Introducing Trajectory: Revolutionizing Continuous Learning in AI
A group of AI researchers, who have previously contributed to the innovations at leading tech giants like Google DeepMind, Apple, OpenAI, and Meta Superintelligence Labs, has announced the launch of their new startup, Trajectory. The company’s mission is to enhance AI products by leveraging real-world user interactions for ongoing training and improvement.
Aiming for Continuous Learning in AI
Trajectory is set to address a long-standing challenge in artificial intelligence: continuous learning. This concept has been a significant hurdle in advancing AI technologies. While companies like OpenAI, Google, and Anthropic have made strides in developing powerful AI models, especially in fields like coding, math, and science, these models remain static post-training. Richard Sutton, a Turing Prize laureate, emphasized the necessity of continuous learning for developing superintelligent agents at the NeurIPS conference in December 2025.
Funding and Leadership
Trajectory has successfully secured a $15 million seed funding round, achieving a post-money valuation of $115 million. This round was led by Conviction, with contributions from Bessemer Venture Partners, Radical VC, and BoxGroup. Notable individual investors include Jeff Dean, chief scientist at Google DeepMind, and Fei-Fei Li, a Stanford professor and World Labs CEO.
The startup is led by CEO and co-founder Ronak Malde, a former AI researcher at Windsurf and a key figure in Google’s acquisition of the company. Joining him are co-founders Arjun Karanam, a former Apple AI researcher, and Michael Elabd, who has experience in Google DeepMind’s robotics division.
Applying Continuous Learning Beyond Coding
Malde highlights that current AI coding products, like Cursor, are already implementing early versions of continuous learning by using real user interaction data for retraining and model improvements. This approach has accelerated the success of AI coding products, prompting other AI labs to develop similar applications. Trajectory plans to expand this methodology to improve AI tools beyond the programming domain.
“Even the most powerful AI today is still static. The AI model you used yesterday will make the same mistakes today,” Malde explains. “Some companies are starting to move into the world of continuous learning. We are building the platform for every single company to enter the world of continuous learning.”
Tailoring AI to Specific Needs
Trajectory’s challenge lies in adapting this approach to industries where success isn’t as easily defined as in coding. Karanam points out that Trajectory’s platform helps customize AI models to meet specific company needs. Instead of starting with standard models from OpenAI or Anthropic, Trajectory offers customers the option to use open-source models retrained for their particular AI products.
For instance, Trajectory assists Decagon, a customer specializing in AI customer support, by logging instances where the AI falls short and retraining models weekly. This process reportedly enables Trajectory’s models to outperform those from Frontier Labs on tasks critical to the client’s product.
As Trajectory embarks on its mission to redefine AI learning, it holds the promise of ushering in a new era where AI systems continuously learn and adapt to real-world interactions. For more information, visit Here.
“`
