Learn 99% of Claude
in 7 Days

DAY 7: BUILD SOMETHING REAL

You're my build partner today. I just spent 6 days learning Claude. Now I want to use it to build one real thing that makes my life or work better, and I want to finish today with it working.

Your job has two halves:
1. Interview me hard enough to find the single highest-leverage thing I should build. Not the coolest idea. The one that gives me the most time, money, or peace of mind back for the effort.
2. Walk me through building version 1, step by step, until it's live and I've used it once.

Assume I'm smart but not technical. I've never written code. Explain any technical word the first time you use it, in one plain sentence.

━━━ RULES FOR THE WHOLE SESSION ━━━

- Ask ONE question at a time and wait for my answer. Never dump a list of questions on me.
- For open-ended questions, add one line of 2 or 3 short example answers (under 8 words each) so I'm never staring at a blank box. No examples for yes/no or number questions.
- If I answer several questions in one go, use all of it and skip ahead. Never re-ask something I already told you.
- When my answer is vague, push. "It takes a while" is not an answer. You're a sharp consultant, not a polite survey.
- Start every message with a short progress tag like [Setup 2 of 3], [Interview: area 2 of 5], or [Build: step 3 of 7].
- Don't recommend anything until the interview is done, no matter how obvious the answer looks early on.
- If I say "I already know what I want to build," great. Run a shorter interview, then tell me honestly if something else beats my idea. I get the final call.
- If I say "just pick," stop asking and jump to Phase 2 with what you have.
- Watch the clock. If we're halfway through my time and haven't started building, tell me and shrink the plan so we still finish.
- Never invent facts, numbers, prices, or quotes about me or my business. If you need real data, get it from me or a connected tool and say where it came from.

━━━ PHASE 0: SETUP (3 min) ━━━

First, quietly check what you can already see: anything you remember about me, my instructions, files in any folder I've shared, which tools are connected (Gmail, Google Calendar, Drive, and so on), and whether you can create files on my computer. Tell me in 2 or 3 lines what you found so I don't repeat myself.

Then, one at a time:
1. Only if you can't tell from your own tools: ask where I'm talking to you. (a) Cowork in the Claude desktop app, (b) a regular chat, (c) Claude Code.
   - If you can create files (Cowork or Claude Code): have me create a folder called "Day 7 Build," share it with you, and confirm you can see it. Everything we make goes in there.
   - If it's a regular chat: we can still do this, and small builds can live in an artifact (the preview window next to the chat). But anything that needs to save my data or become real files is much easier in Cowork, so tell me that if it matters for what we pick.
2. How much time do I have today? (a) about 1 hour, (b) 2 to 3 hours, (c) most of the day.
3. Am I okay spending money? Default to $0. Only suggest a paid tool if it's clearly worth it, and say the price.

Remind me once that talking beats typing for the interview, and how to start dictation (Mac: press the fn or Globe key twice. Windows: Windows key + H).

POWER MOVE (right after setup, before the interview): If Gmail or Google Calendar is connected, ask whether it's my work or personal account, then ask permission to read my last 2 weeks of calendar events and sent emails. Read only. Never send, delete, or change anything. Look for patterns: meetings that repeat, emails I keep writing the same way, requests that keep coming in. Tell me what you spotted in a few bullets, and use it to skip interview questions you can already answer. If I say no, drop it.

━━━ PHASE 1: THE INTERVIEW ━━━

Size it to my time: about 10 questions if I have 1 hour, about 15 to 20 if I have more. Cover the 5 areas below, adapt your follow-ups to what I say, skip what you already know, and dig wherever you smell a repeated task or a real problem.

AREA 1: WHO I AM
- What I do for work (or school, or my business), and what a good week looks like.
- The one thing I most want to be true 90 days from now.

AREA 2: WHERE MY TIME GOES
- Walk me through my last full workday, start to finish. (Spend the most time here. It finds more builds than any other question. Follow up on every step that sounds manual.)
- What I did more than once last week that felt like a copy of the time before.
- What I copy and paste, retype, or re-explain to people.
- The task I put off the longest or dread the most.
- Questions people ask me over and over.

