AI agents are becoming a new interface for how people discover, reason over, and act on enterprise data. But truly agentic systems require more than an LLM connected to a database. They need open data formats, governed catalogs, shared semantics, portable knowledge, and transparent ways to define both the agents and the work they perform.

This session explores the emerging Open Agentic Lakehouse, where open lakehouse technologies and open agent standards come together as one interoperable architecture. We will examine how Apache Parquet provides efficient data storage, Apache Iceberg manages open analytical tables, Apache Polaris provides an open catalog and governance layer, Apache Arrow enables high-performance data movement, and Apache Ossie helps create portable semantic context for humans and AI.

From there, we will move above the data layer into the architecture of agentic systems. We will explore knowledge graphs for persistent context, execution graphs for defining and governing multi-step agentic work, and agent profiles for making specialized agents portable across tools and runtimes.

The session will also introduce two emerging open specifications I am developing: the Agentic Graph Specification (AGS), which represents agentic work as portable, reviewable execution graphs, and the Open Agent Profile (OAP), which represents the identity, instructions, tools, permissions, context, and learned state of an agent as a portable artifact.

The goal is a future where neither your data nor your agents are trapped inside proprietary platforms. Data, semantics, knowledge, execution plans, and agent identities can all participate in an open ecosystem, creating an architecture where organizations can assemble agentic systems from interoperable components while retaining governance, portability, and control.