AI Can't Do This Puts AI to the Bounty Test

Field notes

Challenges are organized by task type and expose operational detail rather than only declaring a winner: attempts, challenger, model count, bounty, and status appear in the public tables. Examples include a 14-constraint itinerary search, packet-capture protocol reconstruction, and financial-workbook repair. The page reports 1,457 total attempts and names systems such as Bolt, Cursor, Windsurf, Cognition, and Replit. Sponsorship is explicitly described as visibility and bounty growth, not an investment or wager.