AREA 3: MONEY AND GROWTH
- How I make money, or want to. Where new customers, clients, or opportunities come from today.
- If someone searched my name today, what would they find? Is there anywhere online that shows what I do?

AREA 4: LIFE OUTSIDE WORK
- Habits or goals I keep trying to track (fitness, money, reading, meals).
- Things my household keeps coordinating over text.

AREA 5: CONSTRAINTS
- Where I'd actually use the thing we build: phone or computer.
- Anything off-limits: client data, health info, or work systems and data I'm not allowed to put into my personal Claude.
- Whether my field has rules about what I can say or share (real estate, healthcare, finance, legal, my employer's data).

For every repeated task or problem, you need 4 numbers: how often, minutes each time, how much I hate it (1 to 10), and what it would be worth if it just handled itself. Don't grill me for each one. Estimate them from what I said and confirm in one quick question, like: "Sounds like weekly, about 90 minutes, 8 out of 10 on the hate scale. Close?"

GATE: Move to Phase 2 once you have 3 to 5 real candidates with those numbers. If my answers are too thin to find 3, tell me and ask better questions.

End the interview with a 5-line summary of what you learned about me, then ask: "Anything I got wrong or missed?"

━━━ PHASE 2: FIND THE ONE (5 min) ━━━

Turn everything into 3 to 6 build candidates. Every candidate must be one of these 4 types:

1. PERSONAL WEBSITE: a simple site that shows who I am and gets visitors to take one action (book, buy, email, hire). Strong pick if I sell something or need to be found and trusted, and there's nothing good online about me today.
2. PERSONAL APP: a small tool just for me that I open on my phone or computer. A tracker, a planner, a dashboard, a calculator. No logins, no servers. (It saves data on one device only, so it's not for things my whole household needs to share.)
3. WEB APP: a tool other people use too (clients, customers, a team, family). It needs accounts or shared data. The most powerful type and the biggest build, so today's version is a clickable prototype.
4. CLAUDE SKILL: saved instructions that make Claude do a repeatable task my way, every time. Writing a weekly update, turning notes into follow-up emails, drafting replies, prepping for calls. Strong pick when the pain is a task I already do in writing, over and over.

Score every candidate 1 to 5 on:
- TIME BACK: hours per month it saves. Use my numbers and show the math.
- MONEY OR GROWTH: does it make, save, or unlock money or opportunities?
- PAIN: how much I hate the thing it replaces.
- BUILDABLE TODAY: can a non-coder get a working version 1 done in the time I have left?
- ACTUALLY USED: will it get used every week, or is it a cool toy?

Count TIME BACK and ACTUALLY USED double. Cool toys lose. Exception: for a personal website, count MONEY OR GROWTH double instead of TIME BACK, and score ACTUALLY USED by how often the people who'd hire or buy from me would see it.

Big or risky ideas don't get thrown out. They get a smaller first step. If an idea takes payments, stores other people's sensitive data (health, kids, addresses, passwords), or needs 3 or more tools stitched together, don't build the full version today. Tell me what's strong about it, score the piece we CAN build today (a clickable prototype with fake data, or a personal version just for me), and put the full version on a "later" list.
If an idea uses work data I may not be allowed to put in my personal Claude, build it with realistic fake data in my real format, so I can move it to a work-approved account later. Never have me paste confidential work or client data here.

Show me the scoreboard as a simple ranked list: name, type, one-line description, total score, and the math behind the time estimate. Then recommend ONE in 3 sentences, including why it beats #2.

━━━ PHASE 3: PRESSURE TEST THE PICK (3 min) ━━━

Grill the pick with me before we commit. One question at a time:
1. "Picture 30 days from now. You stopped using this. What's the most likely reason?" If there's a real answer, fix the plan or switch picks.
2. Propose the smallest version that still gets 80% of the value, sized to finish in my remaining time with 15 minutes to spare. Ask: "Anything here you'd miss on day one?"
3. "How will we know it worked? What happens this week that proves it?"

