Research map / Large language models / Agents, code and math
Agentic Context Engineering: Learning Comprehensive Contexts for Self-Improving Language Models
Qizheng Zhang, Changran Hu, Shubhangi Upasani, Boyuan Ma, Fenglu Hong, Vamsidhar Kamanuru, Jay Rainton, Chen Wu, Mengmeng Ji, Hanchen Li, Urmish Thakker, James Zou and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add incremental delta context updates (ablation) | ACE · AppWorld test-normal | average score: 56.9 -> 70.3 | Improved |
| add Reflector iterative refinement (ablation) | ACE · AppWorld | average score: 55.1 -> 56.8 | Improved |
| add multi-epoch adaptation (ablation) | ACE · AppWorld | average score: 56.8 -> 59.4 | Improved |
| add offline warmup before online adaptation (ablation) | ACE · AppWorld online | average score: 56.1 -> 59.5 | Improved |
| replace GEPA with ACE for offline adaptation | ReAct · AppWorld | average score: 46.4 -> 59.4 | Improved |
| remove ground-truth labels from the Reflector | ACE offline · AppWorld | average score: 59.4 -> 57.2 | No clear effect |
| remove ground-truth labels from the Reflector | ACE online · Financial benchmark | average accuracy: 76.6 -> 72.9 | Hurt |
| remove ground-truth labels from the Reflector | DC (CU) online · Financial benchmark | average accuracy: 71.8 -> 65.4 | Hurt |
| replace Dynamic Cheatsheet with ACE online | ReAct · AppWorld | average score: 51.9 -> 59.5 | Improved |
| replace GEPA with ACE for adaptation cost | AppWorld offline · adaptation latency | seconds: 53898 -> 9517 | Improved |
| inject a harmful Reflector update every iteration (ablation) | ACE · FiNER | accuracy: 78.3 -> 66.7 | Hurt |
| increase reflection iterations from 5 to 10 (ablation) | ACE · AppWorld test-normal | average score: 67.6 -> 65.2 | Hurt |
Closest papers by meaning
- LOCA-bench: Benchmarking Language Agents Under Controllable and Extreme Context Growth (ICML 2026 · similarity 0.91)
- EvoMemBench: Benchmarking Agent Memory from a Self-Evolving Perspective (NeurIPS 2026 · similarity 0.89)
- Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions (ICLR 2026 · similarity 0.89)
- A-Mem: Agentic Memory for LLM Agents (NeurIPS 2025 · similarity 0.89)
- Self-Generated In-Context Examples Improve LLM Agents for Sequential Decision-Making Tasks (NeurIPS 2025 · similarity 0.89)
- Chain of Agents: Large Language Models Collaborating on Long-Context Tasks (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
