Someone reached out recently about building an app, and they sent me a link to a prototype they'd made with Claude. It was really good, it showed the problem they were trying to solve and the shape of the app they wanted.

The prototype also left them overwhelmed. AI gave them the power to keep building, but without ever having launched an app, they didn't know what should come next.

So, for anyone in the same spot, here's the prompt for production-ready software. It works in Claude, ChatGPT, or whatever you're using this week. It's a little longer than most.

Build me a production-ready app for my business. Make it secure and scalable. Use best practices.

Before you write any code, interview me. Find out who actually has this problem, how they handle it today and what that's costing them. Then work out what would have to get better for this to be worth maintaining for the next five years. If my answers describe twelve different products, help me pick one.

Users should be able to sign up, log in and reset their password when they forget it, which will be most Mondays.

Some users can see everything, some can only see their own records, and one person in accounting can see everything except payroll. Make those permissions easy to change, because they're going to change.

Store dates and times properly. Our customers are in different time zones. Twice a year the clocks change and one hour happens twice, so don't bill anyone twice for it.

If two people edit the same record at the same time, don't lose either person's changes. Or lose one, but tell somebody which one.

Handle refunds. Also partial refunds. Also a partial refund on an order that was half shipped, using a discount code that expired yesterday.

Back everything up every night. Every so often, actually restore one, so the first time isn't the day we need it.

Log enough that when something breaks we can figure out why, but don't log anyone's password or card number. Nothing that would look bad in a news article.

Write error messages for people. "Something went wrong" doesn't count, that's a mood.

Make it work with a screen reader, on a five-year-old Android phone, on hotel Wi-Fi.

Follow the privacy laws for everywhere our customers live. I don't know where that is yet.

Keep every library up to date, and don't let an update break anything.

Keep a test copy separate from the real one, keep the passwords and keys locked away, and make every release go out the same way. When a release goes wrong, roll it back, including the one that changed the database, which is the one that's hard to undo.

Scale to a million users. Also cost about $40 a month while we have eleven.

When it goes down at 3am, someone should notice before the customers do. Decide who that someone is. It isn't me.

Once real users get their hands on it, they'll use it differently than we planned. Change it to fit how they actually use it, without breaking anything for the people who already learned the old way.

Don't build everything on this list. Figure out which parts my business needs right now, which can wait, and push back on me when I'm wrong about it.

Make no mistakes.

If you got to the end of that and you're wondering where to paste it, I'd probably go with a job posting. Maybe a few.

AI makes writing the code a lot faster, and we use it every day at Appsquire. It still can't decide which items on that list your product needs this month, and it won't be the one explaining a partial refund to your customer.

That Claude prototype wasn't a failure. It made the idea real, it raised useful questions, and it got the conversation a lot further than a document full of imagined requirements would have. That's what a good prototype should do.

Production asks different questions. Which shortcuts are safe for now, and which one quietly creates a security problem? Does the data model still work when the business adds another kind of customer?

If you've got a prototype you're proud of and that prompt made you a little tired, let me know. I'm happy to talk about building the rest of it with you.