Microsoft Foundry (formerly Azure AI Foundry) and the Gemini Enterprise Agent Platform (formerly Google Vertex AI, evolved into this developer platform at Google Cloud Next 2026) are agent platforms for building agents and agentic applications.
The main difference comes down to the cloud each platform lives on and each vendor’s starting point. Start from your existing cloud commitment and preferred model families, since both platforms cover model access, deployment, governance, and agent orchestration.
Here’s where the two platforms actually diverge:
- Core design differences
- Ecosystem of services
- Models these platforms can run
- Process of building, deploying, and maintaining agents
- How agents retrieve and use context
Azure AI Foundry vs. Google Vertex AI: at a glance
| Dimension | Azure AI Foundry | Google Vertex AI |
|---|---|---|
| Official name | Renamed Microsoft Foundry in 2025 | Evolved into the Gemini Enterprise Agent Platform at Google Cloud Next 2026. Not to be confused with Gemini Enterprise, the separate business-user app Google launched in October 2025, which took over what Google had been calling Agentspace |
| What is it? | Microsoft’s app and agent platform on Azure, launched at Ignite in November 2024 and rebranded Microsoft Foundry a year later | Google Cloud’s ML and generative AI platform, generally available since May 2021 and evolved into the Gemini Enterprise Agent Platform in April 2026 |
| Design organization | Control plane, resource hierarchy | Build, scale, govern, optimize |
| Entry point | Azure AI Foundry resource and projects | Agents |
| Agent definition method | No-code prompt agents and hosted agents | No-code agents with Google Agent Studio, and hosted agent-equivalent with Google’s ADK |
| Agent runtime | Prompt agents run on Foundry’s fully managed runtime; hosted agents run in micro VMs on per-session sandboxes | Both agent types run on Google’s managed runtime, Agent Runtime (formerly Agent Engine) |
| Agent development | Build prompt agents from the Foundry portal or the Foundry SDK. Build hosted agents using an agent framework like LangGraph or Microsoft Agent Framework | Build agents using the open-source and model-agnostic ADK by Google or any other open-source framework of your choice |
| Tools catalog | Built-in tools like file search, code interpretation, etc. Add new tools with remote MCP. Foundry also has its RAG solution in Foundry IQ | Built-in tools, such as Grounding with Google Search, Agent Search (formerly Vertex AI Search), RAG Engine, and code execution. You can also register your own APIs and MCP servers |
| Models used | Most frontier models, plus hundreds of Hugging Face Hub models on managed compute, which need a hub-based project in the classic Foundry portal | Most frontier models, and the 200-plus published in Model Garden |
| Multi-agent support | A2A in both directions: outbound calls to remote agents, and inbound endpoints that expose a Foundry agent. v1.0 is GA, v0.3 is preview | Any framework built for multi-agent orchestration and using A2A is supported |
| Context from enterprise data sources | Context grounding with Bing Search, Foundry IQ, and Fabric | Context grounding via Google Search, along with access to Cloud Storage, Google Drive, BigQuery, and the Apigee API Hub |
| Context from RAG sources | Foundry IQ knowledge bases built on Azure AI Search | Agent Search, RAG Engine, and Agent Retrieval (formerly Vertex AI Vector Search) |
| Knowledge graph, semantic layer, ontology, glossary | Microsoft locates ontologies, semantic models and graphs in Fabric IQ; Foundry IQ is a knowledge and retrieval layer on Azure AI Search | Google’s semantic and graph surfaces sit outside the agent platform, in Knowledge Catalog, BigQuery’s knowledge engine and Looker’s LookML |
| Agent memory and session state | Supports short-term, long-term, and procedural memory (fully managed) | Separates session memory from cross-session memory |
| Best fit | Enterprises already standardized on Azure and Microsoft 365 | Enterprises on Google Cloud that want native Gemini models, and direct BigQuery data integration |
How do Azure AI Foundry and Google Vertex AI differ in their core design?
Azure AI Foundry (now known as Microsoft Foundry) is designed as a control plane for agents, models, connections, and other services that interact with agents. It follows a resource hierarchy, with a Foundry resource at the top that handles security and deployments, and Foundry projects as the isolation boundary.
The Gemini Enterprise Agent Platform, which is what Vertex AI became, handles the organization differently, structuring everything around agents across four pillars: build, scale, govern, and optimize.
