Case Study: Yutori achieves 2-3x faster web task completion with lower inference costs with Cord

Yutori achieves 2-3x faster task completion with Cord

Yutori, a startup founded by former Meta AI leaders, is reimagining web interaction with autonomous agents. Their challenge was that foundation models struggled to navigate the dynamic, real web reliably, and they faced critical data hurdles in obtaining high-quality human-annotated trajectories for supervised fine-tuning and scalable evaluation.

Encord built a scalable data pipeline for Yutori, creating a tailored system for high-quality training data and a comprehensive evaluation framework. This included a custom error taxonomy and thousands of weekly trajectory evaluations. The partnership enabled Yutori's Navigator to launch as the highest-performing web agent, achieving 10-20 percentage points higher accuracy than competitors and 2-3x faster task completion with lower inference costs.

Yutori