arXiv 2510.04618
Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models
By Qizheng Zhang, Changran Hu, et al.
Published 2025-10-06
Discussion
Read the public discussion and references gathered around this paper.
Large language model (LLM) applications such as agents and domain-specific reasoning increasingly rely on context adaptation -- modifying inputs with instructions, strategies, or evidence, rather than weight updates. Prior approaches improve usability but often suffer from brevity bias, which drops domain insights for concise summaries, and from context collapse, where iterative rewriting erodes details over time. B…