This startup wants to build AI that learns from real-world use

A new startup called Trajectory is building a platform that helps AI products improve continuously by learning from real user interactions. Maxwell Zeff reports for WIRED that the company was founded by researchers who previously worked at Google DeepMind, Apple, and OpenAI, and has raised a $15 million seed round at a $115 million valuation. Conviction led the round, with participation from Bessemer Venture Partners, Radical VC, and BoxGroup. Investors also include Google DeepMind chief scientist Jeff Dean and Stanford professor Fei-Fei Li.

Most AI systems today stop improving once their training ends. CEO and cofounder Ronak Malde puts it plainly: “The AI model that you used yesterday is going to make the same mistakes today.” Trajectory aims to close that gap through a process called continual learning, updating a company’s AI model as often as every week based on real interaction data.

How it works in practice

Rather than using off-the-shelf models from OpenAI or Anthropic, Trajectory starts customers with an open-source model that is post-trained for a specific use case. For customer service platform Decagon, the system logs instances where the AI fails, such as a customer query being handed off to a human agent, and uses those failures to retrain the model. Other current customers include enterprise sales startup Clay and legal AI startup Harvey.

Cofounder Arjun Karanam acknowledges a key challenge: unlike coding, where success is easy to verify, many industries have less clear-cut definitions of what a good AI response looks like. Part of Trajectory’s offering is helping businesses define and measure success for their specific needs.

The startup currently updates models weekly, meaning they remain static between those updates. Critics may argue this does not yet qualify as true continual learning. Cofounder Michael Elabd says that is only the beginning. The team’s stated goal is to eventually update models every hour or even after every single user interaction.

Trajectory currently works primarily with AI-native companies but plans to expand its platform to Fortune 500 enterprises.