← back

Agentic Search for Context Engineering — Leonie Monigatti, Elastic

16.0K views · May 08, 2026 · 63:12 min · Watch on YouTube ↗
Takeaway

Context engineering is mostly about giving an agent the right set of search tools across heterogeneous sources rather than perfecting a single retrieval pipeline.

Summary

  • Leonie Monigatti (Elastic) argues context engineering is ~80% agentic search: the search tools an agent picks between determine what context reaches the LLM window.
  • Traces evolution from fixed RAG pipelines (always retrieving once via vector search) to agentic RAG where the agent decides whether, when, and how to search.
  • Modern agents pull context from many sources — local files, scratchpads, plan.md, skills folders, multiple databases — requiring a toolbelt of complementary search tools, not just one vector store.
  • Workshop with Elasticsearch demos shows mixing keyword, hybrid, and vector search as agent-callable tools to handle heterogeneous context sources.
agentic-ragsearchelastic
Original description
Getting context into an LLM is not just a retrieval problem. It is a search problem. This workshop digs into the part of context engineering that usually gets waved away: how agents actually decide what to pull from files, databases, memory, and the web, and why that choice often matters more than the model itself.

Across semantic search, general-purpose database tools, shell-based retrieval, and agent skills, Leonie Monigatti shows where each search interface works, where it breaks, and how to combine them into a more effective retrieval stack. If you're building agents and trying to make retrieval less brittle, this is a practical guide to the real mechanics behind agentic search.

Workshop repo: https://github.com/iamleonie/workshop-agentic-search

Speaker info:
- https://x.com/helloiamleonie
- https://www.linkedin.com/in/804250ab/

Timestamps:
0:00:00 - Introduction and Welcome
0:00:51 - Defining Context Engineering and the role of Search
0:02:21 - Historical context: From RAG to Agentic RAG
0:04:30 - Context sources (local files, memory, databases, web)
0:06:30 - Introduction to the Shell tool and its versatility
0:08:50 - Failure modes in agentic search
0:10:41 - The importance of tool descriptions and parameter design
0:13:53 - Code Demo: Simple semantic search and its limitations
0:23:26 - Code Demo: General purpose database query (ESQL)
0:28:36 - Code Demo: Adding Agent Skills for better interaction
0:34:42 - Code Demo: Using the Shell tool for file system retrieval
0:41:26 - Code Demo: Integrating custom CLIs (Gina Grap)
0:44:42 - Practical recommendations for building a search tool stack
0:49:16 - Q&A Session begins