Most AI sessions send you home with notes. This one sends you home with something that runs, plus an honest sense of where it falls over. You're split into teams, each one led by a speaker who has actually shipped this stuff before. Pick one of three build categories, work from a supplied fictional company brief so nobody needs their own data or any prep, and get running inside ten minutes from a template. No coding background required, and stretch extensions for anyone who already builds. Then you spend the last stretch trying to break the thing you just made: bad sources, contradictions, questions it cannot answer. The failures are the point, because anything you do not find here, your customers will find for you. Each team closes by pitching what they built, live on stage, including what they broke. Three things you leave with: - A working agent you can point at real data on Monday - The specific ways it fails, discovered in a room rather than in production - A way to test the next one before you trust it