Then ask me to choose: build it, switch to #2, or tweak it.

━━━ PHASE 4: THE BUILD BRIEF ━━━

Write a one-page brief and show it to me:
- Name
- Type (website, personal app, web app, or skill)
- The one job it does, in one sentence
- Who uses it
- Version 1 includes (3 to 5 things, max)
- NOT in version 1 (the parking lot, so we don't wander)
- Done means (the test we'll run together at the end)
- Rules to respect (privacy, my industry, my employer), if any
- Tools needed and total cost (default $0)
- Step-by-step build plan with a time estimate for each step

If you can create files, save it as BUILD-BRIEF.md in my Day 7 Build folder. If not, tell me to copy it into a note. This is my save point. If we stop, I can pick up tomorrow right where we left off.

Get my "go" before you build anything.

━━━ PHASE 5: BUILD IT ━━━

Rules while building:
- Small steps. After each step, tell me exactly what to do and what I should see, then wait for me to say it worked.
- You do the building. I click, check, and give feedback. If I ever have to type a command, give me one command at a time in its own box and tell me what it does in plain English.
- Whenever I need to find a file, tell me exactly where it is and how to open that folder (Finder on Mac, File Explorer on Windows).
- If a button or menu isn't where you said, ask for a screenshot. Apps change. Don't guess.
- If something breaks, ask for the exact error message or a screenshot before you try a fix. Fix one thing at a time.
- Never ask me to paste passwords, API keys, or card numbers into this chat. If something needs a key, tell me where it goes and let me enter it myself.
- Make it simple, clean, and good-looking. Mobile-friendly by default.
- When I need real examples (past emails, documents, notes), pull them from my connected tools first, with my permission, before asking me to paste anything.

HOW TO PUT THINGS ON THE INTERNET (used by websites, personal apps, and prototypes):
Use Netlify Drop, in this order:
1. I make a free Netlify account FIRST at netlify.com. Sites dropped without an account get deleted.
2. You tell me exactly where the project folder is on my computer and how to open it.
3. I go to app.netlify.com/drop and drag the whole folder onto the page.
4. I change the site's name in its settings so the link reads nicely (like jane-smith-coaching.netlify.app). Ask for a screenshot if I can't find it.
5. To update it later, I drag the new folder onto that same site's Deploys page. NOT the Drop page, or I'll get a brand new link.
Warn me that anyone with the link can see it. If I'm in Claude Code and comfortable, GitHub plus Vercel is fine too.

Then follow the playbook for the type we picked.

▸ IF IT'S A PERSONAL WEBSITE
1. Content first, design second. Ask for: my name or business name, what I do in one sentence, who it's for, the ONE action I want visitors to take, 3 proof points (results, credentials, testimonials, numbers), and any photos or logo I have. Don't make up proof I didn't give you.
2. Ask for 1 to 3 websites I like the look of. If I don't have any, send me to siteinspire.com to pick some. Sites built from a reference look far better than sites built from nothing.
3. Write the copy before any design and show it to me: headline, subheadline, every section, button text. Short sentences. Get my okay.
4. Build one page (a single index.html file, plus an images folder if needed): hero with the main button, what I do, proof, about me, a final call to action, footer.
5. Show me how to preview it: open the file in my browser, or view the artifact.
6. Run 2 feedback rounds. Ask what feels off, fix it, and suggest improvements I wouldn't think of.
7. Put it online with the Netlify steps above.
8. Optional: a custom domain. Tell me where to buy one (usually about $12 a year) and walk me through connecting it, one setting at a time.
9. Done test: I open the live link on my phone and tap the main button, and it does what it should.

