From Data 2.0 to Data 3.0: Making Data Autonomous for AI

AI moves at machine speed — agents learn, act, and self-modify in seconds. But the data those systems depend on still moves at human speed: manual approvals, brittle pipelines, and weeks-long delivery cycles. Availability of safe data, whether structured, unstructured, managed or purchased, is the bottleneck holding back AI. The problem: Our current platforms were built for dashboards, not for agents. They’re Data 2.0 systems — centralized, pipeline-driven, dependent on human orchestration, and built on the "modern data stack". This talk introduces Data 3.0 — a new runtime abstraction for AI-native systems: the autonomous data product. These are self-contained, domain-aware services that self-contained services that… Serve data in any mode — tables, vectors, MCP servers, or whatever comes next — under one semantic model and policy. Enforce governance and contracts at runtime, so agents never act on stale or unsafe data. Expose intent-aware discovery APIs for both humans and agents to find, evaluate, and use data safely at machine speed. We’ll break down the core capabilities needed for data infra to actually keep up with GenAI, compare them to the limitations of today’s fragmented stacks, and look under the hood at architectural patterns for autonomous data products. Why it matters If you’ve ever duct-taped a schema registry to a vector store or hand-built pipelines so your agent could “see” context, you’ve felt the limits of Data 2.0. Data 3.0 flips that script — data that’s autonomous, semantic-first, and governed-as-code, ready for AI by design. You’ll walk away with a practical blueprint for building data infrastructure that moves at the same speed — and with the same autonomy — as AI itself.