How GitHub works with AI coding tools

What are the four words you need?

Everything else in Git is an elaboration of these four. You can work productively for months knowing only these.

Why is the commit history the best AI audit trail you have?

A chat transcript tells you what the AI said it did. The diff tells you what it did. When a feature stops working after a session, comparing the last two commits is usually faster than re-reading the conversation — you see the exact lines that changed and nothing else.

1. 1Open the repository's commit list.
2. 2Find the last commit from before the problem appeared.
3. 3Open the next commit and read the diff.
4. 4Look specifically for changed config, changed permissions and deleted lines — deletions cause more breakage than additions.

What does an AI tool actually do to my repository?

• Reads files to build context before answering.
• Writes new files and edits existing ones.
• Creates commits, usually with a generated message.
• Sometimes works on a branch and opens a pull request instead of committing directly.

It does this with a token you granted. That token has the permissions you approved — which is worth reading rather than clicking past, since 'write access to all repositories' and 'write access to this one repository' are very different grants.

Do I need branches if I am working alone?

For small changes, no. For anything risky — a payment flow, a database change, a redesign — yes, because a branch gives you a preview deployment and a clean way to abandon the attempt. The cost is one extra click; the benefit is that your live site never sees the experiment.

What breaks most often at this seam?