Former Google, Apple, and OpenAI Researchers Launch Trajectory to Build Continual Learning Platform for AI
Key Takeaways
- ▸Trajectory launches with $15M seed funding to build the first platform for continual learning in AI systems
- ▸Backed by prominent investors including Google DeepMind's Jeff Dean and Stanford AI pioneer Fei-Fei Li
- ▸Platform enables AI systems to improve continuously from real-world user interactions rather than remaining static after training
Summary
A group of AI researchers from Google DeepMind, Apple, OpenAI, and Meta Superintelligence Labs announced the launch of Trajectory, a new startup focused on solving one of AI's most pressing limitations: the inability of AI systems to improve continuously after training. The startup has secured $15 million in seed funding at a $115 million post-money valuation, led by venture capital firm Conviction with participation from Bessemer Venture Partners, Radical VC, and BoxGroup, along with notable individual investors including Google DeepMind chief scientist Jeff Dean and Stanford professor Fei-Fei Li.
Trajactory's platform enables companies to regularly improve their AI products by training on real-world user interactions—a capability that leading AI researchers, including Turing award winner Richard Sutton, have identified as essential for advancing toward superintelligent AI agents. The startup's approach is inspired by the success of AI coding products like Cursor, which continuously learn from user behavior to ship regular model improvements. CEO and cofounder Ronak Malde previously worked at Google DeepMind and believes continual learning is the primary reason AI coding products have proliferated rapidly among developers.
Rather than relying on off-the-shelf models from OpenAI or Anthropic, Trajectory customers like Decagon (an AI customer support agent builder) start with open-source models post-trained for their specific use case. Trajectory logs instances where the AI falls short and uses this feedback to post-train new models as frequently as weekly. The startup claims these post-trained models outperform frontier lab models on narrow tasks critical to a company's product, addressing a key pain point for enterprises seeking to deploy specialized AI systems without maintaining large teams of forward-deployed engineers.
- Early success in AI coding products validates continual learning approach; Trajectory aims to extend this to customer support, coding assistants, and other domains
