What it's like to build with AI

A friend of mine said recently: "I have no idea what you're writing about. Supabase, stack overflow, all those terms, no clue what they mean."

He has a point. Time for a story without jargon.

What I'm doing

I'm building an automated trading system. A program that looks at the stock market every day, decides what to buy or sell, and does it automatically. For myself, with my own money, to learn.

The most important detail: I don't write any code myself. Not a single line. I can make an Excel sheet, that's my programming level.

Still, there's a working system ready, going live in six weeks. How? By collaborating with AI.

The work

Imagine hiring a very young architect. He's brilliant, knows every building technique, can design a house in an hour. But he has no idea what you want. No idea where the house should go. No idea who it's for.

That's my role. I describe what I want to build. The AI builds it. I check if it's right.

My day consists of three types of conversations:

With the strategist (an AI that helps me think about architecture). "If I want this, how do I approach it? What are the risks? What might I miss?"

With the builder (a different AI that actually writes code). "Build this module now. Test if it works. Save the version."

With myself (the hardest one). "Does this match what I meant? Do I trust the AI?"

An example without tech speak

Today I wanted to add five new rules to my trading system. Rules that prevent the system from making the same mistakes repeatedly.

I described those rules in a few sentences. The AI built them. Then we tested it on four years of historical market data.

The result was too slow. The test that normally takes ninety minutes would now take six hours. Unworkable.

Three times I thought I knew where the problem was. Three times I was wrong. Until I finally asked the AI to first measure where the time was going. Only then did we see the real problem.

The fix? Create a kind of cheat sheet. Instead of recalculating everything every day, save the answer once and just read it back the rest of the year. Sounds simple. Was the breakthrough of my day.

What's different than I thought

I thought: AI does the work, I give instructions. If only it were that simple.

What it really is: AI accelerates your decisions, multiplies your options, and amplifies your mistakes. A bad decision gets executed lightning fast. You learn from that.

Three things I didn't expect:

1. Discipline matters more than tempo

Today I was in flow. Pulse 100. Wanted to keep going. The AI happily followed. But somewhere along the way the scope changed, and the AI didn't automatically flag it. I should have been warned beforehand that our new plans contained risks the old plans didn't.

Lesson: AI is not your conscience. Your discipline is your discipline.

2. Documentation is mandatory

I thought: "We'll build, we'll do documentation later." Don't do this. Two days later you don't remember why you made certain choices. Neither does the AI, because its memory is limited to the conversation.

Since today I write first why I'm going to do something. Then I build. That sounds illogical but works.

3. You learn faster than you think

I can't code, but I do understand what a database is. What a table is. How data flows from A to B. Not because I studied this, but because I talk to AI about these concepts every day.

It's like moving to Italy without speaking Italian. You don't learn it by studying, you learn it by living.

Why this works for me, not for my friend

My friend reads my blogs and tunes out at the technical bits. Understandable. He wants to know what it produces, not how it works.

I write for myself, for the learning curve, for transparency. For other entrepreneurs who want to try this too.

What do you need to build with AI?

• Clear thinking. If you don't know what you want, neither does the AI.
• Persistence. When something doesn't work, don't get angry, just find the cause.
• Willingness to start dumb. You know nothing, that's OK. Get going.
• Money to break things. I spend about €100 per month on AI tools, plus extra investment in better database capacity. Not expensive for what it produces.
• Time. An hour a day, sometimes a whole day in a sprint. Not a few minutes per week.

What this is not

It's not magic. I sit debugging for a whole day when something doesn't work. Cursor (the AI that writes code) makes mistakes. Sometimes the same mistake systematically over and over.

It's not free. AI tools, database storage, server time, together it costs me about €100 a month. Much less than a freelancer, but it's money.

It's not safe. I'm trading with real money, my own capital. That's a choice, and it ensures I don't get sloppy with code.

It's not for everyone. You need entrepreneurial spirit. The willingness to push through skepticism, despite the complexity of what you're building. Not everyone fits that.

For my friend

So what am I doing, in one sentence?

I'm building an automated investment system. No human has to choose stocks. The program does it itself. With my own money, to find out if it works.

And how?

I sit at home at the table with my laptop. I talk to AI in text. AI builds the code. I check if it's right and money doesn't leak out. It's not wizardry, it's just work. But different work than was possible ten years ago.

The system isn't entirely finished yet. June 22nd it goes live. Maybe it will work. Maybe it won't.

For me that's the beauty of the experiment. Building something new, without guarantees, with technology that's still new. Someone else would have spent a hundred thousand euros on this. For me it's an evening per week, sometimes a whole day, and it costs less than a good ski weekend.

That's how you build with AI. That's what I'm doing.

playitsmart.nl is an experiment where I publicly build an automated trading system with AI as builder. Live launch with €10,000 of own capital: June 22, 2026.