Claude is for entreprise-grade, Codex is for UI/UX + state mngt work

I've been using Claude end-to-end on Hanko: Coffee Passport — landing site, admin panel, backend, and the Kotlin Multiplatform mobile app itself. For enterprise-grade systems work it's hard to beat: architecture, backend logic, infra, cross-cutting concerns.

Last night I realized I'd been under-using Codex.

Codex is noticeably stronger at QA and PO-style review work — catching edge cases in flows I'd glossed over — and at general UI/UX development and state management. Cleaner reducers, less accidental complexity, better feel for the interaction layer.

Part of why: Codex can drive iOS and Android simulators directly. It runs the build, navigates the app, and verifies behavior on the actual surface instead of reasoning about code in the abstract. For a KMP project where UI correctness has to hold on both platforms, that shortens the QA loop dramatically.

The takeaway: these tools aren't interchangeable. Claude handles the system, Codex handles the surface. On Hanko that now means Claude keeps the KMP shared module, backend, and admin; Codex takes iOS/Android UI passes, state refactors, and PO review.

Small shift, but it's already tightening my iteration loop.