New Startup Tackles Model Drift, But Production Teams May Resist Retraining Overhaul

Trajectory, founded by former Google DeepMind, Apple, and OpenAI researchers, enables AI models to improve through continuous learning from live user interactions instead of periodic batch retraining. The approach addresses model degradation in production—a real operational pain point. Adoption faces friction: organizations with mature ML pipelines may hesitate to replace proven batch systems with costlier, operationally complex continuous frameworks, particularly in risk-averse industries requiring extensive validation before self-modifying production systems.