Agentic Context Engineering
A way for an AI agent to get better over time by rewriting its own notes, instead of retraining the model.
Definition
Agentic Context Engineering (ACE) is a technique for improving an AI agent's performance without changing its underlying weights. Instead of fine-tuning, ACE keeps a growing, structured "playbook" of concrete lessons — what worked, what failed, and why — and updates it after each task. The process splits into three roles: a Generator that attempts the task and produces a reasoning trace, a Reflector that pulls out concrete lessons from what succeeded or failed, and a Curator that folds those lessons into the evolving playbook. Because the playbook accumulates detail rather than being rewritten from scratch each time, it can grow into a genuinely useful reference, and researchers found it noticeably improved agents on tasks like multi-step tool use and financial analysis.