Vibe coding and AI agents classes in Oadby

The learner downloads the roads and buildings of a rectangle around Oadby through the Overpass service: 161.1 km of road and 1,375 mapped buildings. A trial places a team of agents at random points on the network and the same number of jobs at random buildings. Under the contract net, jobs are announced one at a time; each agent still free bids its shortest driving distance to the job, and the lowest bid wins. For comparison, the Hungarian method finds the allocation with the smallest possible total distance, and a third run simply allocates at random.

Bidding is cheap and mostly decent. With three agents it matches the ideal plan more often than not; with ten it never does, and typically drives 12.2% further, because an early job can grab the agent a later job needed far more. Still, that is a fraction of the 71.2% wasted by allocating at random. The practical lesson is a trade: the planner gives the shortest total but must know every job and every agent in advance, while bidding works as jobs arrive and nobody is in charge. Which one a team should use depends on what it knows and when.

Knowing how work is shared keeps Oadby teenagers in charge of the agent teams they design, and that starts with learning to code in 2026. The longer argument is in why teenagers should still learn to code in 2026.