Skills vs. Memory vs. Retrieval—The Context Management Stack Every AI Agent Needs

If there's one thing that can make or break your agent's performance, it's context management. In this video, we break down the three ways LangChain's Deep Agents handles it: Skills, Memory, and Retrieval — and why feeding the right information at the right moment is what separates accurate agents from ones that fall apart. We build a real skill from scratch (a roll-dice skill), walk through SKILL.md structure and progressive disclosure, then dig into long-term memory (AGENTS.md, agent-scoped vs. user-scoped, read-only vs. writable, background consolidation) and wrap up with the three flavors of RAG: 2-Step, Agentic, and Hybrid. ⏱️ Timestamps 00:00 – Why context management matters 00:43 – Part 1: Skills (progressive disclosure, SKILL.md, building a skill from scratch) 04:50 – Part 2: Memory (AGENTS.md, agent-scoped vs user-scoped, read-only vs writable, background consolidation) 09:30 – Part 3: Retrieval (RAG pipeline, 2-Step vs Agentic vs Hybrid RAG) 12:00 – Wrap-up: how skills, memory, and retrieval stack together 🔗 Docs referenced: Skills: https://docs.langchain.com/oss/python... Memory: https://docs.langchain.com/oss/python... Retrieval: https://docs.langchain.com/oss/python... If this helped, drop a like and subscribe for more Deep Agents and LangChain content. #,aiengineering #AgenticAI, #AIAgents, #BuildSaaS, #SaaSDevelopment, #SaaSArchitecture, #AISaaS, #ProductionAISystems, #LLMApplications, #AIWorkflows, #AIAutomation, #LangGraph, #MultiAgentSystems, #AIAgentFramework, #AISystemDesign, #RAGArchitecture, #ProductionRAG, #AIInfrastructure, #VectorDatabase, #AIDataPipelines, #FastAPI, #BackendArchitecture, #MicroservicesArchitecture, #APIDevelopment, #CloudArchitecture, #AWSBackend, #SoftwareArchitecture, #DeveloperTutorials, #BuildAIProjects, #AIDeveloper