This is part three of my Basics series. The first post was about how I structure CLAUDE.md after 1,000+ sessions, the instructions file that tells your AI agent who it is and how to behave.
This one is about Git and GitHub, the version control layer underneath all of it, written for builders who are shipping with AI agents but never formally learned the workflow.
Here's what's inside:
- Git vs. GitHub, the actual difference: Why Git is the tool on your machine and GitHub is just one of many hosts (GitLab, Codeberg, Forgejo). If you've been using the words interchangeably, this is the part that clears it up.
- The setup that actually works: Installing Git, configuring identity, initializing a repo, first commit, pushing to remote. The exact commands I run on a fresh machine, in order.
- Commits as checkpoints, not chores: When to commit and what to write in the message. The rule I follow: commit before any big change, so I can roll back when the AI agent goes sideways.
- Branches and worktrees for parallel AI work: How I isolate experiments and run multiple agents at once without them stepping on each other. This is the part most tutorials skip.
- Why AI agents read your commit history: Claude Code, Codex, and the rest use your git log to understand context. Sloppy commits give you sloppy sessions. Clean commits make the agent smarter.
The part that surprised me most when I started doing this seriously is in the full post.
Full post: https://thoughts.jock.pl/p/how-to-use-github-ai-builders-basics-2026
Free weekly: https://thoughts.jock.pl
Top comments (0)