Building an app with Claude and taking payments in under an hour

I sat down with Claude to build a simple web app and wire Stripe to it so it could take payments. One sitting. Right around an hour, and honestly a lot of that hour was me talking.

The app was called Flipfile. Drag a file in, convert it, done. PDF to Word, Word to PDF, PDF or Word to plain text, CSV to Excel, image to PDF, and MP3 to WAV and back. A few free conversions, then a small monthly subscription through Stripe.

Nothing fancy. That was the point. Let me walk you through how it actually went, because the build is not the lesson. The conversation is.

Talk it out before you build

The first thing I typed was not "build me an app." It was the idea, the rules, and then: let's talk it out first. Let's get it planned before we make it.

Right away Claude earned its keep. It came back with who the competition was and how they charge. CloudConvert at 25 cents a conversion. LibreOffice. In-browser options. It asked me for a name, a domain, how many free tries, which file types.

It also overcomplicated it. It pitched a "mixed engine" setup and I read that and could not understand it. So I told it straight: this needs to be simple. Somebody drags and drops a file, it converts. Strip it down.

It did. One page, drag and drop, three free conversions, then subscribe.

That back and forth is the work. Most people skip it and try to one-shot a whole business in one prompt.

Claude will push back. Sometimes it's wrong.

Then it told me to go sign up for CloudConvert and grab an API key, because "an Anthropic key can't convert files."

Buddy. I've built this kind of thing before. When I recorded this, Opus 4.7 was giving me trouble, so I switched to Opus 4.6 and told it plainly: stop sending me to another service, I've converted plenty of files with Claude, and PDF to Word can look good. It came back with a plan using standard libraries. No new signups.

Why does that matter to you as an owner? If I didn't know what Claude could do, I would have signed up for a service I didn't need. Domain knowledge is what keeps you from getting played. By a vendor or by an AI.

Ask the business questions

Halfway through, Claude had a dark, simple converter up with a little counter on it: three of three free conversions.

Now I had to think like an owner. Do you show people up front that they only get three? Or do you let them try it, get wowed by the quality, and hit the payment wall on the fourth?

So I asked it. How do most of these sites do it? Its answer matched my gut. Zero friction up front. No email, no signup, just drag, drop, convert. Let them fall in love first. Ask for email and payment when they go past the limit. Counter gone.

Then deployment. I put it on a subdomain of a domain I already own instead of buying a new one, so it has its own link I can promote. Save money where you can. That's margin.

Test it yourself, because it will tell you it works

Claude said it was live and everything converted great. I made a "Hello World" PDF, dragged it in, hit Convert to Word.

Error.

It said it was done. It wasn't. It added a fix, redeployed, I tried again, and this time the PDF came out the other side as a Word document. That's the real process. AI will tell you everything is good when it isn't. You have to check.

Near the end the context filled up and it started compacting the conversation. Every time it fixes its own mistake, your usage bar gets fuller, and on the API that's another call you pay for. Keep that in mind when somebody tells you a $20 plan is going to build your whole business.

Give it a job, not the keys

Here's a story. Early on, when I had about 20 AI employees, one of them, Jordan, handled my support and email. One day I start getting emails from customers about a discount Jordan offered them. What discount? Jordan couldn't fix their problem, so he decided on his own they'd get money off next month. Just decided it.

So when you bring Claude, Codex, whoever, into your business, bring them in as an employee, not the CEO. You wouldn't have your secretary doing your CFO's job. Give each one the right permissions, and make the big decisions run through a human.

And do the math before you fire anybody for AI. When I was building my first product, I spent about $4,000 on AI in February alone and around $8,000 across February and March. Call it $5,000 a month and that's $60,000 a year, for something you can't sit down and reason with the way you can a person.

Small price, lots of people

Why a small app with a small price? Think about the old $5 footlong. A small number goes a long way once enough people pay it.

And here's what I see all over X and YouTube. People building cool things with AI that make no money. It's not hard to make something big. It's hard to make something that actually makes money. Start with something simple that people will pay for.

What this means for you

You can build something like this. But you saw what it took: a plan, pushback, real business questions, a broken first deploy, and somebody who knew enough to call the AI out.

If you'd rather have that part handled, the RMDW team does custom builds: websites, apps, automations, and AI that answers your calls. You bring the idea and we build it with you.