Vibe coding and AI agents classes in Newcastle-under-Lyme

The learner fetches Census 2021 tables from the Nomis API for both districts at once: population, age, cars, travel to work and distance to work. Then comes the key step. For each question, the program works out the answer under both readings of "Newcastle" and compares them. If the two answers are close enough for the user's purpose, the agent can reply without asking, mentioning that both places give about the same figure. If they are far apart, a wrong guess would mislead, and the agent should ask a clarifying question first.

Our rule was to ask when two shares differ by more than one percentage point, or two counts by more than 10%. Five of the nine questions need a clarifying question. The contrast is sharp: 36.61% of households in Newcastle upon Tyne have no car or van, against 19.42% in Newcastle-under-Lyme, so an agent that silently assumed the Tyneside reading would give a Staffordshire user an answer almost twice too high. Four questions barely depend on which town is meant. About 11.6% of workers in both places travel less than 2 km, and the counts for 30 to 40 km commutes are 596 and 608.

The train question shows why the rule itself is a design decision. The two shares, 0.63% and 0.39%, are only 0.24 percentage points apart, so the absolute rule says answer. Yet one is about 1.6 times the other, so a relative rule says ask. Neither is wrong; it depends on what the user will do with the number. The motorcycle counts, 235 and 256, differ by 8.9%, just inside our 10% line. Small census counts also carry deliberate tiny adjustments that the ONS adds to protect privacy, so a close call on small numbers deserves a note either way.

Judging when a machine ought to check with you is a skill Newcastle-under-Lyme teenagers can build now, and it is a big part of why coding is worth learning in 2026. The longer argument is in why teenagers should still learn to code in 2026.