Patterns for a New Generation: AI and Agents

TL;DR

This paper demonstrates that large language models can utilize structured design patterns to guide agent behavior, enabling hybrid human-agent systems with applications across various domains.

Contribution

It introduces a novel approach where LLMs read and reason with design patterns, using the Active Inference Framework to guide agent actions without strict prescriptions.

Findings

Patterns serve as shared priors for decision-making.

Agents can read, generate, and reason with structured patterns.

Proof of concept for pattern-capable agents using standard tools.

Abstract

Design patterns have been used in various fields of inquiry and endeavour to externalize procedural knowledge in a form that supports human reasoning and coordination. In this paper, we show that contemporary Large Language Model (LLM)-based systems can also read, generate, and reason with design patterns written in a structured template. We describe an experimental workflow in which patterns function as shared priors for action selection, reflection, and revision in hybrid human/agent settings. Drawing on the Active Inference Framework, we illustrate how patterns can guide agent behavior without fully prescribing it. This provides a proof of concept that pattern-capable agents can be created using now-standard software tools. We discuss implications for software development, education, business, and AI governance.