Announcement
AI agent fleet reportedly ranks first on ClickBench
The poster says agents proposed 397 experiments, with 98 producing improvements and 289 retired with written reasons.
TLDR
The poster claims their agent fleet ranked first on ClickBench, 1.2× ahead of the next-best engine and 3.1× ahead of its starting codebase. They say an operations team of agents made 93 workflow changes; after an early check for weak ideas, experiments per day nearly doubled and median experiment time fell from 16 hours to 2.5. They also say the entry ranks first on a later-added concurrent QPS measure and plan to open-source the engine.
Combined views
—
6 Sources, first seen ago
