THE AGENTIC ENGINEER

The thirteen ideas beneath every agent framework you will ever use

Every week, a new agent framework arrives, and every week the ground seems to shift faster than you can learn it. THE AGENTIC ENGINEER is built on a quieter truth: you don't fall behind by missing the newest tool — you fall behind by never learning the idea the tool is built on. The frameworks churn. The concepts underneath them barely move at all.

This is a field guide to those concepts. Not a tutorial for any one library, and deliberately naming few products, it teaches the thirteen load-bearing ideas — organized into six layers — that appear, in some form, in every agentic system ever built. Learn them once and every new framework stops being a thing to learn from scratch and becomes a new dialect of a language you already speak.

WHAT'S INSIDE

•  The Building Blocks — what an agent actually is (a loop, not an answer), what it knows (state and the context window), and how to make many agents cooperate.
•   The Configuration Layer — config and workflow files, prompt caching, and the quiet danger of context rot.
•   The Capability Layer — MCP, live retrieval, and persistent memory: how an agent reaches past its own training to what's true now.
•   The Orchestration Layer — subagents and agent loops that scale work across space and time without drowning the main context.
•   The Guardrails Layer — sandboxing, permissions, hooks, injection defense, and gates that keep a capable agent from causing harm and, done well, teach it.a
•   The Observability Layer — tracing what the agent actually did, and the crucial difference between behavior metrics and outcome metrics.

Each of the thirteen chapters opens with a concrete scene, teaches one idea through a recurring cast of working engineers, and closes with takeaways, a hands-on exercise, and a short reflection. Thirteen custom figures, an at-a-glance field map, a "Where to Start" guide, and a full glossary make it as useful as a reference as it is as a read.