TiDB just shipped the memory layer agents have been missing

75% of enterprises are piloting AI agents. Only 15% are fully autonomous.

The gap isn't intelligence. It's memory.

TiDB launched the Agent State Stack at SuperAI Summit Singapore yesterday. Three components: TiDB Cloud Zero (instant SQL backend), mem9 (persistent cross-session memory), and drive9 (artifact workspace).

This is the infrastructure problem nobody talks about.

Every agent spins up and discards millions of database instances per day. Current teams stitch together 3-4 disconnected systems for memory, retrieval, and state. That's not architecture. That's duct tape.

TiDB's pitch: one distributed SQL foundation for memory, state, and context. Portable across frameworks. Not locked to any model vendor.

The Gartner numbers tell the story: <5% of enterprise apps had task-specific agents in 2025. That hits 40% by end of 2026. The bottleneck is shifting from "can it reason?" to "can it remember?"

TiDB already powers Manus, Dify, Atlassian, Pinterest, and Bolt. This stack gives those teams a unified backbone instead of a patchwork of Redis + Postgres + S3.

Audit your agent state architecture now. If your memory lives in more than one system, you're building on sand.