Git Algorithms

A branch does not copy commits. It is a label pointing at a single commit, so creating one costs nothing — and the line forks only once you move that label on its own. When the lines meet again, choosing between fast-forward, merge commit, squash and rebase is really choosing what shape to leave in the history, and that shape sets how hard it will be to undo later.

Working alone, nothing stands between a commit and production. That is not because you skip review but because there is nowhere for a review to stand, and a branch is the device that creates that place. Once AI writes the code a solo developer gains a reviewer for the first time — but handing that review to the session that wrote the code makes both fields of view coincide, so they miss the same things.

The branch picture barely changes as a team grows. What changes is what stands in front of the merge — and once AI writes the code, output multiplies while reading speed stays put, so a queue forms in front of approval. Adding approvers does not catch up; narrowing where human attention is spent does.