Edit Claude's work inline instead of re-prompting

Step 1 of 7

01 / 07Overview

Understand the problem this solves

The old loop: Claude writes 800 words, paragraph three is wrong, you reply "fix paragraph three", and get back 800 *different* words — the good parts rewritten along with the bad.

Inline editing breaks that loop: you select the exact words that are wrong inside the artifact, and Claude changes only the selection. Everything else stays byte-identical.

Checkpoint
You can explain why 'fix paragraph three' in chat is riskier than selecting paragraph three.