Benchmarking semantic code retrieval on Claude Code — Kuba Rogut, Turbopuffer
Claude Code defaults to agentic grep search, but semantic code retrieval yields measurable accuracy and retention gains.
“they just kind of found agentic search which is actually you know, grepping through your file system kind of worked better and you know, seems to work simpler for cloud code”
Turbopuffer's Kuba Rogut benchmarks semantic code retrieval, noting Claude Code abandoned semantic search for agentic grep while Cursor indexes codebases into Turbopuffer for ~24% answer-accuracy gains and improved code retention. The talk frames a real architectural tradeoff between agentic file-system search and embedding-based retrieval for coding agents, but reads as a vendor-flavored engineering deep-dive rather than a major industry signal.