AI landing zone: do you need one for D365 F&O agents?

An AI landing zone is a reference Azure architecture, published by Microsoft on GitHub, for deploying AI apps and agents securely and resiliently, with Portal, Bicep and Terraform implementations. It is still in preview.

You need one if you host, in your own Azure subscription, agents or models that work around your ERP: private networking, identity management, a gateway to models, monitoring. You do not need one if your agents are built in Copilot Studio and connected to F&O through the Dynamics 365 ERP MCP server, or if you use a third-party service that hosts its own AI: the infrastructure is then managed by the provider.

According to the Azure/AI-Landing-Zones repository:

• It is an "application landing zone": a foundation for AI app and agent workloads, with or without a "platform landing zone".
• It includes two building blocks that can be deployed together or separately: the Agent Landing Zone, built on Microsoft Foundry, and the AI Gateway Landing Zone, which uses Azure API Management as a gateway to centrally manage and serve Foundry models.
• The Bicep and Terraform implementations are based on Azure Verified Modules.
• It follows the Cloud Adoption Framework (AI scenario) and the Well-Architected Framework (AI workloads), with a design checklist.
• A full deployment is only supported in some regions, including, in Europe, francecentral, germanywestcentral, norwayeast, polandcentral, spaincentral, swedencentral, switzerlandnorth and uksouth.
• It may use preview services, and the project itself is in preview.

• You build your own agents in Microsoft Foundry that call F&O (through the MCP server, OData or your services) and handle sensitive data.
• Several teams consume models and you want a single gateway (quotas, logging, access control): that is the role of the AI Gateway Landing Zone.
• Your security policy requires private networking, data residency in a given region and Azure Policy governance for any AI workload.

In these cases, starting from a reference architecture avoids rebuilding the foundation and eases the CISO's review.

• Your agents live in Copilot Studio and use the Dynamics 365 ERP MCP server: hosting is managed by Microsoft, security goes through the agent's F&O roles. See Dynamics 365 ERP MCP server and Copilot Studio agents and X++.
• You use a third-party AI service: it is up to the provider to document its hosting. Ask for the region, how your data is used and access control.
• You are at prototype stage: an isolated test environment is often enough before industrialising.

As an example of what to ask a provider: Skalp AI states that it hosts its processing in Azure in Europe and never trains a model on its customers' data; the generated X++ code is delivered to your Azure DevOps and validated by your IT lead. See the CISO page and MCP for D365 F&O.

Terms such as Foundry, agent, MCP or system of action are defined in our agentic ERP glossary. For the bigger picture of AI-assisted X++ development: building X++ without a consultant in 2026. Pricing: Basic, Pro and Max plans.

• GitHub, Azure, "AI Landing Zones" (README), accessed 11 Oct 2026.
• Microsoft Learn, "What is Microsoft Foundry?", updated 13 Aug 2026, accessed 11 Oct 2026.