A habit that pays for itself immediately: commit (or stash) before asking the agent to make changes.

AI agents touch multiple files at once. Without a clean baseline, it’s hard to tell what the agent changed vs. what was already there — and harder to undo just the parts you don’t want.

With a checkpoint:

  1. git commit -m "checkpoint" (or git stash)
  2. Let the agent work
  3. git diff to see exactly what changed
  4. Keep what’s good, git checkout -- <file> to revert what isn’t

This is especially valuable for agentic workflows (Copilot CLI, Claude Code, Cursor) where the agent runs multiple steps autonomously. In chat-based workflows you review inline diffs, but in agent mode it might edit five files before you look.

I’m a fan of small commits — even micro-commits. Getting AI agents to commit that granularly is still hard, but at least make sure you commit before handing over control. That way the agent’s changes are always one git diff away from a clean state.

Save your game before the boss fight.

💡 Try this: Before your next AI-assisted change, run git stash or make a quick commit. After the agent finishes, run git diff and review the changes as if they came from a colleague’s PR.

🔗 Review AI-generated code edits (VS Code docs)

Do you create a checkpoint before AI-assisted changes?

🟢 Always — commit or stash first
🟡 Sometimes, when it’s a big change
🔴 No — I rely on undo / inline diffs
⚪ I mostly use chat, not agent mode

Any git tricks that work well with AI workflows?