How far can Codex take a first personal webpage? One case review

01The result: a usable page, with a verification gap

Beginners are the audience of this review, not the participants in an experiment. An existing Codex session performed the task. This demonstrates one result, not a general first-attempt success rate for new users.

02Task and evidence

The task requested a warm, light HTML/CSS page for fictional Lin, with photography, hiking, reading, and City Walk Notes. External assets and services, sign-in, and payments were excluded.

Environment and sample

October 2, 2026; local macOS workspace; Codex in-app browser; one task, one draft, one planned revision.

Missing records

App version, exact model identifier, reasoning settings, token usage, and cost. No repeated trials.

Original Chinese requests ↓ · Full walkthrough →

The original task was local only. The example was later published with this handbook; that publication is outside the original demonstration’s completion criteria.

03What changed

The draft already contained the four requested content items in index.html and style.css, with CSS decoration and a native project disclosure.

The next change was planned teaching feedback, not a discovered beginner usability problem or a retry needed to rescue a failed task.

Desktop bio: 17px → 20px; mobile bio: 16px → 18px; spacing above interests: 26px → 38px. HTML was unchanged. The change matched the request, but no reader study measured a readability improvement.

04Separate passed checks from missing checks

• Content: all requested fictional content was present.
• Responsive layout: no overflow at 1280 / 390 / 320px in simulated viewports, not physical-device tests.
• Interaction: disclosure expanded and collapsed; no complex business features were tested.
• Revision scope: CSS type and spacing changes only; feedback was planned.
• Dependencies: no external font, image, script, or service references in this simple static example.
• Direct file opening: unverified because automation disallowed file://; this is not proof of a product failure.
• Other browsers and real devices: unverified.

The supported conclusion is that local HTTP preview worked. A perfect score or overall success rate would hide missing checks.

05Time and cost

The recorded window was 16:20:42–16:22:23 Beijing time, or 101 seconds. It included reading the task, tool round trips, browser checks, screenshots, revision, and packaging. Later article editing and deployment were excluded.

This is not “a beginner built a website in 101 seconds.” Beginner setup and understanding were not measured. Cost and token usage were unavailable. No recorded hand edits does not imply zero total cost.

06What this means for your first attempt

Under these recorded conditions, Codex delivered an inspectable static page and applied specific feedback when content, constraints, and checks were clear.

1. Prepare a nickname, bio, interests, and one project description.
2. State your constraints and completion criteria.
3. Open the actual files and check content and interactions.
4. Request one concrete revision and verify both the change and what should remain unchanged.

This case did not test accounts, databases, payments, complex apps, or competing models. Those require separate tasks and evidence.

07Inspect the artifacts

Start with the first webpage lesson or read the review method. A future beginner study would need independent participants, environment details, failures, and effort records. Original screenshots and source downloads remain in Chinese.