Authors

Keywords: Survey, LLM-Based Agents, Training Process

Abstract: Autonomous agents based on large language models (LLMs) are becoming an increasingly prevalent paradigm for tackling complex and real-world tasks. Despite the remarkable zero-shot capabilities of modern LLMs, specialized agent training is often essential to obtain reliable and improved performance for specific target tasks. In this work, we present a structured survey dedicated exclusively to the training process of LLM-based agents. We establish a clear taxonomy by examining key methodological steps: environment setups, data preparation strategies, the formulation of effective learning signals, as well as the training objectives and schemes. Finally, we conclude with discussions on potential future directions.

Paper Type: Long

Research Area: AI/LLM Agents

Research Area Keywords: Survey, Agents, Training

Contribution Types: Surveys

Languages Studied: English

Submission Number: 5191