Generating Data with ChatGPT

From Idea to Automation: How We Taught AI to Build Venn Puzzles

Initially, we asked AI to write some documentation. Then we realized ChatGPT could design a domain-specific language to simplify our data files, and then we realized our job is to bring the ideas and see what ChatGPT can do for us.

Does this scare anyone else?

The Problem

We are building a Venn diagram puzzle game. Each puzzle would be printed as a card, with players guessing the categories based on overlapping items. It was meant to be funny, educational, even political—but also repeatable. We wanted 1,000s of cards.

The issue? Every card must follow a specific format:

  • Pick categories
  • Place items in overlapping regions
  • Make sure it rendered correctly
  • Avoid typos, formatting mistakes, or sloppy metadata

This is not hard—just tedious and time-consuming. And easy to screw up.

Step 1: Document the Process by Explaining It to AI

Instead of writing traditional documentation, we explained the process to ChatGPT like we were training a new intern.

We said:

“Here’s how the Venn diagram is structured in HTML/CSS. Here’s how we use slots like a1, ab1, and abc1. Help us design a format so interns can create these cards easily and reliably.”

The result? A clean, Python-style Domain-Specific Language (DSL):

card(
  name="Fruits by Color and Texture",
  template="3-circle",
  categories={"A": "Fruits", "B": "Colors", "C": "Textures"},
  slots={
    "a1": "apple",
    "ab1": "cherry",
    "abc1": "grape jelly"
  }
)

Just like that, we had:

  • A format interns could learn in minutes
  • Machine-friendly structure for automation
  • An auto-validating metadata layer

Then we asked ChatGPT to write a how-to guide for new contributors. The AI wrote better onboarding docs in 5 minutes than most projects get in 5 months.

Step 2: Automate the Creation of the Files

Next realization? Wait… if the AI knows the format, and it can generate creative content based on a theme, why not have it generate entire puzzle cards directly?

We fed it prompts like:

Theme: Generational Tech Use Tone: Playful but educational Goal: Help players infer circle labels like 'Boomers', 'Gen Z', 'Millennials' based on tech artifacts

Bazinga (thank you Sheldon)—out came ready-to-use card(...) blocks, ready to drop into the build pipeline. Valid. Clever. Printable. Repeatable.

The Impact

Before this automation:

  • Each puzzle took ~2 minutes to write by hand
  • We needed thousands

Let’s say 2,000 cards x 2 minutes = 66 hours. That’s two full work weeks just placing puzzle pieces. Now?

  • We generate 10–20 cards in a batch
  • We spend 5–10 minutes reviewing, curating, and approving
  • Quality control stays human
  • Generation and formatting? Fully automated

More importantly, we removed dozens of potential failure points:

  • No more bad slots
  • No more typos in keys
  • No misaligned cards

Step 3: Shift From Labor to Leverage

This isn't just about Venn cards—it's about a new pattern for work:

  1. Explain the process to AI
  2. Design a format or structure together
  3. Write the docs to train others
  4. Let the AI generate the actual files
  5. Review, curate, ship

The human's role? Have the ideas, check the output, and improve the system. This is productivity through building systems that do the work.

The Road Ahead

We’re just getting started. Each time we recognize a repeating task, we now ask: Can we explain this to the AI? If yes, we’re not just speeding things up—we’re creating clarity, unlocking scale, and building systems that compound.

In every IT project, game, creative experiment, or data pipeline, there are processes like this. Most go undocumented. Many are never automated. But all of them could be—if you start by just explaining it.

So while we might save less than 100 hours total while creating Venn diagrams, we can expect similar efficiency gains on every project going forward when we remember to use what is in our toolbox.

Want to Try It?

Here’s a prompt to use with ChatGPT, Claude, or your favorite assistant:

I'm going to explain a process I normally do myself. I want your help designing a structure to let someone else (or AI) do it instead. Then, generate a how-to guide that teaches someone else to follow it, and validate the inputs. Once we’re aligned, I want you to generate actual examples in that format based on themes I provide.

The trick I am learning, is after every successful gain, take a break, and then ask your AI: How do we continue to improve this process?

Let Your Light Shine

We don’t always need to work harder. We always need to work smarter , and we need to do as our parents taught us — share. If this resonates with you, give it a try. Refine it. Teach it to anyone willing to learn.

And if you’re already using a system like this? Please drop a comment and let us know. Let’s build the future of work together.