A group of former Google and Apple researchers have launched a new startup called Trajectory, aiming to build an AI platform capable of continuous learning. This is in response to the current limitations where AI models stop improving after their initial training. Trajectory’s CEO, Ronak Malde, believes that by integrating real-world user interactions, companies can regularly improve their AI products.

The startup has secured a $15 million seed round at a valuation of $115 million, with major investors including Google DeepMind’s chief scientist Jeff Dean and Stanford professor Fei-Fei Li. Trajectory plans to offer an open-source model that customers can post-train for specific applications, such as customer support, allowing the AI to adapt based on real-world interactions.

While other AI coding products are already implementing some form of continual learning, Trajectory aims to extend this capability across various domains. Malde argues that without continuous improvement, today’s most powerful AI models remain static and prone to making the same mistakes over time. The challenge lies in applying this logic to industries where success criteria may be less precise than code running successfully.

Companies like OpenAI and Anthropic are rushing to develop their own AI solutions, seeing a market for tools that can continuously learn and improve on their own without constant human intervention. Trajectory’s goal is to build such a product, reducing the need for in-house engineers to troubleshoot AI stacks. The company already has customers across various fields, including enterprise sales and legal AI.

Despite its ambitious goals, critics argue that Trajectory’s current models only update once a week, which may not meet the needs of all industries. Nonetheless, Trajectory is optimistic about the future, envisioning a platform capable of daily or even hourly updates for companies.