Video by freeCodeCamp.org · taught by Tom Chant · streamed from YouTube. All rights remain with the creator.
Lesson 1: Agentic Coding: Directing an AI Coding Agent
About this lesson
Coding agents can write and change software for you, but getting output you can rely on is a skill of its own. This lesson teaches that skill using OpenAI's Codex as the working tool, through realistic tasks. It is organised in six parts. The first covers prompting and project setup: how to describe what you want clearly and how to prepare a project so that an agent starts with the right information. The second introduces plan mode, where the agent proposes its approach before touching any file, and skills, which package instructions for tasks you repeat. Part three is about testing the changes an agent makes, including characterization tests that record how existing code behaves before it is modified. Part four deals with Git and with keeping secrets such as passwords and keys safely out of the agent's reach and out of your repository. The fifth part goes deeper into managing context, setting permissions that control what the agent may do without asking, and the configuration file where these choices live. The final part connects the agent to outside tools through the Model Context Protocol and uses a component library to build interface elements properly. The course is aimed both at experienced developers who want to add AI to their workflow and at people with little coding background who want to build software with AI help. Afterwards, read Chapter 1 of the course book, which explains the agent loop, shows how quickly a context window fills, and sets out permission levels in a table.