Version control

The benefit is fearless experimentation. Because any change can be reversed, trying something risky costs almost nothing. This is basic infrastructure even for a one-person project, and it is what other practices assume: branches, pull requests, automated testing and deployment all rely on a history existing. The log also works as documentation. Months later, the trail of commits is often the only record of why something behaves the way it does.

It matters even more when an AI agent writes code. Agents make fast, large changes, and sometimes wrong ones. A history lets a person see exactly what changed and step back to the last good state.

• Commit small, coherent changes, each with a message that explains why, not only what.
• Work on a branch for anything beyond a trivial fix, and keep the main line working.
• Write messages for the person who will read them in six months.
• Give an AI agent the same rules as a human contributor: its work goes through a branch and a review, never straight onto the main line.
• Note significant points in the history, such as releases, so a rollback target is easy to find.

Version control keeps a complete history of a project. Each saved change is called a commit: a snapshot of the files plus a short message saying what changed and why. Instead of one fragile folder that a single bad edit could wreck, the project becomes a timeline that can be searched, compared and rewound. Git is the tool almost everyone uses, and GitHub is a popular place to host the history online.