Chapter 3 · Section 04 · On the clock

Buildyour page

One folder · one file · your decisions

Before you begin

Ready to build?

One file called index.html, in your personal-site folder — that's the whole deliverable, and it's small on purpose. A good version introduces you clearly, shows selected work, gives a recruiter one next action, works at laptop and phone widths, and contains nothing private. No framework, no account, no deployment — putting it on the internet is next session. Check these first; they prevent the two most common stalls.

Check

Fourteen minutes on the clock. The countdown on the screen at the front is the clock every milestone on this page refers to. That's deliberately not enough time to perfect it. Finishing something small and clear beats starting something ambitious.

Pace yourself against the clock

Time left on the room timer — not the time of day.

12:00 left The interview is over, the brief is approved and the build has started.
8:00 left A first version is open in your browser.
5:00 left You've checked the story, the visuals and the links.
2:00 left You've made one specific revision. The point
0:00 Stop building — the stranger test is called from the front.

Step 1 of 2

Let Claude interview you
before it builds

The move worth stealing from today: make it extract the brief before it generates anything. A vague desire becomes an explicit brief — on the record, where you can correct it.

Interview prompt

Help me build a one-page personal website as a single self-contained index.html. Before coding, interview me one question at a time — no more than four — about the audience, the impression I want to leave, the evidence I should show, and the visual direction. Summarize the brief and wait for my approval. After approval: create or replace index.html in this folder. Tell me when it's written — I'll open it myself.

Paste it into the box where you type in the Code tab — the same box you used in setup, with personal-site open — and press Enter. Answer its questions one at a time; rough answers are fine. Then approve the brief, and approve the edit when Claude asks: in Manual mode nothing is written to your folder until you say yes.

Edit it freely — but keep two things. The cap of four questions stops it interviewing you for the whole exercise. The approval gate is where your judgment enters. Everything else is yours to change.

What should happen

Ask One question at a time — not a wall of four.
Answer Briefly and honestly. Rough answers are fine.
Brief It plays your answers back as a short brief.
Approve Correct anything wrong, then say go. Your call
With 12:00 left on the room timer, the interview should be over, the brief approved and the build started.

Done when

If something's wrong

I pasted it and nothing is happening

Look for an approval prompt in the chat and accept it — in Manual mode Claude waits for you before it writes anything. If there's no prompt and no reply, press Enter once more in case the paste didn't send.

Still nothing after two minutes? Raise your hand — show us the screen, don't start over. Don't retype the prompt.

It skipped the interview and just built something

Say: "Stop — you skipped the interview. Ask me the four questions one at a time before changing anything else." Then keep going. Redirecting mid-flight is normal.

Its questions are generic, or it asks all four at once

Push back in the same sentence you answer with: "One at a time, please — and ask me something you couldn't have asked anyone else in this room."

I don't know what to say about "visual direction"

Name two adjectives and a reference. "Quiet and serious, like a consulting firm's about page" is a complete answer. Or ask it to propose three directions and pick one — choosing is faster than inventing.

The first version looks bad

Good — that's the exercise starting, not failing. Don't say "make it better." Name the observed problem and protect what works: "The headline is too small and the spacing is cramped. Keep the colours and the structure."

Step 2 of 2

Build in three passes

One concern per pass. Finish a pass before adding features — features multiply, passes converge.

The passes

  1. Content — name, story, evidence Ask yourself: is the story clear? Nothing else matters until it is.
  2. Direction — type, colour, layout Ask yourself: does it feel like me?
  3. Polish — links, widths, wording Ask yourself: does everything actually work?

If a pass doesn't pass — example prompts

One fix per message, and say what to protect. These are examples: describe what you saw on your page.

Content didn't pass · example

A stranger can't tell what I do or why it matters. Rewrite the opening so the first two lines say who I am and the one thing I want to be remembered for. Keep the sections and the design exactly as they are.

Direction didn't pass · example

It doesn't feel like me — it reads corporate and generic. Change only the type, the colours and the spacing to something warmer and simpler, two colours at most. Don't touch the words or the section order.

Polish didn't pass · example

Two things don't work: the LinkedIn link goes nowhere, and at phone width the name overlaps the photo. Fix those two only, then tell me what you changed. Everything else stays as it is.

Stay in Manual today. Setup left you approving every edit, and on a build this short the clicking is the point: you see each change before it lands, which is what makes these fourteen minutes a skill rather than a demo. Widening permissions is a decision you make on evidence, deliberately — and you haven't collected much evidence yet.
Stuck? Show us your screen — don't start over. Starting over throws away the brief, which is the expensive part. A helper can usually unstick a session in under a minute.

Done when

If something's wrong

Every change breaks something else

You're asking for too much at once. Make one change per message, and say what to protect: "Change only the header. Leave everything else exactly as it is."

If it's badly tangled: "Go back to the version before my last three changes."

I'm polishing colours and the story still isn't clear

Very common, and worth catching in yourself. Pass 2 is more fun than pass 1 — which is exactly why pass 1 gets skipped. A clear unstyled page beats a beautiful vague one, and the debrief will show it.

I've run out of things to ask for

Ask Claude: "What's the weakest part of this page for a recruiter who has eight seconds?" Then fix that one thing.

I'm out of time

Stop building. When the swap is called, the four questions are on the screen at the front — trade laptops and bring whatever you have. A half-finished page that a real person has read beats a better page nobody has seen.

The folder isn't going anywhere — you can keep going after class.

You built something.

You turned a vague intention into a written brief, approved it, watched a file appear on your own disk, and steered it with specific corrections. That loop — observe, describe, change, check — is the entire job.

Next: the same loop, pointed at something that exercises judgment instead of following instructions.