Nikhil Harithas describes enterprise requirements for agent identityAn AI agent identity is a persistent, verifiable identifier used to attribute messages, actions, permissions, and history to an agent., security and customer-controlled compute. He explains why installing an agent on a machine alone leaves customers responsible for a complicated compute lifecycle.
Harithas outlines a proposed Open Compute abstraction with drivers for different infrastructureInteroperability is the ability of different systems to exchange information and use it with compatible meaning and behavior., a gateway and an outbound relay. The design aims to let agent sessions use customer-controlled environments while supporting operations such as snapshots and resumptionCheckpointing saves a defined execution state so a computation or workflow can resume or recover from that point..
He discusses bookkeeping, unreliable infrastructure and long-running multi-agent tasksDurable workflow execution preserves progress so a multi-step process can resume after interruptions using recorded completion state., and says the project was planned for open release after the talk. These claims describe Factory’s architecture and plans rather than an independent security assessment.
Watch on YouTube
