Trajectory raises $15M to build AI's missing feedback loop for continuous learning
Former DeepMind and Apple researchers unveil startup to make AI learn from mistakes in real time.
Deep Dive
Trajectory, founded by ex-Google DeepMind, Apple, and Meta AI researchers, raised $15M seed at $115M valuation. The startup builds a platform for continual learning, enabling AI to improve from real-world user interactions.
Customer Decagon uses its weekly post-trained open-source models to beat frontier labs on specific tasks. Other customers include Clay
Key Points
- Trajectory logs AI failures and retrains open-source models weekly, enabling continuous improvement from real user interactions.
- Raised $15M seed at $115M valuation from Conviction, with backing from Jeff Dean and Fei-Fei Li.
- Already deployed at Decagon (customer support), Clay (sales), and Harvey (legal); claims to beat frontier models on specific tasks.
Why It Matters
Enables continuous AI improvement without in-house engineers, unlocking real-time learning for enterprise applications.