A city GenAI team is building a permitting chatbot and already has a classical Azure Machine Learning workspace. What should they create for GenAIOps?

Creating a Microsoft Foundry project for the permitting chatbot separate from the classical AML workspace is correct because GenAIOps for municipal generative apps is organized around Foundry projects and related GenAI resources, while classical Azure Machine Learning workspaces remain the home for traditional MLOps. The city GenAI team should stand up a Foundry project for the permitting chatbot rather than stretching the classical AML workspace into a pretend GenAI control plane. Clear platform boundaries keep identity, content safety, model deployments, and agent tooling aligned with AI-300 GenAIOps skills.

Hosting the permitting chatbot in the existing AML workspace using managed online endpoints for inference fails because managed endpoints alone do not create the GenAI project boundary, connections, and governance surfaces Foundry provides. Creating a new AML workspace for the permitting chatbot and registering GenAI models in the model registry fails municipal security and availability because it splits permitting workloads onto a separate classical workspace outside the city's Foundry-based identity, content safety, and governance controls. Building an AML pipeline for the permitting chatbot with prompt flow evaluation steps in the workspace fails because classical AML pipeline steps do not provision Foundry capabilities or compliance boundaries.

Exam caveat: classical AML versus Foundry/GenAIOps separation is a recurring exam theme—generative municipal apps get a Foundry project; do not treat an AML pipeline with prompt flow evaluation steps as a Foundry project.

Operational check: create a Microsoft Foundry project in the approved subscription for the permitting chatbot, connect required municipal knowledge sources under managed identity, keep the classical AML workspace for non-GenAI models, and document which workloads live in which control plane.