Beyond grep: The case for a context-rich AI coding harness

A harness is software built around AI models that determines what the models see, what actions they can take, and how they interact with a codebase. Different companies building AI coding tools make different choices about whether to build lean harnesses with minimal features or more structured ones that pre-process code repositories. One approach uses grep-based search similar to traditional code searching, while another uses semantic retrieval with embeddings and vector databases to find conceptually relevant code. The choice matters because these different harnesses affect how efficiently AI agents use their limited context windows when working on coding tasks, particularly in large private codebases that the models have not encountered during training.