The prototype already runs on iPhone and Mac, but its subscription path and bring-your-own-key setup make the first release harder to understand and operate. Ray Fernando concludes that asking ordinary users to create an OpenAI platform key and store it in Keychain creates too much friction for an App Store product.
His first simplification is to make the initial iPhone version work without an external server. The plan is to use an on-device foundation modelOn-device artificial intelligence runs AI models directly on a phone, computer, vehicle, or other local hardware instead of sending every task to a remote server. for transcription, preserve live text, adjustable text size and saved notes, and postpone the Mac version and paid cloud features until the core experience is stable.
Ray also describes small product touches that distinguish a functional prototype from a polished app. After the first transcription, the app shows a personal thank-you card with a creator message and avatar. He treats this kind of human context as part of the product design, not a feature that should wait until the technical work is finished.
Rather than immediately deleting code, Ray asks a Cursor cloud agentAn artificial intelligence coding agent is a software system that can plan and carry out bounded development tasks by reading code, editing files and using development tools. to inspect the repository and produce a detailed specification for the replacement. The agent records the desired iOS version, local-model requirement, retained interactions, iCloud syncing plan and useful existing snippets so the rewrite has a tested behavioural target.
The workflow uses a stronger model as the planning brain and fans lower-cost agents out across file reading and research tasks. Because the work runs in Cursor's cloud environment, Ray can leave his laptop, monitor progress from the mobile app and answer questions remotely. The practical lesson is to rewrite only after converting the working prototype into an explicit specificationSpecification-driven development begins with an explicit description of required behavior and uses that specification to guide implementation and verification. that protects its useful behaviour.
Watch on YouTube
