Designing reviewable AI edits in Snippet Editor

The current proposal-card flow in Snippet Editor, and a planned study of understanding, accepting, and recovering from AI edits.

A suggested change needs to be understandable before it becomes an edit. Snippet Editor already separates those steps: the assistant proposes a card, and the user chooses whether to apply it.

What exists today

The current interface shows a diff and Apply and Dismiss controls. After applying, the card records the result and offers a link to the file. The case-study captures are real interface tests with synthetic snippets and simulated assistant responses.

The next question

Can someone understand the proposal and recover when the file changes elsewhere? That is the question for the next prototype and participant study. The research plan compares inline changes with proposal cards and includes interrupted and conflict states.

What the study should change

The plan is to observe relevant participants completing realistic tasks, record where they misunderstand or need help, then revise the affected interactions. The published case study will add those observations and the resulting design changes after the work has happened.

This is a design plan. Recruitment, prototype validation, and participant findings are not complete.

See the current case study · Read the research plan


More from the journal