Skip to content

Prompt engineering

Writing instructions for an AI model so that it does what you actually meant, reliably.

It sounds grander than it is. Mostly it's being specific.

The habit that helped me most with the Base44 builder is using its words. It knows exactly what an "entity" and an "access rule" are. Say "database table" or "RLS policy" and you're making it translate, and things get lost in translation.

A few more that earn their place. Say what done looks like. Show it one good example. Say what not to touch, because "don't change the layout" saves a lot of undoing. And split big jobs into steps, so it checks before it changes anything.

That first habit is basically why my open-source Base44 prompt engineer exists. It's a Claude skill that writes prompts for the Base44 builder in the platform's own terms, so they land first time more often.

Related:Agent skill and Claude Code.

Where I've used it: