Lesson 1 of 9 · about 15 minutes

What is a branch?

A sticky note on a commit — and the cheapest safety net you will ever own.

The problem branches solve

Picture this. You ask Claude Code to "refactor the cleaning code in clean.py and speed up the penguins notebook." It edits six files and makes three commits. The plots now look wrong, and every one of those commits went straight onto main, the branch your teammates (and your future self) trust. Undoing it means picking through history by hand.

A branch fixes this before it starts: the agent works beside main, not on it. If the work is good, you bring it in. If not, you throw the branch away and main never knew. GitHub describes branches as a way to "develop features, fix bugs, or safely experiment with new ideas in a contained area of your repository."2

A branch is just a pointer

You already know commits: snapshots of your project, each pointing back to the one before it. A branch is not a copy of your files. The Pro Git book puts it this way: a branch is "simply a lightweight movable pointer to one of these commits."1

Three consequences matter for everything that follows:

• Branches are nearly free. A branch is a small file holding a commit hash. Creating one is "as quick and simple as writing 41 bytes to a file."1 Make them freely, one per task.
• Committing moves only the current branch. Every other branch, main included, keeps pointing where it was.
• Git always knows where you are. A special pointer called HEAD is "a pointer to the local branch you're currently on."1 The next commit moves whichever branch HEAD names.

On GitHub, new repositories call their default branch main.2 Plain git init on your laptop may still call it master until Git 3.0.3 They are the same idea. This course always says main.

Try it: protect main in the simulator

The simulator below is a tiny git that draws its commit graph. main already has two commits. Type the commands into the terminal line. Up-arrow recalls the previous command, and help lists what's supported. You only need two new commands:

$ git branch clean-data   # create a pointer here (does NOT move you)
$ git switch clean-data   # move HEAD onto it

Look at the graph after the last task. Your two commits sit on the clean-data lane, and the main label never moved. That is the whole safety net.

Check yourself

Answer from memory. Retrieval is what makes it stick.

Why this matters for agentic coding

Claude Code can edit many files quickly. A branch per task gives that speed somewhere safe to land. In later lessons we'll call this an agent branch: the agent commits there, you review the result as a pull request, CI tests it, and only then does it reach main.