Let's Play Brisca.

Couldn’t find a Spanish deck on the web that didn’t look like clip art, so I drew one. That was hour two of the forty eight. The Claude Builder Club App Challenge had a deadline, so the deck design and the rules engine had to land in parallel.

Brisca is a forty card baraja española. No eights, no nines. Trumps shift the game. The point is not the rules. The point is what happens when someone drops an As at the right moment.

The AI is not magic

Three difficulty levels, all rule based Python. No LLM in the loop at game time. The hardest strategy counts which cards have been played, tracks which high value cards are still out, conserves trumps until the trick is worth the spend, leads with zero point cards in the early game, and switches to high point non trumps once the deck is down to five.

Built with Claude Sonnet 4.5 as the coding assistant through Claude Code. The shipped opponent is deterministic and inspectable. That was a choice. A judge can read the strategy file and see exactly what hard mode is doing. An LLM at runtime would be slower, more expensive, and harder to debug, and it would lose on consistency.

The moment that actually mattered

About thirty hours in, the advanced strategy played a hand the way I’d coded it to. It dumped a zero point card to bait a trump out of me, then took the next trick on points. That was my code working. It wasn’t emergent behavior. It was the satisfaction of watching a thing I built do the thing I built it for.

The animations are Framer Motion. The art is pixel work I did between dev sessions. None of it is replacing game night with friends in Spain. All of it is closer than nothing.