Collaborative Software Architecting With Large Language Models
by Claudine Allen
In this talk, we will primarily examine my findings on how software architects can utilise AI as an architecture design collaborator throughout the architecture analysis and design process. We will examine the literature, real case studies, and the results of my own experiments with LLMs at different stages of architectural design.
The talk is concluded with what I propose as a practical framework for doing software architecture work through collaboration with LLMs. In doing this, I also highlight the ways in which human judgement is essential and will likely remain essential while staying cognisant of the ever-growing capabilities of these LLMs.
The framework addresses:
1. What to tell LLMs – Giving the LLM context
2. What questions to ask LLMs – Which questions can LLMs reasonably provide accurate or useful answers to?
3. How to ask the LLMs – Writing effective prompts (and prompt sequences)
4. Assessing the risk when co-architecting with LLMs – What is at stake?
5. Taxonomy of tasks and subtasks based on documented software architecture design and analysis methodologies and how LLMs can be used for them
6. A software architectural task and AI collaboration value matrix