## Microsoft Foundry is the new name for Azure AI Foundry In January 2026 Microsoft rebranded **Azure AI Foundry** to **Microsoft Foundry**. The product is the same unified platform for building, evaluating, and deploying AI applications and agents -- but the name now reflects that Foundry sits across the Microsoft estate (Azure, Microsoft 365, Dynamics 365, Power Platform, Copilot) rather than inside Azure alone. If you architect AI solutions on Microsoft, this rename matters more than it looks. URLs, SDK package names, portal labels, certification blueprints, and customer-facing docs are all being updated through 2026. The AB-100 exam blueprint, in particular, refers to **Microsoft Foundry** throughout -- so candidates studying from older Azure AI Foundry material need to mentally translate every reference. This post walks through the full brand history, what actually changed in the rebrand, the services every architect should know, code examples that still work today, and what AB-100 candidates should update in their notes. ## A short brand history: Azure AI Studio to Microsoft Foundry The product has been renamed twice in 18 months. Each rename reflected a meaningful expansion of scope, not just marketing. | Date | Name | What it was | |------|------|-------------| | Nov 2023 | **Azure AI Studio** (preview) | Unified portal for building generative AI applications on Azure: prompt flow, model catalog, evaluations, deployments | | May 2024 | **Azure AI Studio** (GA) | General availability with project-based workspaces, hub architecture, and content safety filters | | Nov 2024 | **Azure AI Foundry** | Renamed and expanded to include **Foundry Agent Service**, the **Models-as-a-Service** catalog, and tighter integration with Copilot Studio and the Agent Framework | | Jan 2026 | **Microsoft Foundry** | Rebranded again to reflect cross-product positioning across Azure, Microsoft 365 Copilot, Dynamics 365, and Power Platform | The November 2024 rename (Studio -> Foundry) was the more substantive change technically. The January 2026 rename (Azure AI Foundry -> Microsoft Foundry) is primarily about positioning and naming convention -- but it is also when several previously-preview services (Foundry Agent Service, Foundry Evaluations 2.0, Foundry Local) reached general availability. ## What actually changed in the 2026 rebrand Most of the change is naming. A smaller, more important slice is product capability. ### Naming changes you will see - **Portal**: ai.azure.com is being progressively redirected to foundry.microsoft.com. Both URLs work in 2026. - **Documentation**: `learn.microsoft.com/azure/ai-foundry/` paths are being mirrored under `learn.microsoft.com/foundry/`. The Azure paths still resolve. - **SDK packages**: the Python `azure-ai-projects` and `azure-ai-inference` packages keep their names for backward compatibility. New helper packages under `microsoft-foundry-*` are arriving for cross-product scenarios. - **Resource provider**: the underlying Azure Resource Manager provider remains `Microsoft.CognitiveServices` and `Microsoft.MachineLearningServices`. Existing Bicep, Terraform, and ARM templates do not need to change. - **Pricing meters**: meter names retain the `Azure AI` prefix for billing continuity. ### Capability changes that matter Under the new branding, three previously-preview capabilities reached GA in January 2026: 1. **Foundry Agent Service (GA)** -- the managed runtime for hosting agents built with the Microsoft Agent Framework, with built-in support for MCP tools, A2A handoffs, identity, and observability. 2. **Foundry Evaluations 2.0** -- a redesigned evaluation pipeline with golden-dataset CI gates, multi-turn agent evaluation, and content-safety scoring as a first-class metric. This is the version AB-100 expects. 3. **Foundry Local** -- a developer-mode runtime for running Foundry-compatible models on a laptop or edge device using the same `AIProjectClient` interface as cloud Foundry. A few capabilities also moved or were renamed: - **Prompt flow** is now positioned as the authoring surface for *workflow-style* agents and is increasingly being absorbed into the Foundry Agent Service designer. New work should default to the Agent Service authoring experience. - **Model Catalog** is now organised by *capability* (reasoning, multimodal, code, embeddings, speech, vision) rather than by publisher first. - **Content Safety** (formerly Azure AI Content Safety) is now exposed as the default safety layer in every Foundry deployment, including agents, with no extra wiring. Official reference: https://learn.microsoft.com/en-us/azure/ai-foundry/what-is-azure-ai-foundry ## The five Foundry services every agent architect should know Foundry is large. For AB-100 and for day-to-day architecture work, five services do most of the heavy lifting. ### 1. Foundry Agent Service The managed runtime for production agents. It hosts agents authored with the **Microsoft Agent Framework** or the lower-code Foundry Agent designer, exposes them through stable endpoints, and handles: - Tool invocation via **MCP** (Model Context Protocol) - Multi-agent handoffs via **A2A** (Agent-to-Agent protocol) - Identity and delegated permissions through **OAuth 2.0 OBO flow** - Observability (traces, evaluations, content safety scores) - Autoscaling and SLA-backed availability If you are designing an enterprise agent solution that needs to run reliably, this is the default destination. ### 2. Model Catalog The discovery surface for foundation models inside Foundry. The catalog spans Microsoft and OpenAI models (GPT-4o, GPT-4o-mini, o1, o3), open-weights models (Llama, Mistral, Phi, DeepSeek), specialised models (Whisper for speech, Florence for vision, text-embedding-3-large for embeddings), and industry-tuned partner models. **GPT-3.5 is deprecated** and should not appear in any new architecture answer on AB-100. ### 3. Foundry Evaluations The evaluation pipeline. You author evaluators (correctness, groundedness, fluency, safety, custom code) and run them against golden datasets either ad-hoc or as **CI gates** in your ALM pipeline. The exam treats golden-dataset evaluation gates as the default control before promoting an agent to production -- this service is how you implement that control. ### 4. Content Safety The responsible-AI guardrails layer. It scores prompts and completions for hate, sexual, violence, and self-harm categories and supports custom blocklists, prompt-shield (against prompt injection), and protected-material detection. In 2026 it is enabled by default on every Foundry-deployed agent and model. ### 5. Prompt Flow The visual authoring tool for chained prompt and tool workflows. Still supported, but in 2026 the default new-work surface for agent authoring is the Foundry Agent Service designer plus the Microsoft Agent Framework SDK. Treat Prompt Flow as the answer when the question is about a deterministic, DAG-shaped LLM workflow rather than an autonomous agent. ## What stayed the same For architects worried about migration cost, the short list of *no-change* items: Azure Resource Manager structure (hubs, projects, connections), the identity model (Entra ID, managed identities, OBO flow), networking (private endpoints, VNet injection, CMK), existing SDK package names (`azure-ai-projects`, `azure-ai-inference`, `azure-ai-evaluation`), pricing meter names, and existing Bicep/Terraform templates. If you have an Azure AI Foundry deployment in production today, the rename has zero operational impact. You change links in your wiki and move on. ## Code: calling Microsoft Foundry from Python The SDK still ships under the `azure-ai-projects` package name. Here is a minimal example of connecting to a Foundry project and invoking the Foundry Agent Service. ### Connect and run a chat completion ```python from azure.ai.projects import AIProjectClient from azure.identity import DefaultAzureCredential project = AIProjectClient.from_connection_string( conn_str="", credential=DefaultAzureCredential(), ) chat = project.inference.get_chat_completions_client() response = chat.complete( model="gpt-4o-mini", messages=[ {"role": "system", "content": "You are a concise assistant."}, {"role": "user", "content": "Summarise the AB-100 exam in one sentence."}, ], ) print(response.choices[0].message.content) ``` The `AIProjectClient` is the entrypoint for everything Foundry does: model inference, agent invocation, evaluations, and the model catalog. ### Invoke a hosted agent via Foundry Agent Service ```python from azure.ai.projects import AIProjectClient from azure.identity import DefaultAzureCredential project = AIProjectClient.from_connection_string( conn_str="", credential=DefaultAzureCredential(), ) agent_client = project.agents # Create or load a thread, post a user message, run the agent, poll for completion thread = agent_client.create_thread() agent_client.create_message( thread_id=thread.id, role="user", content="Find the latest support ticket for customer 7421 and draft a reply.", ) run = agent_client.create_and_process_run( thread_id=thread.id, agent_id="", ) messages = agent_client.list_messages(thread_id=thread.id) for m in messages.data: print(m.role, m.content) ``` In 2026 the `project.agents` namespace replaces the older `agent_service` import path that appeared in some November 2024 preview docs. Old code still works through a shim, but new code should use `project.agents`. ## How the rename affects AB-100 candidates If you are sitting AB-100, three things to update in your notes: 1. **Vocabulary**: anywhere your notes say *Azure AI Foundry*, mentally substitute *Microsoft Foundry*. The exam blueprint uses the new name throughout. Both names are still acceptable on the exam itself in 2026 -- Microsoft has confirmed scoring is unchanged -- but the question stems use the new name. 2. **Evaluations 2.0 and Agent Service GA**: any question that hints at evaluation pipelines, golden-dataset CI gates, or hosting production agents on a managed runtime is referencing the post-GA capabilities described above. *Public preview* is no longer the right answer for those services. 