Best practices

Small habits dramatically improve speed, credit efficiency, and reliability when building with AI.

Prompting for code output

Be explicit about layout, component boundaries, data dependencies, and empty states. Mention target record types and workflow names so the assistant wires APIs instead of inventing dead mocks.

Request incremental changes after each successful apply.
Paste error snippets from preview when something fails to compile.
Call out accessibility or SEO requirements up front.

Data and workflows

Model entities once in record types, then reference slugs consistently in workflows and React hooks. Prefer save_record + email notification patterns for lead gen rather than only client-side state.

Publish record schemas before advertising the app so public forms never 404 against stale type slugs.

Translations and locales

Freeze copy keys before handing off to translators. Generate the skeleton in English, export strings if needed, then ask the assistant to add locales with parity.

Reliability and backups

ZIP-export before risky refactors or tier downgrades. Keep test projects for experiments so production stays stable. Monitor workflow execution history weekly to catch silent integration failures.

Treat service role keys and webhook secrets like production credentials—never commit them to generated repos.