Using AI to Improve My SFMC SQL Reference: A Version 2 Case Study

Version 2 Is the Point

I just released the second version of my SFMC SQL Reference, and what surprised me most wasn’t the technical improvement. It was how much my thinking changed while building it.

The first version worked, but it wasn’t great. It wasn’t organized the way I wanted, and it didn’t fully match on how I envisioned the usage. There wasn’t a guidebook for how to build something like this, so I learned by doing: building, breaking, and rebuilding.

This new version took about eight hours of focused work. It’s clearer, more usable, and closer to what I originally imagined. But it’s still not perfect, and that’s intentional.

Changing How I Think About "Ready"

I used to believe you shouldn’t share something until it was flawless. That mindset kept a lot of ideas stuck in my head instead of out in the world.

I’m slowly changing how I think about that.

Projects like this are helping me realize that real learning doesn’t come from waiting until something feels perfect. It comes from:

• Putting something out into the world

• Seeing what’s wrong with it

• Improving it based on what you now understand

Each time I go through that cycle, I learn more than I would have by trying to get everything right the first time.

That’s how this reference evolved. The structure is better. The examples make more sense. The flow feels more natural for real-world use. None of that would have happened if I waited for a “final” version before publishing.

Where AI Fits In

AI played a much bigger role in this version than just polishing words. A lot of the work came from giving it very specific instructions and then reasoning through the results.

It wasn’t a single prompt and done.

It was:

• Asking for a structure

• Reviewing what it suggested

• Realizing what didn’t make sense

• Refining the request

• Testing the output against how SFMC SQL actually works

• Repeating the process

If I didn’t understand how the JavaScript and HTML components behaved in real situations, I wouldn’t have known when the output was wrong or incomplete.

AI could suggest code and structures, but it couldn’t see how everything fit together in practice. I had to test it, spot what didn’t work, and adjust the prompts until the steps actually matched how the page and scripts needed to behave.

That back-and-forth — giving instructions, checking results, correcting them, and trying again — was a big part of the work. AI helped move things faster, but only because I could reason through what it produced and guide it toward something usable.

AI helped surface patterns and organize ideas faster than I could alone, but it still needed direction. It needed context. It needed correction. It needed boundaries.

In many ways, the work shifted from “writing” to “reasoning.” Instead of just typing content, I was:

• Defining what the reference should cover

• Deciding how examples should be grouped

• Checking that each query pattern was actually useful

• Making sure the steps were practical, not theoretical

AI accelerated the process, but it didn’t remove the need to think. It amplified whatever understanding I brought to the table.

That’s the part that doesn’t get talked about enough. Using AI well isn’t about letting it do the work — it’s about knowing how to guide it, challenge it, and shape the output into something real.

Why This Matters

What surprised me most is how much confidence comes from iteration. Each version teaches you something new:

• What’s unnecessary

• What’s unclear

• What’s missing

• What actually helps

And each version gets closer to something truly useful.

That feedback loop is what turns something rough into something useful.

This won’t be the last version. There will be another one — and probably another after that. That’s how this reference will grow, and that’s how I’m learning to build.

If you’re working on something — a guide, a tool, or even a skill — don’t wait until it feels perfect. Build it. Learn from it. Make it better.

That’s how useful things are made.

www.kushlife.net

SFMC SQL Reference | Kush Life