We ran AI agent building competitions at 12 universities across 4 countries. Teams of students had 48 hours to build an AI agent that could complete a set of real-world tasks using our open-source tools.
Results: - 340 students participated across all events - Top teams built agents that outperformed our internal baselines - 8 competition participants joined our team as researchers or engineers - 3 winning projects were integrated into our open-source toolkit
The competitions proved that practical AI agent building skills are best developed through hands-on challenges, not lectures.