The original title is "How to write 'done when' criteria for AI-assisted app development"

Short summary

The author argues that telling an AI coding tool to 'finish' a feature is vague and leads to incomplete work. Instead, every feature request should include a 'Done when' line specifying a testable user outcome. The post breaks down four components of a good done-when line: naming user state, using concrete verbs, defining trust conditions, and scoping ambition for version one.

• •'Finish' is too vague for AI coding tools — use a 'Done when' line with a testable user outcome
• •A good done-when line names user state, uses concrete verbs, defines trust conditions, and scopes ambition
• •Includes a link to free AI App Builder Starter Prompts on Gumroad

Generated with AI, which can make mistakes.