​Building the AI-Agent-Native Data Platform​

In this session, Junling Hu, Sr Director of Agentic Platform @ PayPal, explores building an "AI-Agent-Native Data Platform" by introducing a "Semantic Fabric" layer that transforms siloed enterprise data into programmable memory for autonomous agents. 🌟 Key Highlights [00:01:22] Evolution from Human-Centric to Agent-Native: Traditional platforms were built for human analytics, but the future requires architectures that prioritize agent-to-data interaction. [00:05:40] The Semantic Fabric Layer: A core engine is needed to unify entity definitions and enable real-time data joining across disparate systems. [00:07:07] Memory as Infrastructure: Agents require a structured memory system, including Episodic (events), Semantic (facts), Procedural (steps), and Temporal (sequences) memory. [00:08:24] Governance and Portability Protocols: Security and privacy policies should be "portable," traveling with the agent via standardized envelopes for auditing and compliance. [00:13:37] Multi-Agent Collaboration Flow: A semantic layer allows multiple agents to share context, enabling a seamless cycle of perception, reasoning, and action. [00:15:00] Protocol Standards (MCP, A2A & ACP): New communication protocols like MCP (tools) and ACP (internal) are essential for agents to discover and interact with the data ecosystem. [00:17:50] Implementation Roadmap: Enterprises can upgrade by auditing semantic gaps and building a memory fabric without rewriting existing data layers.