Context Engineering for Agentic Workflows

The success of agentic applications using LLMs depends largely on the ability to properly manage the context - the collection of prompts, tools, history, memory, and RAG-indexed content. When you take all of these elements into account, you are able to get the most out of the LLM while avoiding hallucination, conversational drift, and relevance issues. Topics to be covered include: dynamic LLM selection, automated prompt development, context compression, tool decoration, RAG optimization, and LLM-as-a-judge quality assessment of LLM responses. Each of these smaller parts together build a complete context engineering approach that allows an agentic workflows to perform at high quality while keeping cost to a minimum. This session will inspect worked examples using open source technologies such as Langchain, AGNO, and N8N.