Part 2: How a Tutor with Zero Coding Experience Built a SaaS with AI — The Great ChatGPT Struggle

Previously

In Part 1: The Origin Story, I wrote about my six years as a private tutor, struggling with manual attendance tracking.

Then I saw a social media post about someone who had an AI write code and built an app — and I decided to try it myself.

That's when the nightmare began.

Meeting ChatGPT — A Ray of Hope

At the time, the most well-known AI in the world was ChatGPT.

When I heard "an AI that writes code for you," it was the first thing that came to mind. I created an account right away and asked:

I copied the code it gave me and pasted it into Google's Script Editor, just as instructed.

It worked.

In that moment, I genuinely believed, "Maybe I can actually do this."

That was the trap.

Nightmare #1: The Endless Copy-Paste Deploy Cycle

ChatGPT can write code for you. But that code lives inside the chat window.

Here's what that means in practice:

I did this dozens of times a day.

One file would have been manageable. But as features grew, so did the files. Three, four, five… Each one had to be copied from the chat, pasted into the correct file, saved, and executed — every single time.

I was trying to "automate with AI," yet the process of working with the AI itself was completely manual. The despair I felt when I realized this contradiction is something I'll never forget.

Nightmare #2: Fixes That Break Everything

This is where the real nightmare began.

Bugs appeared. Of course they did. This was code written by an AI at the direction of someone with zero programming experience. Bugs were inevitable.

I'd send ChatGPT the error message. It would send back a fix. I'd paste it in.

Something else would break.

This went on forever. Fix A and B breaks. Fix B and A comes back. Whac-A-Mole. Whac-A-Mole with no exit.

The reason was simple: ChatGPT couldn't see the entire file. It would guess based on the error message I pasted and the recent conversation, then generate code from that. But without understanding the full codebase, its fixes were often off the mark.

Nightmare #3: The AI That Forgets

And then came the ultimate despair.

ChatGPT has a conversation length limit. As conversations drag on, it loses the earlier context.

Here's what that looks like in practice:

This didn't happen just once. It happened over and over.

The longer the conversation, the more it forgot: specs we'd agreed on early in the session, function names it got wrong, variables that didn't exist. It would casually suggest code that contradicted what it had written just minutes ago.

A person with no programming experience had to judge whether the AI's output was correct — while the AI itself couldn't even remember its own prior context.

It was nothing short of hell.

Almost Gave Up

I kept at this for weeks.

Day after day: copy-paste, error, fix, copy-paste, different error, fix, original error returns, copy-paste, context lost, re-explain everything, copy-paste.

I seriously considered giving up.

But six years of frustration doesn't just disappear. The end of the month keeps coming. The manual work never ends.

That's when I discovered something called Claude Code.

Next Time

Part 3: Discovering Claude Code — Not "AI writes code" but "AI writes code directly into your files." That distinction changed everything.