Paper 2510.04618
Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models
- Published
- Oct 2025
- Research lab
- Independent
- Citations
- 236
- GitHub
- 1.2K stars
01 In brief
Summary
The paper introduces ACE (Agentic Context Engineering), a framework for context adaptation in LLMs that treats contexts as evolving playbooks, accumulating and refining strategies through generation, reflection, and curation.
ACE addresses two limitations of prior methods: brevity bias (over-compression of prompts) and context collapse (loss of detail during iterative rewriting).
It uses incremental delta updates and a grow-and-refine mechanism to preserve detailed knowledge and scale with long-context models.
Evaluated on agent (AppWorld) and domain-specific benchmarks (FiNER, Formula, DDXPlus, BIRD-SQL), ACE consistently outperforms baselines, achieving average gains of +10.6% on agents and +8.6% on finance, while reducing adaptation latency by up to 86.9% and token costs.
ACE works without labeled supervision by leveraging execution feedback, and on AppWorld it matches or surpasses the top-ranked production agent (IBM CUGA) using a smaller open-source model (DeepSeek-V3.1).
The framework is model-agnostic, showing consistent gains across GPT-OSS-120B, GPT-5.1, and Llama-3.3-70B-Instruct.
Ablations confirm the importance of the Reflector, multi-epoch refinement, and incremental updates.
Limitations include reliance on a strong Reflector and potential degradation without reliable feedback signals.
02 From the paper
Abstract
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. We introduce ACE (Agentic Context Engineering), a framework that treats contexts as evolving playbooks that accumulate, refine, and organize strategies through a modular process of generation, reflection, and curation. ACE prevents collapse with structured, incremental updates that preserve detailed knowledge and scale with long-context models. Across agent and domain-specific benchmarks, ACE optimizes contexts both offline (e.g., system prompts) and online (e.g., agent memory), consistently outperforming strong baselines: +10.6% on agents and +8.6% on finance, while significantly reducing adaptation latency and rollout cost. Notably, ACE could adapt effectively without labeled supervision and instead by leveraging natural execution feedback. On the AppWorld leaderboard, ACE matches the top-ranked production-level agent on the overall average and surpasses it on the harder test-challenge split, despite using a smaller open-source model. These results show that comprehensive, evolving contexts enable scalable, efficient, and self-improving LLM systems with low overhead.