05 September 2026

Microsoft Foundry: agents, models, and tools in one plane

What Learn calls Microsoft Foundry (formerly Azure AI Foundry / Azure AI Studio naming). Enterprise build path beside Microsoft 365 Copilot.

microsoftfoundryazureagents

DIGITAL SKILL WORKS LTD / 12 PANES

Learn now brands the unified Azure AI build platform as Microsoft Foundry. Older names (Azure AI Studio, Azure AI Foundry, hub-based classic portal) still appear in migration docs. For Horizon briefs, use the current Learn name and point builders to the new portal.

What it is

Microsoft Foundry unifies agents, models, and tools under one management grouping with enterprise capabilities: tracing, monitoring, evaluations, RBAC, networking, and policies under one Azure resource provider namespace. The Foundry portal is at ai.azure.com. Hub-based projects remain reachable via Foundry (classic); new investment focuses on Foundry projects in the new portal. Azure OpenAI resources can upgrade to a Foundry resource while preserving endpoint and keys per Learn guidance.

What it does

Builders can:

  • Create prompt agents (declarative) or hosted agents (bring your own code/framework in a container)
  • Browse a large Foundry Models catalogue (models sold by Azure vs partners/community)
  • Attach tools, memory, and retrieval via Foundry Toolbox
  • Use portal, SDKs, Azure Developer CLI, VS Code, and coding-agent skills/MCP
  • Apply observability, content filters, private networking, and Azure Policy through the Foundry control plane

Roadmap items also describe publishing Foundry agents toward Microsoft 365 Copilot experiences and grounding in SharePoint via Graph — treat those as Horizon signals and verify status in MRC / Learn before promising them.

Why you should care

Board. Foundry is how custom AI becomes an estate, not a lab demo. Demand the same Steady bar as Copilot.

CISO. Model choice, network isolation, and agent hosted code change the attack surface. Align with Entra and Defender patterns; connect to Agent 365 when agents enter Microsoft 365.

CFO. Model deployment, provisioned throughput, and hosted agents are Azure cost lines. FinOps ownership required.

Adoption. Pro-code agents need maker standards and review boards — not only citizen Agent Builder policy.

How it maps to 12 Panes

  • Pane 03 Horizon: platform naming and model catalogue shifts.
  • Pane 04–05: control plane, content safety, audit.
  • Pane 06: Azure consumption vs M365 seats.
  • Pane 09: custom agents in work redesign.
  • Pane 12 Frontier: experimental models and hosted agents in sandboxes.

Suggested actions

  1. Standardise language: “Microsoft Foundry” in architecture docs; note classic hubs only where still in use.
  2. Require project RBAC, network, and evaluation gates before production endpoints.
  3. Decide when declarative M365 agents suffice versus Foundry hosted agents.
  4. Register Foundry-origin agents in Agent 365 / M365 inventory when published to Copilot channels.
  5. Keep Wave 0 Steady (identity, DLP, cost) ahead of wide model catalogue access.

Communication needed

Security, GRC, and FinOps for platform standards. Engineering and Adoption for builder pathways. Comms: avoid “Azure AI Studio” in external board packs unless explaining history.

Benefits and risks

Benefits. One plane for models and agents. Upgrade path from Azure OpenAI. Observability and enterprise controls documented on Learn.

Risks. Naming drift causing wrong portal/docs. Unreviewed partner models. Cost surprise on serverless and hosted agents. Publishing into M365 without CCS/Agent 365 alignment.

Sources

Back to blog