A team of former AI researchers from Google DeepMind, Apple, OpenAI, and Meta Superintelligence Labs announced in May 2026 the launch of Trajectory, a startup building a platform that enables AI products to improve continuously by training on real-world user interactions.
Trajectory has raised a $15 million seed round at a $115 million post-money valuation. The round was led by Conviction, with participation from Bessemer Venture Partners, Radical VC, and BoxGroup. Notable individual investors include Google DeepMind chief scientist Jeff Dean and Stanford professor and World Labs CEO Fei-Fei Li.
The company was founded by Ronak Malde, who previously worked as an AI researcher at Windsurf before joining Google DeepMind when it acquired the coding startup in a $2.4 billion deal. His cofounders are Arjun Karanam, a former Apple AI researcher who worked on the Vision Pro, and Michael Elabd, who worked in Google DeepMind’s robotics division. The team currently numbers 11 researchers and engineers.
Trajectory’s core argument is that most AI models stop improving once training ends. Malde points to AI coding tools like Cursor as early examples of continual learning — using real interaction data to regularly post-train and ship improved models. He attributes the rapid growth of AI coding products partly to this feedback loop, and says Trajectory aims to bring the same approach to AI products outside the coding space.
Rather than building on off-the-shelf models from OpenAI or Anthropic, Trajectory starts customers with an open-source model post-trained for their specific use case. For customer Decagon, which builds AI customer support agents, Trajectory logs instances where the AI falls short — such as routing a query to a human — and uses those failures to post-train a new model as frequently as every week.
The challenge, the founders acknowledge, is that coding is easily verifiable in ways other industries are not. Karanam says part of Trajectory’s platform is helping businesses define and optimize for their own specific measures of success.
Source: WIRED