Five days, then fifteen minutes
A friend told me, “We use AI for everything. Why would we need a product if we can build any solution ourselves?”
Then she mentioned she’d spent five days trying to get ChatGPT to make a deduction puzzle for a game she was creating. She wanted a Sudoku and a “Murdoku,” but couldn’t get puzzles that worked.
That evening, I asked Claude to create a puzzle plugin. Fifteen minutes later, it generated both. The puzzle logic worked on the first attempt. It took a couple more iterations to get the plugin packaged in ChatGPT’s latest format.
Claude did the coding. What I contributed was framing the problem: define the puzzle rules, generate candidates, and check that each puzzle has one solution a person can reach by deduction.
The lesson I took from it isn’t that everyone needs to buy a puzzle plugin. It’s that access to AI and a useful result are two different things. Often the hard part is working out what to ask it to build, what constraints matter, and how to check the answer.
Access to AI and a useful result are two different things.