Notes on version control for the AI era
Practical writing about provenance, intent, recovery and what changes when half your commits come from a model. No filler — each piece exists because someone asked the question.
Undo is a design decision, not a feature
Compare undo models: git reflog, jj op log, kal ops. What makes a safety net complete vs partial. Why operation logs beat ref
How to tell which code an AI wrote — and prove it later
Git records one author string, so an agent looks exactly like a colleague. The approaches teams use to track AI-generated code, and where each breaks.
You ran git reset --hard. Here's how to get your work back
A practical recovery guide for the commands that eat your work — reset --hard, a bad rebase, a dropped stash, a deleted branch.
Your git history records what changed, never why
Commit messages describe the diff you can already read. Why the reasoning behind a change keeps getting lost, and what to do about it.
Four questions about AI-generated code your team can't answer yet
Auditors, acquirers and insurers have started asking what share of your codebase a model wrote. What to have ready before you're asked.