Vibe coding and AI agents classes in Kettering

The learner downloads a small rectangle of central Kettering from the OpenStreetMap API and keeps the roads, paths, footways and steps. Each junction becomes a square on the board: 640 junctions joined by 756 links, including 101 dead ends. The rules are simple. The cop moves first; on each turn a player either walks along one link or stays put; the cop wins by landing on the robber's junction. Eight real places are pinned to their nearest junctions, among them Kettering Library, the Manor House Museum and Newlands Shopping Centre.

A minimax agent chooses a move by imagining every reply. The cop tries each of its moves; for each, it assumes the robber will answer with the move that is worst for the cop; then it looks another turn deeper, and so on, up to a limit called the horizon. At the horizon it scores the position by how many links still separate the two. Alpha-beta pruning is the clever part: as soon as a move is shown to be worse than one already found, the agent stops exploring it, because a sensible opponent would never allow it. The answer is identical; the work is not.

Every extra pair of turns multiplies plain minimax's work by roughly eleven, and at ten turns it examines 167,653 positions. With alpha-beta pruning the same search needs 4,217 and reaches exactly the same verdict. The savings grow the deeper the agent looks, which is why pruning matters so much to any game-playing program. The verdict itself is sobering, though: after ten turns the cop is still 24 links away. Looking ahead does not help if the chase cannot be won.

To settle that, the learner stops searching forwards and works backwards through every possible position, a method called retrograde analysis. Of all 408,960 starting pairs, the cop can force a capture from just 5,781, or 1.41%. In 77.4% of those, the robber starts on one of the 186 junctions that sit on branches leading to dead ends rather than on a loop; the wins take a median of 13 turns and at most 43. From Kettering Library the cop cannot force a capture against a robber starting at any of the seven other landmarks, not even the Manor House Museum six links away. Wherever streets form a loop around a block, the robber can keep the block between them forever.

Understanding how agents plan against an opponent gives Kettering teenagers a real head start with AI, a strong reason to learn coding in 2026. The longer argument is in why teenagers should still learn to code in 2026.