Why even say this? Codebase Indexing…
AI Summary
This video discusses the contentious topic of codebase indexing in AI coding agents. The creator reviews an article by Nick Balman criticizing client-side codebase indexing and shares insights from tests comparing several AI coding agents: Augment Code, Klein, Rue Code, Root Code, Sourcegraph Cody, Cursor, and Warp.dev. The content covers various indexing approaches, including AST (Abstract Syntax Tree) chunking for logical code grouping, versus blind token chunking. The video explains that while Klein does not use indexing but still uses retrieval-augmented generation (RAG), others like Augment Code use sophisticated semantic search and indexing strategies focused on helpfulness rather than surface similarity. The creator highlights testing on a 100MB codebase with 2500 files, comparing accuracy, speed, and cost of answers on complex code queries. Augment Code emerges as the fastest and most cost-effective despite being a black box proprietary tool. The video concludes that indexing a codebase, especially with AST-based chunking, offers significant value for AI agents, especially for obscure queries with little context. It suggests indexing will become increasingly necessary as codebases grow, saving both time and token cost, and calls for reconsideration of the notion that not indexing is a good thing. The creator invites discussion and comments on the topic.