Developers | Run to reliable on Azure

Empower
Growth

Developers

Who it's for & aim: This page is for developers and technical leads who integrate AI into applications, design prompts, or build internal tooling. It explains how to deliver reliable solutions while controlling costs and maintaining compliance. By following these guidelines, your work will align with the company's AI adoption policy and support overall quality, security and cost control.

Background problem: Hidden costs and brittle prompts can arise when developers call models directly or neglect testing. Without standards, prompt behaviour may drift over time.

Policy overview: Use a thin API gateway for all model calls, version prompts, run evaluations (evals) to track output quality, and monitor cost per task.

What you need to do:
• Route all API calls through the gateway (for logging, redaction and rate limiting).
• Treat prompts like code: version them, document changes and write tests.
• Run automated evals on each workflow and fix prompts that fail quality checks.
• Choose the smallest model that meets quality requirements to control spend.

Owner & exceptions: The development lead owns the gateway rules; any exceptions must be time‑boxed and documented.