Project Every Record Counts: Data Engineering for Backend Engineers
Part 33: "Context Is the Product" — Metadata, Agents and AI-Assisted Data Engineering
Leadership wants a chat box that answers questions about the business. Rohan plans to hand a model the schema and a read-only SQL user. Ingrid hands it the metric definitions, the contracts and the catalog instead, through an MCP server that decides what may be asked: metadata as context, MCP, governed tools, AI-assisted pipeline work with guardrails, and the three risks: fluent wrong numbers, exfiltration and prompt injection through data.