Vibe coding and AI agents classes in West Bromwich

The learner asks an AI assistant for a small Python agent with three tools. One looks up the published population of any ONS built-up area. One counts how many residents of an area actually live inside Sandwell, using the official lookup that places every small census area in a built-up area. The third returns the borough total, 341,832. The agent reads a question, decides which tool calls to make, and writes an answer. The code arrives in seconds and runs first time. Then the learner asks it how many people live in the built-up areas of Sandwell, and it replies 1,565,195.

The answer is 4.58 times the whole borough, so something is badly wrong, and the tool call log shows what. The agent added up the published population of every built-up area that touches Sandwell, including all 1,121,375 people of the Birmingham built-up area because 3,156 of them happen to live on the Sandwell side of the line. Halesowen, Coseley and Darlaston crept in the same way. The code did exactly what the plain-English request allowed. It was the request that was vague, and the agent never stopped to ask. This is the central lesson of vibe coding: the AI writes what you say, not what you mean.

The fix is to count only the residents inside the borough for each area. Adding those up gives 341,843, which looks right but is 11 more than the published total of 341,832, because official small-area counts are adjusted for privacy and never add up exactly. So the learner adds two guardrails. First, no answer about Sandwell may exceed the published borough total. Second, whenever a total is asked for, the agent quotes the published figure and shows the sum of parts only as a check. Finally, every answer prints its tool calls, so a human can see how it was reached. That is what building AI agents responsibly looks like.

West Bromwich teenagers who can test and question an AI agent will be ahead of those who only use one, and in 2026 that makes learning to code more worthwhile than ever. The longer argument is in why teenagers should still learn to code in 2026.