I had decided to move everything onto our own server.

The hospital ran a small NAS — a box on the network that stores files and can run little programs. My plan was to lift the tools I’d been hacking together and put them there, where coworkers could actually use them. Good plan. Moving them turned out to be a slog.

Around then I found Cursor. It’s a code editor with an AI built into it. Before, I’d ask Gemini a question, it would hand me some code, and I’d copy that code by hand, paste it into the right place, and try to fix it. With Cursor, I connected the AI right into the editor and just told it what I wanted. Build this. Fix that. It edited the code itself. People call this “vibe-coding” — making software by talking to it.

It sounds like a dream. And honestly, back then, the tool was terrible.

Dumb, and a little mean

Don’t picture today’s AI. The vibe-coding of that era wasn’t just dumb. Some days it felt spiteful.

Here’s what it did. I’d ask it to fix one part of the screen — call it section A. It would fix A. Then it would quietly delete B, C, and D — the parts that were working fine — and leave A sitting there alone. I asked for one change. I’d come back and find that hours of work had vanished.

And there was no safety net. No real undo. No autosave worth the name. Once something was gone, it was gone.

My stupid, beautiful workaround

So I did the dumbest thing that worked. Before I changed anything, I copied the whole project and saved a spare. Make a copy, edit. Make another copy, edit again. If the AI nuked everything, I could fall back to the last good copy.

I stacked copies on top of copies. Copy 1, copy 2, copy 3. I crept forward one square at a time, dragging a trail of backups behind me.

This is the part I want you to hear, because it’s the part nobody tells you: the embarrassing manual workaround is allowed. You don’t need the elegant method. You need a method that lets you keep going when the tool betrays you. Folders full of final, final_v2, final_REAL are not a sign you’re doing it wrong. They’re a sign you’re still moving.

It was painfully slow

Moving a single program onto the NAS took days. Days.

Today I’d hand the same job to a modern AI and it would be done in about ten minutes. I’m not exaggerating. The thing that now takes ten minutes once cost me several full nights of stacking copies and inching ahead.

If someone had watched me, they might have thought I was wasting my time. Why grind through that by hand? It’s a fair question. I didn’t have a good answer at the time.

The strange part: I had fun

Here’s what I didn’t expect. I enjoyed it.

Every night the program got a little better, and I could see it happen. A button that didn’t work yesterday clicked today. A crooked layout snapped straight. A number that had been wrong came out right. Those tiny wins were oddly addictive. I’d look up and it was 2 a.m. Fix one square, feel good, reach for the next one.

It was hard and it was fun at the same time. Those two aren’t opposites. The difficulty was why the small wins tasted so sweet. If it had been easy, getting a button to work wouldn’t have meant anything.

Looking back, the thing that kept me going wasn’t some grand sense of mission. It was simpler than that. This was fun. That fun alone got me through those long nights with that dumb little tool.

Why this matters if you’re starting now

It’s tempting to wait for the tools to get good. They’re far better now, and that’s real. But the tools were never the point.

The point was the small, daily proof that I made this work. You can get that today with way less pain than I went through — and that should make starting easier, not scarier. The slow version still taught me everything. The fast version just lets you learn it faster.

So pick something small. Let the AI break it. Save a copy first. Then fix one square and feel that little click of progress. That feeling is the engine. Protect it and you’ll outlast every dumb tool and every bad night.

I kept stacking copies and moving programs onto the NAS, one at a time. And then one day the phone rang — a call that would change how I worked entirely. That’s the next story.