Skip to content
Chandler Nguyen
AI6 min read

Before You Share the App You Built With AI

Nine questions to answer before a team or a client depends on something you built with AI, how careful you need to be depending on who will use it, and a prompt that makes your coding tool show you the tradeoffs it made.

I have been working with a lot of vibe coders, and many of them come from the marketing or agency side. What they build often works. However, I keep seeing the same kind of problem. AI makes tradeoffs when it builds your application, and if you don't have software engineering skills, you simply don't know which tradeoffs were made or what to ask.

I am not writing this from above. I learned to code through structured courses, and I still find mistakes like these in my own apps. In January 2024, I wrote about saving generated embeddings in a CSV file as a makeshift database. I knew a CSV wasn't optimal for a database, but I lacked the skills for better alternatives.

So would I tell you to stop and take a course on frontend, backend and databases first? I would say NO. Those words only matter because of the decisions behind them, so I would start with the questions below and learn each concept when your own app needs it.

First, should this be built at all?

When you build a tool, you own its security, maintenance, compliance and updates. My approach is to build when the capability is core to your competitive advantage, and to buy when it is commodity infrastructure. For a task only you do, a spreadsheet and a good prompt may be enough.

How careful you need to be depends on who will use it

Who will use itWhat can be fineWhat you need
Just you, with no sensitive dataKeeping data in a simple fileA way to undo changes, and keys kept out of anything you share
Your teamVery little shortcutting, because people now depend on itA real database, a login, backups and a separate copy to test on
Clients or the publicNo shortcutsAll nine answers below, a second-account permission test, a check with Legal on where the data goes, and one named person watching it

So a simple file is not always wrong. It is a tradeoff that stops making sense once other people depend on the data.

I would also roll out in that order. Start with your team, prove the answers are accurate, and only then let clients in. One wrong answer in the first week of a client-facing tool can cost you months of credibility.

The nine questions

1. Where does the data live, and what happens if two people save at the same time? With a simple file, one person's save can overwrite another person's work. Handling that well is a big part of what a database is for.

2. What runs in the browser, and what runs on a server? That is the difference between frontend and backend, and it matters because anyone using your app can inspect what reaches their browser. An API key there is not a secret anymore, so it belongs on the server.

3. What can someone do with only your public key and no account, and what can the AI inside it do on its own? Having a login is not the same as having the right permissions. So test it on a local copy: log in as a second user and try to open the first user's data. If your app has an AI inside it, ask the same question about the AI. Which tools can it call, and which of them change something? Whatever it can do on its own, a stranger who can talk to it may be able to ask it to do.

4. How do you undo a bad change? Learn enough Git to make small commits and roll back cleanly when you break something. Git looks after your code but not your data, so you need backups too. If a real key was ever committed, revoke it, because deleting the line is not enough.

5. Where can you try a change safely? You want a copy that isn't your live app, such as a local version on your computer or a preview, with its own data and its own settings. If you are testing on the same data your users rely on, one mistake reaches them straight away.

6. How do you know it really works? The AI saying "done" does not tell you. In February 2026 I wrote about a time when Claude built what the design doc said, and the production database disagreed. I opened a podcast that the production backend had generated, and the app crashed. Then there is STRAŦUM, a multi-tenant app where I wrote 83 RLS policies and tested with both SME and Agency accounts, i.e. the rules for which account can see which rows. I had still assumed org_id filtering was enough for isolation. It wasn't, and I rebuilt it. So use your app the way a real person would, with real data, and test each step on its own so you know where a failure came from.

7. Who is watching it, and what does it cost? Software goes stale after you ship it, and usually nothing fails when it does. Run npm audit --omit=dev once today, then make that check automatic. Look at your AI bill too. In August 2025 I found that a temperature set to 0.7 instead of 0 caused 30% failed API calls and cost me $54 a month in wasted tokens. Last but not least, decide whether you will keep each app. With STRAŦUM I retired it and published v1 as a reference implementation, because the dishonest version is an app that still takes signups while nobody is watching it. If nobody is going to watch yours, take it down or say so.

8. Are the numbers and claims it shows right, and can you trace each one to its source? AI will give you a confident number whether or not it is right. In the GreenLeaf case on this site (a fictional company with synthetic data, recorded in October 2026), the AI's draft report put weekly media spend at $110,100. The checked figure for week 1 was $5,462. Pick the few numbers your app shows that people will act on, and trace each one back to where it came from.

9. Whose data is in it, where does it go, and are you allowed to put it there? If you are feeding client data into AI tools, find out where it goes, because consumer and enterprise plans can have very different data retention policies, and personal information should be anonymized first. Check with Legal before your app holds anyone else's data.

What to ask your AI

You don't need to know the answers yet, because you can ask your coding tool to show you. Before you change anything, try this:

Before changing anything, tell me where this app stores data, what runs in the browser, and who can access what. List the three biggest tradeoffs you have made, what would break first with five users, and what I need to decide. Don't print any secret values.

Please treat the answer as a starting point rather than proof. If it says your data is protected, log in as that second user and check.

When to bring in an engineer

If clients or the public will use your app and you can't answer questions 1 to 3, I would stop and get an engineer to look at it. I would do the same if it handles payments or personal data. Knowing when to stop is part of the skill.

This list is not a security audit, and it will not catch everything. I have done security reviews on my own products that still missed things. What it should do is show you the decisions you have been making without knowing it.

If you have built something with AI, which of these questions could you not answer? I would really like to know, especially if I missed one that has already burned you.

If someone else at work will use what you built, I also made Prova, a free checklist for agency staff getting an AI tool ready for their team. It is shorter and less technical than this list: five checks, each with a message to send or a question for your AI, and a one-page summary for your manager. You don't need an account.