Most of the hyperventilated AI “success stories” are as useful as the “ANSIBLE!!!” movement was a few years ago. It’s thus always a pleasure to find someone with well-established software development chops who took the time to describe what works for them.

One cannot argue with Mike McQuaid’s credentials (at least if you happen to be using homebrew on MacOS, which you REALLY SHOULD), and his Sandboxes and Worktrees: My secure Agentic AI Setup in 2026 article is full of relevant recommendations in case you’re brave enough to let AI agents loose on your GitHub repository.