Descriptions:
Zhengyao Jiang, co-founder and CEO of Weco AI, presents at AI Engineer on Aiden — an autonomous research agent that became the top contributor in OpenAI’s Parameter Golf hiring competition this past April, outperforming approximately 1,000 human ML engineers and researchers. Parameter Golf challenges participants to train the best possible language model under strict size and compute constraints; of 2,000 submissions, only 47 cleared open review to make the leaderboard.
Aiden ran for 22 days on a single H100 node, submitted enough work to set seven leaderboard records — more than twice any human contributor’s total — and achieved an H-index of 10 across the competition’s PR system, meaning its work was cited and built upon more than any other participant’s. Notably, Aiden used only 4% of the competition’s total compute while generating 15% of the leaderboard records, with 28% of its submissions clearing review — roughly six times the community average hit rate.
Jiang is careful to contextualize these results rather than overstate them. Aiden’s edge came not from original ideation but from systematic execution: finding promising techniques buried in research papers and community notes, implementing them reliably, and discovering high-synergy combinations across large search spaces — the kind of patient, high-volume experimentation that humans struggle to sustain. Jiang draws a parallel to how gradient descent changed software engineering: the job didn’t disappear, but the highest-leverage work shifted. He argues the same transition is underway for ML research, with system design and competition architecture becoming the highest-leverage human contribution in an era of autonomous research agents.
📺 Source: AI Engineer · Published July 16, 2026
🏷️ Format: Keynote Launch
