AI Agents — A Developer's Primer
A ground-up primer on AI agents for working developers. No hand-waving: every note explains its subject from zero and anchors the new ideas to systems concepts you already know — C, bash, Unix, networking.
The notes form a sequence. Start at Module 1 and read in order — each note states its prerequisites and builds on the previous ones — or jump straight to what you need.
Module 1 — The Core Mechanic
What an agent actually is: a model, tools, and a loop. Read these two first.
Module 2 — Protocols
Sharing tools and delegating work across boundaries: MCP and A2A.
Module 3 — Choosing an Architecture
Patterns before frameworks: how to pick the shape of agent a task needs, and what the framework tiers add.
Module 4 — The Production Picture
The six-layer stack between a loop and production, and where agent knowledge should live.
Module 5 — Hands-On with NAT
The whole stack exercised in one small real project, using the NVIDIA NeMo Agent Toolkit.
Supplements
Short supporting reads: the graph structure under planners, and the observability layer.