Vibe coding with Gemini: helpful, tricky, and why you still need a human.

Ever been handed a document that’s technically correct but practically useless? That was me last week when my daughter’s school sent out the new class timetable. It was a cluttered, ugly PDF that was difficult to read at a glance.

Instead of firing up a code editor, I tried something new: vibe coding. The idea is simple — you describe what you want in plain English, and the AI builds it. I started in ChatGPT, but it floundered at the very first step. So I switched to Gemini.

My request was straightforward: create a tool where parents could upload the ugly PDF and get back a clean, colour-coded timetable. Something you could print, stick on the fridge, and actually use.

And you know what? It worked. Kind of.

The Magic and the Mess

The good news: the final tool delivered exactly what I wanted. Parents could upload their timetable, get back a colour-coded A4 version, and download it in seconds. I shared the link in the parents’ WhatsApp group, and the feedback was brilliant. Lots of parents used it, and many thanked me for making something genuinely useful.

But the road to get there wasn’t smooth. Gemini felt less like a polished assistant and more like a super-intelligent intern who had zero common sense.

The initial idea was to go from this cluttered official timetable: (The images are slightly redacted to hide PII - teachers’ names)

To this clean, simple, and colour-coded version that a human would actually want to use:

The AI Hits a Brick Wall

Gemini started confidently but quickly got stuck. It would generate code that created errors, apologise profusely, and then generate different code that created new errors. We were going in circles. The tool couldn't "see" the timetable structure in the text it extracted from the PDF. It saw a jumbled mess of words and couldn't figure out that a lesson at a specific time was for a specific day.

It hit a dead end, convinced the problem was too complex to solve. I was about to give up, but then, looking at the raw text it had extracted, I spotted something the AI had completely missed. It wasn't random at all; there was a simple, sequential pattern. For any given time slot, the first lesson was always for Monday, the second for Tuesday, and so on.

It was a simple observation that a human would make, but the AI was blind to it. When I explained this, the tone of the conversation shifted instantly.

After this breakthrough, we had a working tool within minutes. But it's a critical point: without that human insight, the project was dead in the water.

Why this matters to you

If you're a business owner looking to use AI to build small tools to boost your team's productivity, this experience holds a vital lesson. AI is fantastic for getting you 90% of the way there. It can build the scaffolding of an application incredibly quickly.

But that last 10% can be a minefield. When the AI gets stuck, it can't always debug itself. It can't spot the simple, context-based patterns that are obvious to you or your team members. If you don't have any technical experience, you'll hit the same wall I did, but you'll have no way of pointing the AI in the right direction.

For now, AI is a powerful assistant, not an autonomous creator. It needs guidance, expertise, and a human hand on the tiller to navigate the inevitable problems.

Vibe coding is helpful, but it's tricky. It can get you started, but it might not get you to the finish line.

Don't Get Stuck - Get Expert Help

If you’re experimenting with AI in your business — maybe building internal dashboards, automating reports, or creating custom tools — the risk is the same. You’ll make progress, then stall when the AI’s code breaks.

With Kimbley IT, you don’t get left at 90%. You get:

• AI speed, with human reliability — fast prototypes that are refined into dependable tools.

• Debugging you can trust — we spot and fix the blind spots that AI misses.

• Tools tailored to your team — so they actually work in your business, day after day.

That’s how you move from “clever demo” to “tool you rely on.”

Vibe coding is helpful, but it’s tricky. If you’re exploring how to use AI to build tools for your business, don’t risk stalling at 90%. Book a video call today and let’s turn your idea into something reliable that your team can actually use.

Frequently asked questions

What is vibe coding?
It’s the practice of describing what you want in natural language and letting an AI system generate the code. Great for quick prototypes, but fragile for production.

Why did Gemini struggle with a timetable PDF?
PDFs don’t store neat tables—they’re more like digital drawings. AI often misreads them, which leads to parsing errors and broken layouts.

Can non-coders still build with AI?
Yes, but expect to get stuck when the AI can’t debug itself. That’s when you need a human partner to step in and fix the problems.

How does Kimbley IT help?
You get the best of both worlds: AI speed and human reliability. We refine the code, handle tricky formats, and make sure the end result works every time.

Is Gemini better than ChatGPT for coding?
Each has strengths. But the constant is this: neither can guarantee working code without human input.