Change review

Added · Seed

Seeded from looking at different AI pattern libraries and sorting them out. It’s not quite ready yet because I haven’t had a chance to consider it closely.

A proposed change to an existing document, dataset, or codebase is shown as a change — what was, what would be — so the actor can judge and accept it in parts rather than take or leave the whole.

Forms

Recurring forms: side-by-side and inline diffs, per-change accept/reject, track-changes marking in prose, and review queues where accumulated proposals wait for a pass.

Consequences

  • the delta is a reviewable object — what was, what would be, side by side or in place — so judgement lands on the change rather than on re-reading the whole, and an approval has something concrete to act on
  • the diff's granularity becomes the decision's granularity: whatever the review can show separately is what can be accepted or rejected separately
  • reviewing deltas narrows attention to what changed, so consequences in the unchanged surroundings are exactly what the review doesn't show

Resources & references

  • AI UX Playground — Smart Diff
  • Long pre-AI lineage: version-control code review and word-processor track changes are this pattern without the machine author.

Appears in

Delegating work

Related patterns

Precedes

  • Approval gate — the delta is a reviewable object — what was, what would be, side by side or in place — so judgement lands on the change rather than on re-reading the whole, and an approval has something concrete to act on

Enacts

  • Agency — the actor disposes over each change instead of inheriting the batch

Complements

  • Undo — rejecting a proposed change is recovery moved before the commit
  • Draft and publish — the proposal holds in a draft state while the review decides what publishes
  • Commenting — review talk anchors to the delta rather than to the document at large

Related to

  • Plan preview — the plan is the intention before the work; the change review is the delta after it
  • Steering surface — what the instruction surface can show of a change it caused, when the change landed on content it does not display

Preceded by

  • Generated content — the proposed change is generated output aimed at existing content
  • Dry run — the run produces a before-and-after over cases, which the actor may want to accept or reject one at a time rather than as a whole