Designing AI that asks before it acts
Why Remora treats AI extraction as a suggestion that a person must confirm, and keeps every core feature working without AI.
Status: Design decision. The AI extraction phase has not been built yet.
Remora helps people keep track of subscriptions, bills, warranties and renewals. A planned feature reads documents and screenshots and suggests the commitment they describe. The product decision behind that feature was made before any of it was built: AI never silently creates a commitment.
Why confirmation is mandatory
In a product whose job is to remind you of the right thing at the right time, a wrong date is worse than a missing one. A missing commitment is visible; a wrong one quietly fails you. So any extracted value — a renewal date, an amount, a provider — is shown to the person to confirm or correct before it becomes a commitment.
AI as infrastructure, not the product
- Manual entry is a first-class path, and core manual functionality is designed to work without AI and offline.
- The AI provider sits behind an abstraction, so it can be changed without rewriting the product.
- Extracted output is validated and records where it came from before it is offered for confirmation.
What this means in practice
Designing the confirmation step first shapes everything after it: the data model stores provenance, the interface is built around review rather than automation, and the product still works for people who never use the AI features at all.