2026 is the year AI agents went from demos to production. Every agent — whether querying data, training models, or orchestrating pipelines — hits the same wall: “Where is the data, and do I have permission?” The metadata catalog is no longer a passive registry; it is the contextual brain agents depend on to discover, reason, and act.

This talk shares how we built TCCatalog on Apache Gravitino to serve this new role at Tencent Cloud scale. We cover cloud-native architecture, extensions for AI-native workloads (Iceberg + Lance + Model Registry + MCP), and hard-won production lessons — including real incident stories, performance data, and engineering trade-offs.

Speakers:

Jinsong Zhou: Staff software engineer at Tencent, PPMC member of Apache Amoro (incubating)

Jin Song Zhou, a staff software engineer at Tencent, brings 10 years of expertise in the fields of databases and big data. Currently leading the implementation of a lakehouse architecture based on Apache Iceberg at Tencent Cloud, also serving as a PPMC member for Apache Amoro (incubating), actively fostering community development.

Jie Zan: Staff Software Engineer at Tencent Cloud.

Jie Zan is a Staff Software Engineer at Tencent Cloud, working in the Big Data Infrastructure Platform team. He focuses on Data + AI technologies, including AI agents, metadata integration, and cloud-native data infrastructure. His work includes building metadata services based on Apache Gravitino and developing cloud-native data platforms such as TCBase BaaS to enable unified governance for data and AI workloads.