
AI, LLMs, and Evolutionary Architecture

Building Evolutionary Architectures: Automated Software Governance - Rebecca Parsons
Topic 9 of 10
Published Jan 13, 2025 · Clip 45:41–52:32 (6:51)
The host asks about the impact of AI and LLMs on evolutionary architecture. Parsons explains fitness functions can assess AI-generated code, noting anecdotal evidence that code generators tend to copy-paste-modify rather than abstract, so copy-paste detectors help. She sees genuine promise in LLMs, citing ThoughtWorks experiments using LLMs to understand information flow in legacy codebases to enable refactoring and replacement. However, she worries about the industry's apprenticeship model: if coding assistants don't improve, future star developers may not emerge, citing a CodeScene study where coding assistants' refactoring recommendations were correct at best 37% of the time, undermining the training of the next generation.