Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AgenticAKM : Enroute to Agentic Architecture Knowledge Management

Rudra Dhar, Karthik Vaidhyanathan, Vasudeva Varma

TL;DR

The paper addresses the challenging problem of Architecture Knowledge Management (AKM), which is often labor-intensive and impeded by the distributed, multi-abstraction nature of architectural knowledge. It proposes AgenticAKM, a multi-agent framework that decomposes architecture recovery into specialized tasks—Extraction, Retrieval, Generation, and Validation—coordinated by a central Orchestrator and overseen by a human architect. An instantiation focusing on generating Architecture Decision Records (ADRs) from code repositories is evaluated through a user study, showing that agentic ADR generation significantly outperforms a baseline single-prompt LLM approach in terms of relevance, coherence, completeness, conciseness, and overall quality. The results suggest that AgenticAKM is a practical and scalable approach to automating AKM, with potential to improve onboarding, design rationales, and maintenance across evolving software systems.

Abstract

Architecture Knowledge Management (AKM) is crucial for maintaining current and comprehensive software Architecture Knowledge (AK) in a software project. However AKM is often a laborious process and is not adopted by developers and architects. While LLMs present an opportunity for automation, a naive, single-prompt approach is often ineffective, constrained by context limits and an inability to grasp the distributed nature of architectural knowledge. To address these limitations, we propose an Agentic approach for AKM, AgenticAKM, where the complex problem of architecture recovery and documentation is decomposed into manageable sub-tasks. Specialized agents for architecture Extraction, Retrieval, Generation, and Validation collaborate in a structured workflow to generate AK. To validate we made an initial instantiation of our approach to generate Architecture Decision Records (ADRs) from code repositories. We validated our approach through a user study with 29 repositories. The results demonstrate that our agentic approach generates better ADRs, and is a promising and practical approach for automating AKM.

AgenticAKM : Enroute to Agentic Architecture Knowledge Management

TL;DR

The paper addresses the challenging problem of Architecture Knowledge Management (AKM), which is often labor-intensive and impeded by the distributed, multi-abstraction nature of architectural knowledge. It proposes AgenticAKM, a multi-agent framework that decomposes architecture recovery into specialized tasks—Extraction, Retrieval, Generation, and Validation—coordinated by a central Orchestrator and overseen by a human architect. An instantiation focusing on generating Architecture Decision Records (ADRs) from code repositories is evaluated through a user study, showing that agentic ADR generation significantly outperforms a baseline single-prompt LLM approach in terms of relevance, coherence, completeness, conciseness, and overall quality. The results suggest that AgenticAKM is a practical and scalable approach to automating AKM, with potential to improve onboarding, design rationales, and maintenance across evolving software systems.

Abstract

Architecture Knowledge Management (AKM) is crucial for maintaining current and comprehensive software Architecture Knowledge (AK) in a software project. However AKM is often a laborious process and is not adopted by developers and architects. While LLMs present an opportunity for automation, a naive, single-prompt approach is often ineffective, constrained by context limits and an inability to grasp the distributed nature of architectural knowledge. To address these limitations, we propose an Agentic approach for AKM, AgenticAKM, where the complex problem of architecture recovery and documentation is decomposed into manageable sub-tasks. Specialized agents for architecture Extraction, Retrieval, Generation, and Validation collaborate in a structured workflow to generate AK. To validate we made an initial instantiation of our approach to generate Architecture Decision Records (ADRs) from code repositories. We validated our approach through a user study with 29 repositories. The results demonstrate that our agentic approach generates better ADRs, and is a promising and practical approach for automating AKM.
Paper Structure (13 sections, 1 figure, 1 table)

This paper contains 13 sections, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Motivation and Overview
  3. AKM Agents
  4. Architecture Extractor Agents
  5. Architecture Retriever Agents
  6. Architecture Generator Agents
  7. Architecture Validator Agents
  8. Experiments
  9. Agentic ADR generation from repository
  10. Evaluation Setup
  11. Results
  12. Related Works
  13. Conclusion and Future Works

Figures (1)

  • Figure 1: Agentic AKM