Author(s): Dave R – Microsoft Azure & AI MVP☁️

Originally published on Towards AI.

Two practical patterns for Fabric-aware Copilot, bounded agents, OneLake, Spark, and CI/CD, with architecture and merge decisions kept in human hands.

Code generation is the easy part. The hard part is delegation.

The article explains that successful AI delegation in data engineering depends on giving agents bounded implementation tasks while engineers retain ownership of architecture, constraints, and merge decisions. It lays out two patterns: first, using Fabric-aware Copilot to plan a streaming/bronze-to-silver medallion pipeline and then constraining the scope so the agent only fills known gaps within an engineer-defined design; and second, building an autonomous, bounded development loop where an agent iterates inside a controlled context, guided by skills, deterministic tools, and shared interfaces, with tests and explicit review gates determining when work can be promoted. It also covers how to make these loops reliable via local-to-Fabric workflows using OneLake mounted into dev containers, CI to run the same checks, and a “scale the loop, not the uncertainty” approach for parallel agent execution. Finally, it provides guidance on choosing interaction modes (web portal vs chat vs CLI) based on how deterministic the task is, and summarizes the key guardrails: explicit architecture, explicit context, narrow scope, deterministic tooling, limited write access, testable acceptance criteria, and human review at merge time.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI