Former DeepMind and Apple researchers launch AI startup Trajectory

In short: A team of former researchers from DeepMind, Apple, OpenAI, and Meta announced a new startup, Trajectory, focused on making AI products improve as people use them.

What happened

Trajectory said it is launching to help companies update and improve their AI tools more often. The team includes researchers who previously worked at Google DeepMind, Apple, OpenAI, and Meta Superintelligence Labs.

The company’s main idea is to train AI on real-world user interactions. In simple terms, that means the AI can learn from what people do with it in everyday use, like which answers users accept, fix, or ignore. Think of it like a customer service script that gets rewritten based on the questions customers actually ask.

Trajectory is also betting that faster “iteration” can spread beyond software coding. Iteration just means making frequent small changes, checking what works, then repeating. This approach became popular in “vibe-coding,” where people build software quickly by giving instructions to an AI coding helper, then adjusting based on the results.

Why it matters

Many AI tools work well in a demo but struggle in day-to-day situations. If Trajectory’s approach works, companies could improve AI features more like updating a navigation app, using real driving data to fix routes and mistakes. It also raises practical questions companies will need to handle carefully, like what user data is collected, how it is protected, and whether people can opt out.

Source: Wired