Saved in:
Bibliographic Details
Main Authors: Li, Zhening, Solar-Lezama, Armando, Yue, Yisong, Zheng, Stephan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.03571
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914178644574208
author Li, Zhening
Solar-Lezama, Armando
Yue, Yisong
Zheng, Stephan
author_facet Li, Zhening
Solar-Lezama, Armando
Yue, Yisong
Zheng, Stephan
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2512_03571
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle EnCompass: Enhancing Agent Programming with Search Over Program Execution Paths
Li, Zhening
Solar-Lezama, Armando
Yue, Yisong
Zheng, Stephan
Artificial Intelligence
Machine Learning
Programming Languages
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.
title EnCompass: Enhancing Agent Programming with Search Over Program Execution Paths
topic Artificial Intelligence
Machine Learning
Programming Languages
url https://arxiv.org/abs/2512.03571