One prompt, one app
The first hands-on Claude Code session for the group. Sebastian and Mary walked attendees through the desktop app in Code mode and then ran two live builds end-to-end: an interactive atlas of every lighthouse in Ireland pulled from OpenStreetMap, and a receipt scanner that turns a folder of images into a categorised spreadsheet. The through-line was that a single well-formed prompt can produce a working app — the craft sits in the prompt, not the code. Homework extended the loop by handing Claude Code its own browser via Agent Browser, so it can open what it built, check its own work, and iterate without the user in the loop.
TK — refine thesis after rewatching.
Tools covered
- Claude Code — Claude's agentic coding companion. Runs in the desktop app (Code mode) or terminal.
- Agent Browser — browser automation CLI for AI agents. Lets Claude Code open what it built, look at the result, and fix issues before handing it back.
- OpenStreetMap Overpass API — overpass-api.de. Public data source for the lighthouse exercise.
- Leaflet.js — JavaScript mapping library used by the lighthouse atlas build.
TK — confirm tool list.