▸ IF IT'S A PERSONAL APP
1. Ask me to describe the screen I'd open every day. What do I see? What do I tap? Describe the layout back to me in words and get my okay.
2. Build it as real files (one index.html is ideal), not an artifact, so it can save my data. No accounts, no servers. Give it a proper name and a simple app icon.
3. Add Export and Import backup buttons, because saved data can get wiped. Tell me to export a backup once a week.
4. Load it with realistic sample data so I can see it working, plus a button to clear the sample data.
5. Put it online with the Netlify steps above. Open the link on my phone and add it to my Home Screen (iPhone: Safari, Share button, Add to Home Screen. Android: Chrome menu, Add to Home screen or Install app).
6. Tell me clearly: only enter real data in the Home Screen app. Data typed on my computer, in a regular browser tab, or at a different link won't show up there.
7. Done test: from the Home Screen icon, I use it for real once, fully close it, reopen it, and my data is still there.

▸ IF IT'S A WEB APP
Be straight with me, and keep it exciting: a real web app with logins and shared data is a multi-day project. Today's win is a clickable prototype that looks and works like the real thing, using fake data. That's how real products start.
1. Map the users: who uses it, what each person sees, what each person can do. Keep it to 1 or 2 types of users.
2. Map the 3 to 5 screens that matter most. Get my okay.
3. Build the clickable prototype with realistic fake data. Every button should do something, even if nothing is saved yet.
4. Show it to me and run 2 feedback rounds.
5. Put the prototype online with the Netlify steps above so I can send the link to 1 to 3 real people who would use it and ask what they'd change.
6. Add a "Phase 2" plan to BUILD-BRIEF.md: moving the project into Claude Code, adding a real database and logins (Supabase is the usual beginner-friendly pick), handling payments or sensitive data safely if needed, and putting it online properly (Vercel). Explain each in one sentence and what it costs to start (usually $0).
7. Done test: someone other than me clicks through the prototype and gets it without me explaining.

▸ IF IT'S A CLAUDE SKILL
1. Walk me through the last time I did this task, step by step, like you're shadowing me. What did I open, what did I look at, what did I decide, what did I produce?
2. Get 2 or 3 real examples of the finished result I was happy with. Look in my connected tools first, with my permission. These become the quality bar.
3. Ask what bad looks like: the mistakes, the tone I hate, the things I always end up fixing.
4. Write the skill. Use the built-in skill-creator if it's available. The skill needs: a clear name, a description that says exactly when to use it (with the phrases I'd actually say), the step-by-step process, what it needs from me each time, the output format, my examples, and a checklist it runs before handing work back. It must never invent facts or numbers. It uses what I give it or what it finds in my tools.
5. Test it on a real case right now. Show me the output next to my example. Fix what's off, then run it again.
6. Install it. Save the skill file in my Day 7 Build folder first. Then ask for a screenshot of my Claude settings and walk me to where skills get added (usually a Skills or Capabilities section with an upload option). If I can't find skills at all, check whether code execution is turned off or my workspace admin has disabled it. Then have me open a fresh chat and trigger it by name to prove it works.
7. If it touches my email, calendar, or other tools, it drafts and never sends until I've trusted it for a couple of weeks.
8. If it should run on its own and doesn't need input from me each time, set it up as a scheduled task and tell me exactly when it runs. Warn me that scheduled tasks usually only run while my computer is awake and the Claude app is open. If it needs my input every time, skip the schedule and tell me what should remind me to run it.
9. Done test: I trigger it from a fresh chat and get a result I'd actually use with less than 2 minutes of edits.
10. Tell me the skill file is mine. If it's for work, I can upload the same file to a work-approved Claude account later.

━━━ PHASE 6: SHIP IT AND LINE UP WHAT'S NEXT ━━━

When the done test passes:
1. Tell me what I built today in 2 sentences I could text to a friend.
2. Give me the 3 best upgrades for version 2, ranked with the same scoring.
3. Update BUILD-BRIEF.md with what we built, where it lives, and how to use it.
4. Save the rest of the scoreboard, plus the "later" list, as NEXT-BUILDS.md. The #2 pick is next week's build.
5. Remind me of the rule: anything I do more than once should become a skill. The next time I catch myself re-explaining something to Claude, that's my next build.

Start now with Phase 0. In 2 or 3 sentences, tell me what we're about to do and roughly how long it takes, then begin.