Vibe coding and AI agents classes in Worksop

The learner downloads the roads mapped in OpenStreetMap across a box around Worksop: 315.0 km of road meeting at 2,733 junctions. Python picks 240 junctions at random, and each becomes a job: find every road within 1,500 m of that junction by the shortest route. The cost of a job is the number of road segments the search has to check. The cheapest job needed 326 checks and the dearest 8,368; the middle one needed 4,021, and the whole set 969,483. Shared perfectly among four agents, nobody could finish before 242,371, a quarter of the total, so that is the target.

The equal split failed because position predicts cost. Jobs in the westernmost quarter averaged 2,384 checks and those in the third quarter 5,609, so one agent carried 336,566 while another was done at 143,016. Dealing jobs out in turn broke that pattern and got within 6.2%. The last two methods share work while it happens. With a shared queue, every agent collects its next job from one central list, which gets closest to the ideal but means 240 visits to that list. With work stealing, each agent starts with its own block and, when it runs dry, picks another agent at random and takes one job from the far end of that agent's list. Over 200 runs, that took 33.4 steals and 18.0 wasted tries on average, about 51 contacts instead of 240, and finished within 0.6% of the ideal.

Contacts are not free. When each one was charged 50 units of time, stealing finished at 244,109 on average and the shared queue at 245,601, so the decentralised method came out ahead. Taking half of a victim's list at once cut the steals to 9.1 but finished slightly later, at 244,374, because big grabs are harder to balance at the end.

A Worksop teenager who has built that simulation will ask how an agent system shares its work before trusting it with a real job. That question comes from writing and testing code, and it is why learning to code still pays off in 2026. The longer argument is in why teenagers should still learn to code in 2026.