Startup Targets Model Drift With Continuous Learning, But Adoption Barriers Remain

Trajectory, founded by former researchers from Google DeepMind, Apple, and OpenAI, lets AI models improve by learning from real user interactions instead of periodic batch retraining. The approach tackles model degradation—a genuine operational problem. Yet adoption faces headwinds: organizations with established ML systems may balk at replacing proven batch retraining with costlier, riskier continuous frameworks, especially in regulated industries.