What we learned running an agentic hackathon with 30k+ hackers (twitter.com)

The recent Orchestrate hackathon, which concluded on September 12th, saw a record participation of over 35,700 developers, an increase from 12,871 in May. This AI-first event, held on HackerRank, encouraged developers to build real-world AI agents over a 24-hour period. Notably, participants showcased a variety of models and coding architectures, sharing transcripts and implementation artifacts that revealed effective AI-assisted engineering practices. The key takeaway was that the strength of submissions was less about the specific tools used and more about how well developers directed, tested, and retained ownership of their projects. Technical analysis indicated that successful submissions demonstrated a complete engineering loop encompassing planning, implementation, testing, debugging, correction, and review. Higher scoring submissions were linked to more visible development stages, affirming that a structured approach is critical for success in AI projects. Interestingly, while tool preferences varied, the strongest performers consistently used Claude Code in the highest-rated submissions, highlighting a shift towards more effective tools as participants gained experience. This hackathon serves as a significant case study for the AI/ML community, providing valuable insights into developer behavior, tool efficacy, and the importance of thorough engineering methodologies in building AI solutions.