Tic-Tac-Toe Coach

Tic-Tac-Toe was solved decades ago. With perfect play from both sides, every game ends in a draw. This is true and also why most online versions of Tic-Tac-Toe are useless — they either let you win because the computer plays randomly, or they always draw because the computer plays perfectly, and either way there's nothing to learn.

This version is built around teaching, not winning. The AI uses the classic Minimax algorithm, which means it will never lose. But when you turn on Coach Mode, every time it's your turn, the game shows you which moves are strong, which are weak, and why. If you make a mistake — say, you fail to block a fork — the game pauses and explains what just happened. You can keep playing and lose, or undo and try again. That's how you actually get better at a game like this.

The strategies worth learning are the same ones competitive players have used since the 1950s. The first move matters less than people think — corners and centre are both fine; edges are weaker. The second move matters more. If you're playing first and you took a corner, your opponent should take the centre, because anything else lets you set up a fork (two simultaneous winning threats). If they don't take the centre, you should respond by taking the diagonally opposite corner and aiming for that fork on move three. Most kids and most adults get this wrong the first ten times they play. That's exactly when Coach Mode is most useful.

Once you can reliably force a draw against the AI, the game stops being interesting — and that's the point. Tic-Tac-Toe is a stepping stone to bigger logic games. The skills here (look at the whole board, find the threat, count moves ahead, recognise that some positions are losing no matter what you do) transfer directly to Cribbage, to Sudoku, and to Jazz Ball, where the same kind of forward-looking pattern matching is the whole skill.

Build notes. Minimax for Tic-Tac-Toe is short enough to fit on a postcard — there are only 9 positions and at most 9 moves, so even a naive implementation finishes the full game tree in microseconds. The interesting code is the explainer: for each potential player move, the algorithm computes how much worse the player's best outcome becomes versus the optimal move, and surfaces a human-readable reason for the gap.