Contents

Agent Engineering

Overview

View as:

12 - Agent Engineering

What You Will Learn

Agent Engineering is the practice of designing the system around the model rather than the model itself: what scaffolding it runs inside (harness), how you decide when work should self-iterate versus stay a single step (loop), and how you decide when a single iterating step should split into a coordinated structure of specialized parts (graph). Where Prompt Engineering optimizes a single call and Agentic AI covers the mechanics of a single agent loop, this module covers the architectural decisions that sit above both.

Why It Comes After Agents and Agentic AI

You need to understand what an agent is (Topic 05) and how the base agent loop actually executes - the LLM/Tool Executor/Loop Controller, ReAct, termination conditions (Topic 06) - before the design-level questions here make sense. Harness Engineering asks what surrounds that loop to make it durable. Loop Engineering asks when building a loop at all is worth the setup cost. Graph Engineering asks when a single loop should become a coordinated structure instead.

Chapter Map

#FileTopic
1Harness EngineeringAgent = Model + Harness; storage, execution, context management, long-horizon support
2Loop EngineeringDesign loops not prompts; verification surface; the six-component practice; loop economics
3Graph EngineeringLoop vs. graph; five signals justifying a graph; failure modes; hybrid architectures
  1. Harness Engineering - what surrounds the model to make it functional and durable
  2. Loop Engineering - deciding when and how to design work as a self-iterating loop
  3. Graph Engineering - deciding when a loop should become a graph instead

Previous: 11 - Prog Langs | Related: 06 - Agentic AI

AI-assisted content - always verify, always explore multiple perspectives·