Olivier Bouchez · Data Governance

Articles / AI & Puzzles

You Love Logic. Logic Loves You.

What I learned building a whole website of 21 logic and word puzzles by directing an AI like a development team, without writing the code myself.

· by Olivier Bouchez

AI Generative AI Prompt engineering Logic puzzles Web development

I'm curious by nature. I wanted to find out how far an AI can go when you treat it like a development team: you bring the goal, the requirements and the quality bar, and it does the design, the drawing and the coding. So I built a whole website of logic and word games without writing the code myself.

Poster for Nalou Génie: “You love logic. Logic loves you.” A website of 21 free logic and word puzzles, designed, written and coded entirely by AI, shown with thumbnails of twelve of the games and Nalou the beagle with her four cat friends.
The site, Nalou Génie: 21 free logic and word puzzles, designed, written and coded entirely by AI, steered by prompts.

The result: 21 free puzzles

There are Sums, Lotus, Constellations, Skyscrapers, Cryptogram, Mystery Word, crosswords, mazes and more. A new grid is generated every time you play, and each logic grid has a single solution you can reach by pure deduction: no guessing required.

13 puzzles

Logic puzzles

Sums, Lotus, Constellations, Nalou & Ziggy, Lanterns, Skyscrapers, 24, Animal Sudoku, Caterpillars, The Grumpy Gnomes, Hungry Tummies, Ice Floe, Cuddles.

3 games

Word games

Crossword, Mystery Word, Cryptogram.

5 games

Arcade and adventure

Maze, Where's Nalou's Bone?, Hide and Seek, Operation Scyannah, Retro Tennis.

Everything is available in English, French and Dutch. It works on a phone, a tablet or a PC, it can be installed like an app, and it keeps working offline.

How: 100% AI-generated

I wrote no code. The design, the code, the drawings, the puzzle generators and the translations into English, French and Dutch were all written by AI. Even the daily gazette starring Nalou, my beagle, and her four cat friends comes from AI.

My part was to steer it with prompts: say what I wanted, say what "done" looks like, and look at what came back.

What I learned

The real lesson was how to specify and direct an AI through the complete design of an interactive website. In practice that came down to four habits.

  1. Write clear specifications and acceptance criteria. "Make a logic puzzle" is not a specification. "Every grid has exactly one solution, reachable by deduction alone, and a new grid is generated on every game" is one, and it is something you can check.
  2. Break a big goal into small, testable steps. A website of 21 games is not a single request. It was built one game at a time, each delivered as its own small change that I could play and check before moving on.
  3. Review every result, give precise feedback, and iterate. The first version of anything is rarely the right one. The retro tennis opponent was tuned until the computer was beatable at every level, and the beagle was redrawn several times, down to the paws. Precise feedback produces better revisions than vague dissatisfaction.
  4. Keep control of quality, privacy and licensing. These are decisions an AI should not make on its own. The games set no cookies, show no ads, do no tracking and collect no personal data; your progress stays in your browser. Code and content are released under a CC BY-NC 4.0 licence.

It's the same discipline I use every day in data governance: define what "good" looks like, check it, and keep a human in the loop.

The same principle runs through my AI governance framework, which describes how I evaluate, classify and document AI applications.

Try it

On your phone, tablet or PC. You can install it like an app, and it works offline.

Play in English Jouer en français Speel in het Nederlands

This article was drafted with Claude from my LinkedIn post and reviewed by me. See the disclaimer.