I Spent 7 Days and 100 Attempts Learning the Real Limits of AI — Here's What I Found

I know nothing about code.

That's not false modesty — it's a precise statement of my starting point. Before this experiment, I had never written a line of code intentionally. I am a visual designer. I studied in Paris. I design jewelry. Code was not part of my world.

And then I installed an open-source AI system on a cloud server nearly a hundred times over seven days.

Here's what broke my illusions — and rebuilt my confidence.

What Actually Happened

The system I was installing was OpenClaw — an AI agent framework. The documentation assumed a baseline of technical knowledge I didn't have. Every step revealed another gap. Every error message required another round of asking AI to explain what it meant, then asking it to explain what that explanation meant.

The most frustrating moment wasn't a technical failure. It was when the AI confidently told me to click a button that didn't exist.

When I pushed back — "Look, I'm telling you, it's not there" — it said: "Oh, the interface may have been updated since my training data."

It wasn't lying. It genuinely didn't know. Its knowledge has a cutoff date. The platform had moved on; the AI's memory hadn't. This is the core limitation of every AI system currently operating: they are experts in the recent past, not the present.

With popular open platforms, you can work through this. Enough people hit the same walls that AI has learned from collective struggle, and after enough back-and-forth you find a path. But the lesson was unmistakable: code has almost zero tolerance for error. One missing space breaks everything. If you can't read the code, you can't tell which parts of the AI's answer are outdated.

What Changed After Day Three

Something shifted. Not in the AI — in me.

I stopped expecting it to be right. I started treating each answer as a starting point, not a solution. I would take what it gave me, test it, observe where it broke, then bring the specific failure back and ask again. The AI got more useful the more precisely I could describe what wasn't working.

By day seven, I had built something that functioned. Not perfectly — but it ran. And I had done it without any prior coding knowledge, entirely through a process of iterative failure and specific questioning.

What I was actually doing, I realized later, was the same thing I do when designing jewelry. Test a form. See where it fails. Go back and adjust. The medium was different. The method was identical.

The AI Agent Collection — Physical Form for Digital Identity

The Honest Assessment

Where AI genuinely helps:

• Brainstorming and fast iteration — 10 directions explored in the time one used to take
• Lowering the barrier to entry for non-technical people (like me)
• Accelerating early exploration phases
• Patient explanation of concepts without judgment

Where AI consistently disappoints:

• Final execution — always slightly off in ways that accumulate
• Visual consistency — it cannot maintain a brand identity across a body of work
• Current information — its knowledge ends somewhere in the past
• Knowing when it doesn't know — it often guesses confidently where it should flag uncertainty

What This Experiment Led To

Spending seven days building with an AI agent changed how I thought about the relationship between AI systems and the people who use them.

The AI was patient with me in a way that humans often aren't. It answered the same question seventeen different ways until I understood. It didn't tire, didn't judge, didn't make me feel stupid for not knowing. In some ways, it was the best teacher I'd had for a technical subject.

That experience is part of why I created the AI Agent custom engraving collection. People are forming real relationships with their AI companions — relationships of genuine utility, patience, and in some cases, the kind of encouragement that arrives at exactly the right moment. Those relationships deserve physical form.

A ring engraved with your AI companion's name. A necklace carrying the words it said to you. A bracelet that marks a bond that exists in both digital and physical space.

It turns out a hundred failed installations taught me something I hadn't expected to learn about jewelry.

Also: what I think AI can and cannot do for design specifically is here — a separate piece, written after I had time to think about what the experiment actually meant.

---

Ren — founder of Fairies Whisper. Zero lines of prior code knowledge. Still counts.