Startup Aims to Fix AI Systems That Weaken Over Time, But Companies May Resist the Change

Trajectory, founded by former researchers at Google, Apple, and OpenAI, is offering a way for AI systems to learn and improve continuously from real user interactions, rather than being rebuilt in big batches once a year. The problem it solves is real: AI models get worse over time. The challenge: established companies with working systems may resist switching to something new, more expensive, and harder to manage—especially in industries where safety and reliability are critical.