// track · 9 lessons
Reversible by default, gate the irreversible
Speed on what you can undo. A human on the one button you can't take back.
- 01Reversible by default, gate the irreversibleLet the agent move fast on anything you can undo. Put a human on the one button you can't take back.
- 02Sort every action by whether you can undo itSplit every action into reversible and irreversible before you touch it. That one line through your workflow decides how much care each step earns.
- 03Speed on the reversible, ceremony on the irreversibleMove fast on anything you can undo and slow way down on anything you can't. Most people get this exactly backwards and pay for it.
- 04The blast-radius testFour questions find your irreversible actions fast: does it move money, deploy, send, or delete? Those are the ones that need a gate.
- 05The autonomy ladderDon't grant an AI full autonomy on day one. Earn it one boundary at a time: read-only, then draft-only, then approve-then-send, then loose.
- 06Human-in-the-loop at the irreversible edgeAutomate the whole workflow up to the irreversible step, then put a human on that one button. It is the highest-leverage seam in any AI system.
- 07Build the gate as a switch you can flipBuild the irreversible gate as one switch: draft-only now, auto-send later. Same code, promotable when the system has earned your trust.
- 08When "we can always send a correction" is a lieThe most expensive reversibility mistake is calling an irreversible action reversible. A correction is not an undo. It is proof the first one shipped wrong.
- 09Reversibility is a day-one design constraintReversibility isn't something you add after a scare. Design it in from the start with drafts, branches, dry-runs, and backups, so undo already exists.