3. **Prompt Flow vs Agent Service**: if a scenario asks you to choose an authoring surface for a *new* multi-tool agent with autonomy, the Foundry Agent Service designer or the Agent Framework SDK is correct. Prompt Flow is correct when the workflow is deterministic and DAG-shaped. The Microsoft Foundry rename does **not** introduce new skills to AB-100 -- the blueprint already accounted for it in late 2025 -- but candidates studying from older material risk getting confused by terminology drift. When you are ready for realistic, exam-format questions that already use the 2026 vocabulary, sit a full-length mock at https://www.prepmycert.com/course/ab-100-microsoft-agentic-ai-business-solutions-architect. ## Foundry alongside Copilot Studio and the Agent Framework A recurring source of confusion: Microsoft Foundry is not the only place you build agents on Microsoft. The platform sits inside a three-surface stack. | Surface | Audience | Strength | Where it runs | |---------|----------|----------|---------------| | **Microsoft Copilot Studio** | Makers, business analysts, citizen developers | Low-code agents, deep Power Platform / Dynamics 365 integration, topics + actions | Power Platform | | **Microsoft Foundry (Agent Service)** | Architects, pro-code developers | Managed runtime, fine-grained control, MCP/A2A, custom models, evaluations | Azure / Foundry | | **Microsoft Agent Framework SDK** | Engineers building production-grade multi-agent systems | Code-first orchestration, MCP, A2A, .NET and Python, can deploy to Foundry Agent Service | Anywhere (Foundry, Azure Container Apps, on-prem) | AB-100 expects you to pick the right surface for the scenario. A common trap: defaulting to Foundry when Copilot Studio is the better answer for a business-user-driven scenario, or defaulting to Copilot Studio when the scenario clearly demands code-first multi-agent orchestration. ## Frequently asked questions ### When did Azure AI Foundry become Microsoft Foundry? January 2026. The product is identical to Azure AI Foundry from November 2024, with three previously-preview services (Foundry Agent Service, Foundry Evaluations 2.0, Foundry Local) reaching general availability at the same time. ### Do I need to migrate my existing Azure AI Foundry projects? No. The underlying Azure resources, SDKs, and APIs are unchanged. You only need to update human-facing references (docs, wikis, training, bookmarks) and pin the latest SDK version if you want the GA agent APIs. ### Is Microsoft Foundry the same product as Azure AI Studio? It is the same product line, renamed twice. Azure AI Studio was the original name (Nov 2023). It became Azure AI Foundry in November 2024 with a substantial capability expansion, and Microsoft Foundry in January 2026 with the GA of Agent Service, Evaluations 2.0, and Foundry Local. ### Does the AB-100 exam still mention Azure AI Foundry? The AB-100 blueprint uses *Microsoft Foundry* throughout. Older Microsoft Learn modules still reference *Azure AI Foundry*; both names are treated as equivalent on the exam in 2026. Candidates should be comfortable seeing either. ### What is the difference between Microsoft Foundry and Microsoft Copilot Studio? Foundry is pro-code and architect-focused: custom models, managed agent runtime, fine-grained orchestration, evaluations. Copilot Studio is low-code and maker-focused: topics, actions, agent flows, deep Power Platform and Dynamics 365 integration. They interoperate -- a Copilot Studio agent can call a Foundry-hosted agent via MCP -- and AB-100 expects you to pick correctly between them. ### Is Prompt Flow being retired? No. Prompt Flow remains supported and is the correct authoring surface for deterministic, DAG-shaped LLM workflows. The default for *new* agent authoring is shifting to the Foundry Agent Service designer or the Microsoft Agent Framework SDK. ### What is Foundry Local? A developer-focused runtime that lets you run Foundry-compatible models on a laptop or edge device using the same `AIProjectClient` API surface as cloud Foundry. It is intended for development, demos, and offline scenarios -- not production at scale. ### Which models should I default to in 2026? For multimodal balanced workloads: **GPT-4o**. For cost-sensitive high-throughput chat: **GPT-4o-mini**. For deep reasoning and multi-step planning: **o1** or **o3**. Use a **model router** when the workload mix justifies it. GPT-3.5 is deprecated. ## Next step If you architect on Microsoft Foundry day-to-day, the AB-100 exam is the most direct way to validate that work in 2026. The blueprint covers Foundry plus Copilot Studio, the Agent Framework, Dynamics 365, the Power Platform, Microsoft Purview, and the responsible AI lifecycle. When you are ready to test yourself with realistic, exam-format questions that already use the 2026 Microsoft Foundry vocabulary, sit a full-length mock at https://www.prepmycert.com/course/ab-100-microsoft-agentic-ai-business-solutions-architect. 599 questions, six mock exams, mapped directly to Microsoft's official skill areas.