Trajectory Launches to Bring Continual Learning to Enterprise AI

A new startup, Trajectory, has emerged from stealth with $15 million in seed funding to address one of the most significant bottlenecks in artificial intelligence: the lack of a feedback loop. Founded by former researchers from Google DeepMind, Apple, and OpenAI, the company aims to move AI beyond static models that remain unchanged after their initial training. By enabling systems to learn continuously from real-world user interactions, Trajectory seeks to bridge the gap between general-purpose models and highly specialized, evolving enterprise tools.

Currently, most frontier AI models are "frozen" once training is complete, meaning they repeat the same errors indefinitely. While some coding-focused AI products have begun implementing basic forms of iterative improvement, Trajectory intends to standardize this process across all industries. By utilizing open-source models as a foundation, the platform allows businesses to fine-tune AI performance based on specific operational data—such as customer support outcomes—effectively creating a cycle where the model improves its accuracy and relevance on a weekly basis.

This development is significant because it addresses the "verifiability" problem in AI. While coding is objectively measurable, other business domains often lack clear metrics for success. Trajectory’s approach helps companies define and optimize for their specific needs, potentially reducing the reliance on large, expensive teams of engineers dedicated to manual model maintenance. With high-profile backing from industry leaders like Jeff Dean and Fei-Fei Li, the startup is positioning itself to be the infrastructure layer that transforms AI from a static utility into a dynamic, self-improving asset for the enterprise.