PlainLogic Arcade · AI Skills
Debug the Build
The AI built something that doesn’t work. What do you tell it next? Eight broken builds, four prompts each — pick the one that gets a real fix.
The game
Debug the builds
Play · Watch · Learn
How it works
1 · Meet the broken build
Each round shows a tiny fictional project the AI built — a sketch of what it looks like and exactly what goes wrong when you use it.
2 · Pick the next prompt
Choose the best thing to say to the AI next. One option is a sharp bug report; the other three are traps that sound tempting and waste the AI’s one shot. Get it first try for 2 points, second try for 1.
3 · Read the why
Every answer explains itself: why the best prompt works, mapped to six debugging principles — and why each trap would have failed. Eight builds, sixteen points.
Under the hood
What this teaches
Debugging is information work. Most failed AI fixes fail before any code changes: the prompt didn’t say what actually happens, didn’t narrow the problem, or asked for six changes at once. The game’s six principles are the fix for that — they work on real projects, in bug reports to humans, and in every AI tool you’ll use.
“It doesn’t work” is not a symptom. Expected vs. actual is the single highest-value sentence in debugging: this should happen, this happens instead. The trap prompts in this game (“Fix everything”, “Make it work”) are the exact messages that send AI assistants \u2014 and junior developers — guessing.
Rewriting everything is the most expensive trap. It feels decisive, but it discards the parts that work and trades one known bug for a family of unknown ones. The game punishes the rewrite option every time, because production punishes it too.