DEV Community: Bobi

Two questions I added after readers replied

Bobi — Thu, 05 Nov 2026 06:55:23 +0000

Last week I shared the one-page test I use to shrink an app idea before building it with AI. Three readers on Indie Hackers replied with points I hadn't thought of. They were good enough that I changed my guide. Here's what changed and why.

1. Who pays, and how much?

The test had five questions: who is it for, what problem do they have, what is the one thing they do in your app, what do they see afterwards, and how will you know it works.

simonzsy pointed out that an app can pass all five and still sell nothing. So there's now a sixth question:

Who pays, and how much? A guess is enough, like "freelancers, $5 a month".

You don't need a pricing page. You need one honest sentence. If you can't name who would pay, better to find out before the weekend than after.

2. Which parts did you reuse?

When the AI finishes a change, I ask it to explain what it did in three plain bullets, so I understand my own app without reading code. hyunjoong suggested adding one more line, and I now use it every time:

Explain what you changed in three plain bullets, and tell me which existing parts you reused. No code.

Why it helps: AI tools sometimes build a fresh copy of something your app already has, like a second save button or a second place to store users. Asking what it reused shows you when it started from scratch. If the answer is "nothing", ask why.

And one exception

My "leave out of version one" list says no notifications or emails. usewello replied that some apps are pointless without them. A salon booking app that never reminds anyone about their appointment isn't much of a booking app.

Fair. The rule now reads: leave out notifications and emails, unless the app is pointless without them. If the reminder is the product, it stays. Everything else on the list still waits.

The updated test

Who is it for?
What problem do they have?
What is the one thing they do in your app?
What do they see afterwards?
How will you know it works?
Who pays, and how much?

If an answer needs more than two lines, the idea is still too big.

Thanks to simonzsy, hyunjoong and usewello for taking the time to reply.

This is from Live by Sunday, a 42-page guide for people who don't code: build your first app with AI in one weekend and put it live on your own domain by Sunday night. 30 copy-paste prompts are included. More at livebysunday.app.

Claude Code for beginners: build your first app without touching a terminal

Bobi — Thu, 29 Oct 2026 08:15:13 +0000

If you've never written code, Claude Code can look like a tool for developers. It isn't only that. You describe what you want in plain words, it writes the code, and you decide what goes live. Here's how to start without installing anything or opening a terminal.

What Claude Code is

Claude Code is the part of Claude that works on a real project. A normal chat gives you code to copy and paste. Claude Code opens your project, makes the change itself, and saves it as a proposal you can accept or throw away.

You can use it in your browser at claude.ai/code or in the Claude desktop app. It's included in Claude Pro ($20 a month, or $17 a month paid yearly).

What you need

A Claude Pro account. That's the only paid part to start.
A free GitHub account. GitHub keeps every version of your code, so nothing gets lost and any mistake can be undone.
Later, a free Vercel account to put your app on the internet.

Your first 20 minutes

On GitHub, create a new repository (GitHub's word for a project folder). Make it private and tick "Add a README file".
Open claude.ai/code, connect GitHub when it asks, and give it access to that one project only.
Ask it to create a file called CLAUDE.md (more on that below).
Ask for something small, like a simple landing page with your app's name.

When it's done, Claude Code saves its work as a pull request: a proposed change waiting for your OK. You read what it did, then press Merge. That's the whole loop.

The one file that makes it work

CLAUDE.md is a short note about your project. Claude Code reads it at the start of every session, so you never explain your app twice. Mine has three parts:

One sentence about the app: "For [who], who struggle with [problem], [app] lets them [one action], so they [result]."
What version one includes: a short list, nothing more.
How to work with me: "I don't code. Explain things in plain words. Plan first and wait for my yes. One change at a time. Never put passwords or secret keys in the code."

That last part matters most. It turns the AI from something that guesses into something that checks with you first.

Four habits that save you hours

Ask for a plan first. End every request with "Plan first and wait for my yes." Read the plan. If it lists five things and you asked for one, say no.
One change at a time. "Add sign-in" is one request. "Add sign-in, a dashboard and payments" is three, and when something breaks you won't know which one did it.
Check before you merge. If your project is connected to Vercel, every pull request gets its own preview link. Open it on your phone and click through it before you press Merge.
Report bugs in four parts. What I did. What I expected. What happened. The exact error or a screenshot. "It doesn't work" gets you a guess. Those four lines get you a fix.

Mistakes almost every beginner makes

Asking for the whole app in one message. You get a big pile of code that half works. Small steps are faster.
Merging without reading. You don't need to understand the code. You do need to read the plain-words summary of what changed.
Letting it add extras. AI likes to be helpful. "Keep it simple, no features I didn't ask for" belongs in CLAUDE.md.
Pasting secret keys into the chat or the code. Keys go into your hosting settings (Vercel calls them environment variables). Ask Claude Code to show you where.

What it costs

While it's a personal project: Claude Pro, plus about $10 to $15 a year if you want your own domain. GitHub is free, and Vercel's Hobby plan is free for personal, non-commercial projects. Once your app takes money, Vercel needs its Pro plan ($20 a month).

Where to go from here

Pick an idea small enough to finish in a weekend, write the one sentence, and ask for the first page. The first merged pull request is the moment it stops feeling like a developer tool.

Live by Sunday is my 42-page guide for people who don't code: build your first app with Claude Code in one weekend and put it live on your own domain by Sunday night, with 30 copy-paste prompts. More at livebysunday.app.

The one-page test that shrinks your app idea to something you can finish this weekend

Bobi — Wed, 14 Oct 2026 07:15:15 +0000

Most first apps never go live. Not because the AI gets stuck, but because the idea is too big.

If you're building your first app with an AI coding assistant, the best thing you can do before opening it is answer five questions on one page. If an answer needs more than two lines, the idea is still too big.

The five questions

Who is it for? One kind of person, like "freelance designers", not "everyone".
What problem do they have? Say it the way they would say it.
What is the one thing they do in your app? One action, like "log a client call".
What do they see afterwards? The result that makes them come back.
How will you know it works? For example: one real person uses it twice in a week.

Then fill in this sentence:

For [who], who struggle with [problem], [app name] lets them [one action], so they [result].

That sentence becomes the first line of your brief to the AI. A tool that knows exactly what you're building makes far fewer wrong guesses.

Cut it down

Big idea	Version one
A platform for personal trainers	Trainers send each client a weekly workout plan as a link
A CRM for freelancers	Log each client and the last time you spoke; show who's overdue
A recipe app	Save recipes from a link and tag them
A team habit tracker	One person ticks one habit a day and sees their streak

Leave these out of version one

Payments
Admin dashboards and settings pages
Teams, sharing and invitations
Notifications and emails
A phone app from the app stores
Several languages

Build a web app, not a phone app. It works in any browser, needs no app store approval, and you can change it in minutes. If people love it, a phone app can come later.

One more thing

My first product had a seven-phase plan. I launched after phase four and moved the rest to a list called "Later". Launching taught me more than building those three phases would have.

If it doesn't fit in a weekend, it's not version one.

This is chapter 1 of Live by Sunday, a 42-page guide for people who don't code: build your first app with AI in one weekend and put it live on your own domain by Sunday night. 30 copy-paste prompts are included. More at livebysunday.app.