When Small Experiments Turn Into Systems

Amy Westlake
May 19
3 min read

Situation

I used to check slides the hard way.

Click on an object. Look at the font. Click on another. Check the color. Click on another. Wonder if that blue is the right blue or just close enough. Repeat this a few hundred times across a deck you didn't build and aren't totally sure about.

The problem wasn't effort. It was physics. Human eyes aren't calibrated to distinguish Arial from Archivo at a glance. They're not built to catch a hex value that's off by thirty points. You can be careful and still miss things - and I was missing things. Probably others doing the same checks were too.

I'd been learning about Apps Scripts. Enough to know they existed. Enough to be curious. Not enough to do anything with them yet.

So I did what I'd started doing with most problems: I described it out loud to an AI and asked if there was a way.

The AI Move

The ask wasn't sophisticated. It was something like: here's what I'm doing manually, here's what I'm working in, is there a script that could catch this for me?

I wasn't thinking about building a system. I was thinking about one specific annoyance.

What came back was a script that could run against any Google Slide deck and flag objects where the font or color was off. Not delete them. Not fix them automatically. Just flag them - little markers I could review and decide what to do with.

I copied it into Apps Script, pointed it at a real deck, and ran it.

My first reaction was: Oh shit. That actually worked.

My second was: Huh. That's pretty cool.

Then I closed the tab and went back to work.

The Shift

The next time I had slides to review, I used it again.

Not because I'd decided to. More because it was there and it had worked, so why wouldn't I?

I used it again after that. And again. Four or five real decks in, something shifted - not in the script, but in how I was thinking about it. I'd stopped treating it as an experiment. I wasn't testing anymore. I was just doing the check this way now.

There was no decision point. No moment where I said: this is now my process. It became my process because I kept reaching for it, and it kept working.

That's when I realized: I hadn't built a system. The system had built itself - out of repetition.

The Pattern

Systems don't start as systems. They start as one attempt to do something slightly faster, or slightly better, or with slightly less friction than before.

What turns them into systems isn't intention. It's repetition. You do it once. It works. You do it again. It still works. At some point, you stop questioning whether to use it and it just becomes how you do the thing.

I've watched this happen with other experiments too. A Gem that started as a one-off for a specific report - now it's part of how I do status reporting every week. A set of quality questions I started using to check my own AI outputs - now they're something I run without thinking. None of these started with architecture in mind. They started with: I wonder if this would be faster.

The mechanism is always the same. Small friction. Specific attempt. Surprise that it worked. Quiet repetition. Then one day you notice it's just how you work now.

The Implication

You don't need to plan a system to end up with one.

Pick one repetitive task - something you do on a regular cadence, something where you've quietly accepted that it's just tedious. Describe the problem to an AI. See what it gives you. Try it once.

If it works, try it again the next time you need to do that thing. Don't decide whether it's a permanent solution yet. Just use it.

That's the whole experiment. The structure will show up on its own if it earns it.

What I'm Testing Next

The compliance script solved one piece. But it got me thinking about how many other checks I do manually that have the same profile - repetitive, visual, easy to miss. I've been mapping those. Some might be scriptable. Some probably aren't.

The interesting question isn't whether AI can automate them. It's which ones, once automated, I'd actually reach for again.

That's the real filter.