[Back to Home]

Free book · 19 chapters · v3

Your agent isn’t
getting dumber.
Its context is getting crowded.

Context Engineering 101 is the practical guide to building AI agents that stay sharp at step twenty — by designing what the model sees, not just what you say to it. Get the full book free when you join the newsletter.

// no spam — one useful email at a time. unsubscribe anytime.

The 6-minute version

Watch why prompt engineering stopped being enough

A quick tour of the core idea: the context window is working memory, and almost every agent failure traces back to what was inside it.

context_engineering_overview.mp4

The grammar of the book

Four moves explain almost every reliable agent

Every technique in agent design — scratchpads, retrieval, summarization, sub-agents — is one of these four moves. The book teaches you when to reach for each one.

W

Write

Put important information outside the fragile context window, so the agent can survive forgetting.

→ rules files · memory · playbooks
S

Select

Load only what the next decision requires. Agents get smarter when they see less — the right less.

→ retrieval · tool masking · MCP
C

Compress

Keep the meaning, cut the tokens. More useful signal per token on every single model call.

→ summaries · compaction · state
I

Isolate

Separate jobs deserve separate context windows. Keep noisy research out of clean planning.

→ sub-agents · phases · handoffs

Inside the book

Built for builders

A weekend is enough for a first version

0
chapters, from the context window to self-improving playbooks
0
original figures that make the architecture visible
0
days to build your first context-engineered agent, step by step