Context Engineering for AI Agents, explained on one whiteboard

Every turn of the agent loop adds to the context. So what should go in, what should stay out, and how do agents keep it lean? We draw context engineering on one board. New to a term? Look for the ▶ badges on the board: each one points to a short "AI agent basics" video (listed below). • What's in the context window: instructions, knowledge, tool feedback • Prompt engineering vs. context engineering • Context rot: why a bigger context isn't automatically better • Lost in the middle: where information sits matters • Four ways long context fails, and the attention budget • Four moves: write, select, compress, isolate (notes, just-in-time retrieval, compaction, sub-agents) • Where curation fits in the agent loop Chapters: 0:00 Intro 0:27 What's in the context 1:09 Prompt vs. context engineering 1:43 Context rot 2:25 Lost in the middle 2:54 Less, but better 3:36 Write, select, compress, isolate 4:21 Back to the loop 4:49 Summary Related videos: Ep. 1 · Tool Calls vs MCP vs Agent Skills:    • Tool Calls vs MCP vs Agent Skills, explain...   Ep. 2 · The AI Agent Loop (and Loop Engineering):    • The AI Agent Loop (and Loop Engineering), ...   Basics · LLMs & tokens:    • Large Language Models (LLMs) and Tokens, e...   Basics · context window & turns:    • Context Window & Turns in AI Agents, expla...   Basics · hallucinations:    • AI Hallucinations, explained on one whiteb...   Basics · Claude Code & sub-agents:    • Claude Code & Sub-agents, explained on one...   Basics · LangChain & Chroma:    • LangChain & Chroma for AI Agents, explaine...   Sources: Anthropic, Effective context engineering for AI agents: https://www.anthropic.com/research/ef... LangChain, Context Engineering: https://www.langchain.com/blog/contex... Chroma, Context Rot: How Increasing Input Tokens Impacts LLM Performance: https://www.trychroma.com/research/co... Liu et al. (2023), Lost in the Middle: How Language Models Use Long Contexts: https://arxiv.org/abs/2307.03172 Narration: synthetic voice (Kokoro TTS). Drawing: rough.js.