[Infographics] Code Search for Agents: Is Grep-Only Retrieval Enough or Do You Need Vector/Semantic Search?
🔍 Big Question:
For AI coding agents, is grep-only retrieval enough… or do you need vectors and semantic search?
In the MLOps Community, Mirco Bianchini sparked a great debate:
Mehdi B ➡️ “Milvus take is biased. Grep works ~99% of the time if used well. RAG adds complexity in fast-moving repos. For structural tasks, AST/Tree-sitter > embeddings.”
Sugam Garg ➡️ “Feels like overengineering. Don’t add complex systems for the last 1%.”
Mirco ➡️ testing semtools (from LlamaIndex) as a sub-agent; constraining Claude to targeted files helps.
💡 Our take: Smarter grep + structure-aware tools (AST/Tree-sitter) often beats full-blown RAG. Semantic search can help on huge repos or cross-cutting contexts, but expect maintenance overhead.
📌 For full coverage, check ou the article below 👇🏾:
🗣️ Voice Agents Evals, Claude Code’s Grep vs RAG, 90%-generated AI Code Hype, Cloud LLM Deploys, and Token “Scams”
Every other Thursday, we surface the most thought-provoking, practical, and occasionally spicy conversations happening inside the AI communities so you never miss out. The MLOps Community Slack is where the sharpest builders and tinkerers ask the hard questions and don't settle for surface-level answers.