Phasis pitches World API to orient AI agents before work

Phasis began by turning to graphs as a way to measure culture, aiming to capture relations not just between people but between a person and their values, beliefs and subcultures. The goal was to map culture so agents could grasp what it is composed of and decide based on its fast-changing nature.

According to Farley, the realization after that work was that the same process of cleaning, verification and building ontologies mapped well to other domains. That became the World API, which he described as representing many domains for agents to orient against, built into systems as they are now and using the primitives needed for a reality layer and expertise. Phasis says it builds and deploys Worlds so agents are oriented before work begins and stay current as circumstances change.

Farley framed orientation as a step added before an agent searches, plans or strategizes — a quick check against reality across a shared or private world. He argued that agents oriented against reality will compound differently from those piecing it together every time, and that the World model is the durable asset. Looking ahead, he said he expects the next year to bring scaling of agent swarms and auto research, alongside smaller but highly capable models.