Amy Boyd

Amy Boyd leads Microsoft Foundry Developer Relations, helping developers build AI agents whose behavior remains visible, reliable, and safe after deployment. A principal cloud advocate at Microsoft, she combines developer education with practical expertise in agent observability and evaluation.

Boyd studied computer science at the University of Sheffield, focusing on natural language processing and machine learning. A placement working on Microsoft’s Bing search engine preceded experience across data visualization, machine learning, large-scale data, and connected devices; she subsequently advanced into AI advocacy and developer-relations leadership.

Her public projects include AI Nights workshop resources, an Azure Machine Learning fashion demonstration, and a Love Island data-science project. She also coauthored an introductory generative-AI workshop with Carlotta Castelluccio in Microsoft’s workshop library, taught Microsoft Fabric data pipelines, and helped troubleshoot agent deployment and search integration.

What dependable agents require

• Observability from the beginning: Instrument agents during development and continue monitoring them as models, customer behavior, and requirements change.
• Workflow-level agent evaluation: Assess whether an agent understood the request, selected appropriate tools, stayed grounded, completed its task, and met quality and safety requirements—not simply whether its final answer looked persuasive.
• OpenTelemetry-based agent tracing: Capture model interactions, tool calls, and execution steps using shared standards that connect Foundry with Azure Monitor and Application Insights across different frameworks.
• Fleet-wide agent monitoring: Give developers and administrators centralized visibility across multiple agents and interconnected systems while integrating with existing operational infrastructure.

In her AI Engineer Europe workshop with Nitya Narasimhan, Boyd illustrated these principles through a travel assistant equipped with web search, execution traces, and built-in evaluation. Her focus is practical: help developers understand where an agent’s actual behavior diverges from its intended job, then use that evidence to improve it.