While the design is slightly different, the methods of creating and deploying agents are similar. Here’s a quick comparison of the core design features across the two platforms.
| Design feature | Azure AI Foundry | Google Vertex AI |
|---|---|---|
| Isolating unit | Foundry resource and projects | Agents |
| Entry point | Responses API | Model-agnostic Google’s ADK (Agent Development Kit) for all communication |
| Per-agent identity | Entra Agent ID | Agent Identity |
What is the difference between building and running agents in Azure AI Foundry vs. Google Vertex AI?
Both platforms support two paths for building and running agents: one that lets you build configuration-only agents without code, and the other is where you bring the code, the agent loop, choose the runtime, and more. Here’s a side-by-side comparison considering agents on the platforms.
| Aspect | Azure AI Foundry | Google Vertex AI |
|---|---|---|
| How is an agent defined? | You can create prompt agents that are no-code configuration-only and also hosted agents where you bring and build your own code | You can have the no-code option with Google Agent Studio. You can have the hosted agent-equivalent with Google’s Agent Development Kit |
| Where does the agent run? | Prompt agents run on Foundry’s fully managed runtime, while hosted agents run in micro VMs on per-session sandboxes | Both types of agents run on Google’s managed runtime, Agent Runtime (formerly Agent Engine) |
| How do you develop an agent? | Prompt agents can be built from the Foundry portal directly or the Foundry SDK. Hosted agents can use an agent framework of their choice, such as LangGraph, OpenAI Agents SDK, Microsoft Agent Framework, among others | You can use the open-source and model-agnostic ADK by Google or use any other open-source framework of your choice and deploy it to Agent Runtime |
| What tools can the agent access? | Foundry agents have access to built-in tools like file search, code interpretation, etc. You can add any new tool using a remote MCP server. Foundry also has its RAG solution in Foundry IQ | Google’s Agent Platform agents use built-in tools, such as Grounding with Google Search, Agent Search (formerly Vertex AI Search), RAG Engine, and code execution. In addition, you can also register your own APIs and MCP servers that you can route through Agent Gateway |
| What models can the agent use? | Azure AI Foundry gives you access to most frontier models, plus hundreds of Hugging Face Hub models on managed compute; those require a hub-based project in the classic Foundry portal | Google Vertex AI gives you access to most frontier models, including Gemini and Gemma, along with the 200-plus models published in Model Garden on the Gemini Enterprise Agent Platform |
| What about multi-agent? | Foundry agents call remote agents over A2A and can be exposed as A2A endpoints themselves. A2A v1.0 is generally available and v0.3 remains in preview; Microsoft tells new integrations to target v1.0. Inbound endpoints are configured through REST or the SDK rather than the portal, and are text-only with no streaming | Google created the A2A protocol and contributed it to the Linux Foundation in June 2025. The ADK or any other framework built for multi-agent orchestration and using A2A can be used with Google Vertex AI agents |
Building the agent is only the beginning. How efficiently the agent works is primarily a function of the knowledge and context it has access to at any given time, especially while it is in action. Here’s a look at the different ways both platforms make knowledge and context available to their agents.
How do Azure AI Foundry and Google Vertex AI connect agents to the knowledge and context they need?
Agents in Azure AI Foundry can be registered in the Entra Agent Registry. Those agents have access to Azure services like Purview, Azure Database for PostgreSQL, Cosmos DB, as well as OneLake and SharePoint via Foundry IQ.
Agent Platform agents ground on Google Search and reach BigQuery datasets through agent-scoped IAM permissions and first-party BigQuery tools. Colab Enterprise is a developer notebook surface under Build, not a tool the agent calls.
But there are many other ways agents can get knowledge and context from within and outside your enterprise systems. Here are the broad categories and how the two platforms support them.
