I have an old Cordova app for iOS I wanted to convert to native
I have an old Cordova app for iOS I wanted to convert to native (Swift for Apple iOS).
I tasked both Codex (OpenAI) and Claude (Anthropic) to generate a high-level document to recreate it, using the same prompt.
The two output files differ mainly in intent and discipline.
Codex: reads like a true behavioral map of the app. It focuses on what the user sees, what they can do, and how the app responds, using plain English to describe screens, flows, and state changes. That makes it well aligned with the goal of documenting the existing Cordova app without drifting into implementation. Its main weakness is that some areas are lightly specified, which leaves room for interpretation when rebuilding visually or handling edge cases.
Claude: feels closer to a hybrid product and engineering spec. It is more detailed, more prescriptive, and immediately actionable, but it frequently crosses the line into implementation assumptions and inferred behavior. In doing so, it introduces contradictions and subtle feature drift that risk rebuilding a slightly different app rather than the same one.
For the stated goal, the Codex version is stronger as a source of truth. Claude's version is valuable as a reference, but only after the behavior has been locked down elsewhere.