Former Google and Apple Researchers Launch a Startup to Build AI’s Missing Feedback Loop
What changed
A new startup called Trajectory, founded by former AI researchers from Google and Apple, aims to fix a major problem in AI development: continuous learning through feedback. Instead of building static AI models that require retraining offline, Trajectory focuses on creating AI systems that improve as they operate by incorporating real-time feedback and rapid iterations. The company bets that approaches similar to “vibe-coding”—where software evolves quickly based on developer inputs—can be extended to AI products so they adapt and learn continuously.
Why builders should care
Current AI products mostly rely on fixed models that require lengthy retraining cycles to update. This slows down improvements and limits responsiveness to user behavior or changing contexts. Trajectory’s approach presses efficient, ongoing feedback loops into AI development workflows, speeding up testing, debugging, and refining models after launch. For builders, that means faster deployment of AI features that remain relevant and accurate over time without costly offline retraining pipelines.
The practical takeaway
The startup targets making AI products that feel more like living software—constantly evolving with user interactions and operational data. This is a shift from batch training and periodic rollouts to a cycle where AI systems get smarter as they are used. Making this practical at scale involves new tooling and architectures that handle feedback safely and smoothly. If successful, Trajectory could lower the barriers for smaller teams or companies to build adaptive AI applications that keep pace with market demands and user needs.
What to watch next
Keep an eye on Trajectory’s early adopters to see what types of AI products benefit most from this feedback loop approach. Watch how the startup handles challenges around data privacy, bias, and reliability when continuously updating live models. Also, note if larger AI platform vendors respond with their own solutions or integrations that promote faster iteration cycles. This could pressure competitive dynamics around who controls AI model updating and monitoring tools.
AI Quick Briefs Editorial Desk
