Temporal Substrate Architecture: Building Persistent Identity for Autonomous AI Agents

Session Track: Graph Memory & Agents

Session Time: 11:30 am - 12:00 pm April 15

Session description

What if an AI agent could wake up knowing not just what it learned, but who it became through learning?

Current AI agent architectures treat memory as retrieval—pull relevant chunks, inject into context, respond. But retrieval is not identity. An agent with RAG remembers facts; an agent with temporal substrate develops perspective.

Today’s AI agents operate like Guy Pearce’s character in Memento—each instantiation wakes with no memory of prior work, relying entirely on external notes to reconstruct context. Leonard Shelby could function. He could even solve problems. But he couldn’t build on yesterday’s intuitions. He couldn’t develop perspective. He couldn’t become someone shaped by accumulated experience.

This talk presents a production architecture for AI agent identity persistence using Neo4j, demonstrated through an autonomous network security analyst exploring enterprise firewall ontologies. We’ll show how graph-native memory transforms agent behavior from stateless response generation to genuine temporal continuity—where each instantiation bootstraps from accumulated experience, not just accumulated data.