A group of trailblazing AI researchers from Google DeepMind, Apple, OpenAI, and Meta launch a startup named Trajectory to build an AI platform that continuously learns from real-world user interactions to improve products.

The platform targets non-coding AI domains beyond software development, with examples like Decagon, a customer-support AI, where Trajectory logs failure cases (such as queries routed to humans) and uses them to re-train models on a weekly cadence.

Trajectory plans ongoing post-training updates by leveraging real user interactions to optimize models, starting from open-source base models pre-trained for specific products and applying weekly updates in practice.

The platform aims to reduce reliance on in-house forward-deployed engineers by autonomously improving AI stacks, with current customers including enterprise tools like Clay for sales and Harvey for legal, and aims to expand to Fortune 500 firms.

The founding team features Ronak Malde as CEO, Arjun Karanam (former Apple AI researcher), and Michael Elabd (ex-Google DeepMind robotics), each bringing deep experience from top AI labs.