Building your own applications with AI
What is left when AI writes the code, and what that makes sellable as marketing work: a small application on your clients’ data.
01
The moment
The build began with a single sentence on an empty page: what a customer gets and what they pay for. Only once that sentence stood was the AI allowed to write the first line of code. I wrote down the build of an application for customers step by step, as a build log and not a manifesto, so that someone with a clear idea can walk the same path.
Yes, an AI wrote most of the code. That is now the normal way to build, and I will not dwell on it. Something else counts for you. Until recently an application of your own was, for someone in marketing, a project with an agency, a specification and a budget you do not have. Today the question is no longer whether you can code. The question is whether you can say clearly what should be built, and whether you notice when something was built wrong.
Which brings us to your offer. The obvious application for someone with your background is not an app for consumers but a small tool on a business’s marketing data: a report the client opens themselves, a form that captures enquiries cleanly, a view that shows which campaign brought which enquiry.
02
What went wrong
What goes wrong goes wrong quietly. The AI will happily build you a cathedral. You want a kiosk that takes coins. Whoever starts without a clear sentence gets a lot of software nobody needs in very little time. It feels productive, because something new appears all the time. In reality it is the most expensive form of procrastination.
The second quiet mistake is ordering a whole system at once. Then you get a whole system you do not understand, and at the first error you do not know where to look. And one of the pieces will be wrong.
The third is about security. In my own build I paid most attention to one thing: that no key is left where it should not be, such as a password in the code, an access that may do more than needed, or a check that only happens in the browser and not on the server. Whoever builds fast leaves keys like that lying around unless they pay attention on purpose.
The AI will happily build you a cathedral. You want a kiosk that takes coins.
03
What I tried
This is how I built it, in order. First the one sentence, then breaking it into the smallest pieces that can each be built and checked on their own: the account, the core action, the payment, the confirmation. For each piece I asked the AI what risk was in it and made it justify why something had to be visible from outside. If the justification was thin, it stayed internal.
The first feature was the account, with sign-in without a password: type your email, get a link, click. The links go out through Mailjet so they arrive reliably. The whole thing is hosted on Google Cloud, as a container on Cloud Run that sleeps when nobody needs it and costs almost nothing during that time. What that comes to per month, I will add. Next to it stands a test copy that mirrors the real environment. The interface came last and was designed in Figma before a line of it was built. The whole way is on the experiment page.
04
What I know now
What I know now: code has become cheap, decisions have not. What first? What not at all? Who may do what? What happens when something fails? An application succeeds or fails on these questions. The AI answers them gladly, but for an average case, and your client’s business is not one.
Order matters more than technology. First the sentence, then the pieces, then security, then running it, and at the very end the look. Many marketers do the exact opposite with their first product of their own: they start with the landing page, because they can do that, and not with the sentence, because it hurts.
On security you need no department of your own, only four habits. Do not store passwords if you can avoid it. Keep secrets out of the code and in the platform’s secret store. Check every real action on the server. Give every piece only the access it needs.
For your exit this means: a small application that solves exactly one problem on the marketing data of one specific group is within reach today. It only has to be better than the spreadsheet or form that group uses today. You probably know a few such spreadsheets from your job. But you will not sell it as an app. You sell it as part of a service: the client pays for their numbers being right and visible every month, and the application is the window through which they see them.
And because everything goes so fast, the temptation to build too much is big.
05
Open question
What is open is maintenance. Building an application is one thing, keeping it running for years while libraries, platforms and models change is another. How much effort that really is with AI in the long run, I do not know yet. I will add here what I see.
06
Changelog
First version. Nothing here is rewritten silently: every later change gets its own entry in this changelog.