AI Agent Conference 2026: Alex Gallego presents the Agentic Data Plane

Agentic Data Plane 2026-05-04 | 14:25 – 14:45 EDT As AI agents move from prototypes to production systems, they need something traditional data platforms were never built for: a unified plane where agents can observe, reason over, and act on enterprise data with strict guarantees around latency, safety, and auditability. This talk introduces the technical foundations of a new systems category — the Agentic Data Plane (ADP) — and breaks down the core primitives engineers need to support autonomous, data-driven agents in real environments. We’ll walk through the key pillars of an ADP: Unified Data Surface: a single logical plane that exposes streaming data, historical data, and system state through a consistent programmable interface, eliminating fragmentation between OLTP, OLAP, and event systems. Deterministic, Low-Latency Eventing: sub-10ms ingestion → processing → propagation pipelines that allow agents to react to real-world conditions with predictable timing. Context Compute Layer: a query/compute engine capable of joining live streams with large historical datasets, shaping agent context on demand without ETL hops or offline preprocessing. Governance as a System Primitive: identity-aware access, per-agent authorization, isolation boundaries, and immutable audit trails built into the runtime rather than bolted on afterward. Action & Side-Effect Control: a safe execution path for agent-initiated writes, workflows, or external system calls, complete with policy gates, replayability, and observability. By the end, engineers will understand the architectural requirements for building an ADP, how it compares to familiar categories (streaming engines, data warehouses, service meshes), and how these primitives enable reliable, inspectable, production-grade agentic applications.