← All tags

Using AI in apps

Notes on where AI helps inside an app, where a plain rule does the job better, and how we check AI output before anyone relies on it.

How do we decide where AI belongs in an app?

We use one test across all five of our apps: AI should take real work off the person using the app and leave them in charge of the result. If it only adds a feature nobody asked for, it stays out.

OwnSites writes the first draft of a whole website from one sentence, and the owner reviews and changes every page. ProfVault reads job application forms and fills them from stored documents, and the applicant checks every field before submitting.

Brandreel drafts a script and lays out the scenes of a brand film from a short brief, then the person directing it changes any scene in a live preview before anything is rendered. In each case the AI does the slow first pass and a person makes the decisions.

What happens when AI gets something wrong?

We plan for it. Homeroom's reading passages pass a set of rules, then a second model answers the questions without the answer key, and a parent approves each passage before a child sees it. Where rules are predictable, instant and free, as with Homeroom's worksheets, we use rules instead.

If your idea includes AI, the posts below show how we find the part where it earns its place.

Got an idea of your own?

Tell us your idea