Why AI works better on existing codebases (www.stromcapital.fi)

A recent exploration of AI-assisted coding reveals that established codebases, or brownfield projects, offer significant advantages over greenfield projects where AI generates code from scratch. Unlike an empty repository, which provides no reference for consistent coding patterns, a populated codebase allows AI tools to index existing code into semantic embeddings, accessing a wealth of working examples and established conventions. This leads to improved consistency and the ability for AI to build upon prior implementations rather than inventing disparate solutions that create a chaotic codebase. However, the quality of the existing code is crucial; poorly structured legacy code can lead to replication of bad practices. Developers are encouraged to establish clear architectural rules within their brownfield projects, outlining the patterns to prioritize, which allows AI to extend systems coherently. For those embarking on greenfield projects, starting with a few completed features and a structured setup before employing AI can yield much better results. Ultimately, leveraging the context of an existing codebase empowers AI, making the case for a "go brown, not green" strategy in AI-assisted development.