The hackathon judging problem predates AI - it's always been biased toward demos that look impressive in a 3-minute pitch. What AI changed is the floor. Previously you needed at least one competent frontend person to make something presentable. Now anyone can generate a polished UI in an hour. Counterintuitively this should force hackathon organizers to judge on harder criteria: does it actually work with real data? Did you solve a genuinely hard integration problem? Is there a novel architecture decision? The best hackathon I attended had judges who asked to see the git log and architecture diagram, not just the demo.