2026-01-13 カリフォルニア工科大学（Caltech）
The EnCompass framework compiles an AI agent program and its workflow into a branching search space object to identify the best pathway.Credit: Li et al. (2025) 39th Conference on Neural Information Processing Systems
＜関連情報＞
- https://www.caltech.edu/about/news/helping-ai-systems-recover-from-mistakes-and-find-optimal-solutions
- https://neurips.cc/virtual/2025/loc/san-diego/poster/118817
- https://openreview.net/forum?id=IKVkpjSJzJ
EnCompass: プログラム実行パスの検索によるエージェントプログラミングの強化 EnCompass: Enhancing Agent Programming with Search Over Program Execution Paths
Zhening Li, Armando Solar-Lezama, Yisong Yue, Stephan Zheng
Open Review Published: 19 Sept 2025
Abstract
We introduce a new approach to agent programming, the development of LLM-based agents. Current approaches to agent programming often entangle two aspects of agent design: the core workflow logic and the inference-time strategy (e.g., tree search). We introduce probabilistic angelic nondeterminism (PAN), a programming model that disentangles these two concerns, allowing the programmer to describe the agent workflow and independently experiment with different inference-time strategies by simply changing a few inputs. We provide an implementation of PAN in Python as the EnCompass framework, which uses a Python decorator to compile agent workflow programs into a search space. We present three case studies that demonstrate how the framework lets the programmer quickly improve the reliability of an agent and easily switch between different inference-time strategies, all with little additional coding.
