A group of former engineers from Google and Apple has quietly launched a new startup called Trajectory, with the goal of making artificial intelligence systems better at learning from their own mistakes in real time. The company is focusing on what it calls continuous learning — a way for AI to adapt without needing constant retraining from scratch.
Who’s behind Trajectory
The founding team includes researchers who previously worked on large-scale machine-learning projects at two of the world’s biggest tech firms. Their backgrounds span both the research labs and the product-side engineering teams at Google and Apple, though the startup has not yet publicly named individual founders. The team’s collective experience suggests they are aiming to solve a problem that has long frustrated AI developers: how to make models that improve as they go, instead of staying frozen after deployment.
What continuous learning means for AI
Most current AI systems are trained on a fixed dataset, then released into the world. When new situations arise that weren’t in the training data, the model can fail or produce stale results. Trajectory’s approach involves feedback loops that let an AI adjust its behavior based on new inputs without forgetting what it already learned. The company claims this could make AI systems far more flexible — able to react to changing environments on the fly rather than waiting for a human to update the model.
Where the technology could land first
The startup says its work has direct implications for robotics and autonomous systems. A robot that can learn from each interaction — adjusting its grip on an unfamiliar object or navigating a cluttered room it has never seen — would be a major leap over current machines, which rely on pre-programmed routines. Similarly, self-driving cars, drones, and factory automation tools could benefit from a system that updates its knowledge continuously as it encounters novel scenarios.
The approach also addresses a well-known issue in AI called catastrophic forgetting, where a model trained on new data suddenly loses the ability to perform older tasks. Trajectory’s feedback-loop method aims to preserve past learning while integrating new information.
For now, the startup remains in an early phase. It has not announced funding, partnerships, or a specific product release date. But the pedigree of its founders and the specificity of its focus suggest that Trajectory is betting that the next generation of AI will need to be alive to the world around it — not just a snapshot of the past.
