claude vs codex: login

Started building the actual app. Decided to run two AIs in parallel on the same brief — Claude (Opus 4.7, 1M context, max effort) in one project directory, ChatGPT (Codex 5.5) in another. Same project bible, same design language doc as input. Port offsets — Claude on +30, Codex on +40 — so both can run at once. Claude got the custom-agents-and-commands setup I’m used to (8 agents). For Codex I asked it to mimic that workflow as best it could.

This entry covers only the login page. Registration flow and the rest come next.

Claude (Opus 4.7):

Claude’s login screen

Built a marketing-style splash at / instead of a form. Cube, “GamaHaus”, “Playground, not workplace.” (the exact tagline from the design doc), product copy, magenta “Get started” + outlined “Log in” CTAs. Top-right has a quieter “Log in” link and “Sign up” pill. Logging in is one extra click; the first impression is positioning.

ChatGPT (Codex 5.5):

Codex’s login screen

Built an actual login form at /login. Two-column split — left side is branding (cube, “GamaHaus”, tagline, decorative pastel blocks), right side is a card with email + password, magenta “Sign in”, “Continue with Google”, and a “Create an account” link. Tagline is improvised: “Plan, publish, and remix playful home ideas with a local-first workspace.” — different voice from the doc.

What diverged:

  • Scope reading. Claude took “login page” to mean “the entry surface for unauthenticated users” and built a full splash. Codex read it literally and built the form.
  • Brand fidelity. Claude pulled the exact tagline and CTA hierarchy from the design language. Codex used the brand colors and fonts but improvised the copy and added decorative elements not in the doc.
  • Auth surface. Codex shows Google SSO, email/password, and signup all in one view. Claude defers all of that to a later page.

Neither is wrong for a brief that just said “login page.” Different inferred intents from the same starting documents. Keeping both projects running through the next slices and seeing whether the divergence widens or they converge as more context lands.

#gamahaus #ai