Beyond Grep: Search for Reliable Coding Agents
summary
Claim: improving retrieval beats blindly enlarging models. Compares grep, lexical ranking, structural search (function/class extraction + LSP-like links) and semantic embeddings using an MCP server that indexes an OpenSearch codebase and runs local embeddings. Finds grep returns tests first and consumes most tokens, structural search returns the correct class top for API-definition queries, semantic retrieval solves natural‑language questions immediately, and GitHub Copilot reports a ~12% improvement with semantic search — but indexing freshness and maintenance add real cost.
AI-written summary · about
chapters (9)
- 0:00Introduction and speakers
- 1:09Why coding agents work well
- 5:40Limitations of grep for agents
- 7:08Retrieval methods overview
- 12:59Market shift in coding assistants
- 19:19Demo comparing retrieval methods
- 26:50Coding agent infrastructure
- 27:54Key takeaways and conclusions
- 30:06Q&A session
related talks
16mBenchmarking semantic code retrieval on Claude Code
11mRAG is dead, right??
22mAgentic Hybrid Retrieval: Orchestrating LLM Agents, Hybrid Search & Explainability in OpenSearch
45mMaking semantic search & RAG real: How to build a prod-ready app
14mHow we taught agents to use good retrieval
37mAgentic AI Meets Search: Real-World Use Cases With...
28mAI Agents in Action: Harnessing MCP and OpenSearch for Modern Use Cases
45mNever Send a Human to do an Agent's Search