My First Prompt Was Way Too Ambitious

The very first prompt I gave Lovable was not small.

I asked it to build a production-ready automated AI stock trading web app focused on short-term U.S. stock opportunities, especially AI and AI-adjacent stocks. I also said it must not promise guaranteed profits and needed strong risk controls.

That is a mouthful. It was also probably the kind of prompt a real software team would break into six months of meetings.

Lovable came back with a plan. It named the platform QuantisAI. It talked about user accounts, watchlists, risk settings, roles, and a database. When it mentioned things like database security rules, I did not fully understand them, but I understood the basic idea: people should only see their own information, and the system should have guardrails.

Then something wild happened.

It built a foundation.

There was a site. There were screens. There was a scanner. There was mock market data, meaning made-up sample stock information so the app could be explored before live data was connected. It looked real enough that my brain wanted to celebrate.

But almost immediately, the other part of my brain kicked in.

I asked it to confirm that the system was absolutely fully functional with a top-rated algorithm and fully executable trading capabilities. I even said not to come back until it was.

Now, I can admit that was not my finest moment. I was excited. I was impatient. And I was treating the AI a little like a contractor I had hired to finish the whole house overnight.

But I am glad I pushed.

Because the first important lesson came fast: an AI builder can make an app look real before every part of it is real.

The scanner appeared to run, but then I noticed something was wrong. I said, "Run scan doesn't seem to be working."

Lovable found that one piece of the app was not properly connected to another piece. In plain English, the scan button was looking for a tool that had not been plugged in correctly.

That was fixable.

Then came the next punch: "Returning 0 candidates."

The app was scanning and finding nothing. Not one stock. Zero.

Lovable found the cause inside the sample data generator. Again, in plain English, the app was using a broken way to create repeatable fake prices and scores. It was like having a deck of cards that looked normal until you realized the shuffle machine was broken.

As a non-coder, this was both amazing and unsettling.

Amazing because I could describe the problem in regular language and the system could investigate it.

Unsettling because I would never have found that on my own.

That first prompt taught me something I keep coming back to: AI can build fast, but I still have to be the adult in the room. I have to test. I have to question. I have to notice when something feels off.

The first version of QuantisAI was exciting.

It was also not enough.

And that was okay, because for the first time in my life, I was not just imagining an app.

I was arguing with one into existence.