Undo lets the happy path stay uninterrupted
Aliases: undo vs confirm · reversible action · no interruption
What it is
The default recovery strategy for a reversible act is do it, then offer undo, not ask “are you sure” first. Undo places the correction after the act, so the happy path stays a straight line. Confirm turns every act into two decisions and taxes the success path too. This entry is only about not interrupting reversible acts with a confirm. It is not about where the undo control sits, how many seconds remain, or why confirmations get clicked through.
Why it happens
Skilled work is a compiled action chain. A confirm in the middle forces a meta-decision after every step, and the chain stops being fluent. Undo lets the chain finish; correction happens after the result is visible: people see “archived” and then decide whether to pull it back, judging a result rather than imagining a future. A slip (right intention, wrong execution) cannot be caught by a second click in the moment, because the second click is the same automation; the slip becomes noticeable only when the result appears. So interrupting the success path to prevent a slip on a reversible act does not prevent it. It only slows everyone.
Studying it
Build the same delete task as “confirm, then delete” versus “delete, undo for several seconds,” using content the participant just organized.
Independent variables: recovery strategy (pre-confirm vs post-undo), how often the act repeats in the task. Dependent variables: erroneous-act rate, time to complete, number of inserted decisions on the main task, later restatement of “what I just did.”
When laboratory loss is too cheap, the confirm group dismisses the dialog faster and undo’s advantage is understated. Using something they just wrote is closer to real hesitation than a dummy file from the experimenter.
Where it stops holding
Irreversible, high-stakes acts cannot be rewritten as do-then-undo by this entry—when there is no equivalent undo, interrupting the success path is intentional. Where law requires prior consent, undo cannot stand in for consent. In collaboration, others may already have acted on the result (notified, paid); undo then harms a third party and needs a different scope rule. Experts on high-repeat reversible acts benefit most; a first-time destructive act for a novice is still better taught by a visible result plus undo than by an empty confirm.
Applying it
- Inventory every “are you sure” on a reversible act (archive, remove from view, unstar, revert to draft) and change it to immediate execution with undo on the result.
- Stop inserting modal questions on the success path; if a slip must be caught, catch it with a visible result and a later correction, not a second click beforehand.
- Spend the confirmation budget on acts that truly have no equivalent undo, so reversible and irreversible do not share the same interruption.
- Verify by walking the main task and counting dialogs that block the success path. If those blocks are reversible acts, the strategy has not switched.