A free, vendor-neutral course on how coding agents and their harnesses work.
Coding agents are moving faster than anyone’s mental model of them. Vendor docs explain one tool, blog posts explain one trick, and teams adopt agents without understanding the harness around the model: what context it sees, what it is allowed to touch, what runs automatically, and how its work gets checked. The result is either an expensive autocomplete or an agent with more reach than anyone intended.
An AI software factory applies agents across the whole delivery lifecycle. There is no standard design for one yet, but every factory is built from the same blocks, and those blocks are learnable.
Coding Agent Fundamentals is the first course on ai-sw-factory.mellicci.dev, the site about AI software factories. It teaches those building blocks, plainly and without selling a tool. Module 1 · Fundamentals covers:
Every concept comes in three layers: the concept, one real use case, and how Claude Code, Codex and Copilot CLI implement it, side by side. A comparison of the three agents and a glossary sit next to the module.
The course is a site of short concept pages around one interactive diagram: the factory’s building blocks, where every block opens its concept. It is free and needs no account for the public modules.
concept ──► real use case ──► Claude Code · Codex · Copilot CLI
│
└─ one building block in the factory diagramBeing concrete and vendor-neutral at once. Three agents, three release cadences, and features that change week to week: every vendor-specific claim is checked against the vendors’ own documentation and kept current, so a page never teaches last quarter’s behaviour as fact.
Module 1 · Fundamentals is published at ai-sw-factory.mellicci.dev, free and open to everyone.