| Context | Azure AI Foundry | Google Vertex AI |
|---|---|---|
| Documents, files, APIs, and databases from the enterprise | Context grounding using Bing Search, Foundry IQ (SharePoint, OneLake, and Blob Storage), and Fabric; agents can reach a catalog of MCP-enabled tools to connect to enterprise systems | Context grounding using Google Search, along with access to Cloud Storage and Google Drive, BigQuery, and the Apigee API Hub |
| RAG context sources like vector and hybrid indexes | Foundry IQ knowledge bases built on Azure AI Search at the heart of agentic retrieval following the RAG pattern | Agent Search, RAG Engine, and Agent Retrieval powering the RAG pattern for Agent Platform agents |
| Knowledge graph, semantic layer, ontology, glossary, and lineage | Microsoft locates ontologies, semantic models and graphs in Fabric IQ, and Foundry IQ is a knowledge and retrieval layer built on Azure AI Search. Fabric IQ has an ontology feature in preview that a Foundry agent reaches as a server-side tool, not as a Foundry IQ knowledge source, and it needs a Fabric F2 or Power BI Premium P1 licence. Otherwise the route is an external tool over MCP | Google’s semantic and graph surfaces sit outside the agent platform, in Knowledge Catalog, BigQuery’s knowledge engine and Looker’s LookML, so an agent reaches them through tools rather than through a layer the platform provides |
| Agent memory and session state | Azure AI Foundry supports short-term, long-term, and procedural memory, all of which is fully managed | Google Vertex AI separates session memory from cross-session memory using Sessions and Memory Bank, both of which are useful for agents |
Notice that there are two clear patterns where neither platform makes it easy to bring context to the agent:
-
While you can connect a lot of enterprise systems to your agent, it involves several separate integrations. These are one-to-one integrations.
-
There’s no knowledge graph or semantic layer or ontology that stitches all the context and knowledge from various sources in your organization into one cogent and coherent context layer.
This is why agents often don’t get the right context, and often guess. What agents really need is a context layer spanning the entire enterprise.
Both agent platforms, Azure AI Foundry and Vertex AI, support this pattern. That’s where Atlan, the Context Layer for AI, comes into the picture. Here’s how Atlan gives agents on either platform the context they need.
How does Atlan help with context in Azure AI Foundry and Google Vertex AI?
Atlan is the Context Layer for AI. It serves agents on both platforms, Azure AI Foundry and Google Vertex AI, with the precise context they need at any given time. Atlan’s MCP server delivers context to agents at runtime by bringing context from all enterprise systems into the enterprise context layer. Atlan accumulates, organizes, and curates context using three core capabilities.
Enterprise Data Graph is a graph that contains the entity relationships between systems, assets, code, products, and more. Active Ontology distills organizational language and knowledge into glossary terms, domains, metrics, and relationships as a semantic layer, aligned with Open Semantic Interchange. And the Context Engineering Studio lets you build Context Repos for your projects, run evaluations across different contexts, and deploy the version that works best for your agents.
Atlan’s enterprise context layer offloads the complexity of making context available to agents. It’s platform-agnostic and speaks MCP, which is why Foundry agents and Vertex agents can both talk to it and reach the same context at any given point.
Moving forward with agentic platforms for Azure and Google Cloud
AI Foundry and Vertex AI are both flagship agentic platforms for Azure and Google Cloud. They reach the same end goal, giving you the capability to create and customize agents, but each implements it slightly differently.
Both platforms support most major frontier models, let you use MCP for tool calls, and allow multi-agent orchestration using A2A.
Whichever platform you choose, the key constraint for agents is their ability to reach the right context at the right time during a run. In large enterprises, organizational knowledge, process documentation, communication channels, and decision registers, in other words, context, is spread across hundreds of systems.
Both platforms contain information within their own boundaries. For agents to act effectively, they need context from multiple systems at once, which is where the need for a context layer comes in. Atlan is the Context Layer for AI: rather than maintaining separate integrations for each platform in your enterprise, you point your platform to Atlan instead.
FAQs about Azure AI Foundry vs. Google Vertex AI
1. Is Azure AI Foundry the same as Microsoft Foundry?
Yes, Azure AI Foundry was rebranded as Microsoft Foundry in November 2025. Microsoft Foundry lets you build agents, access models, and extend agents using enterprise-grade tools and knowledge with a single management plane, governance, security, and observability.
2. Is Google Vertex AI the same as the Gemini Enterprise Agent Platform?
Yes. Google Vertex AI evolved into the Gemini Enterprise Agent Platform at Google Cloud Next 2026. The core functionality of Vertex AI is now part of the Gemini Enterprise Agent Platform, which lets you build and reuse AI agents, choose a runtime, and apply enterprise governance throughout. It is a distinct product from Gemini Enterprise, the business-user-facing app Google launched in October 2025, which took over what Google had been calling Agentspace.
3. Do both Azure AI Foundry and Google Vertex AI support MCP?
Yes, both platforms support MCP servers.
4. Which agent frameworks can you use for creating agents on Azure AI Foundry and Google Vertex AI?
Azure AI Foundry allows hosted agents with Microsoft Agent Framework, LangGraph, or the OpenAI Agents SDK, among many others. Prompt agents don’t need any framework at all. Google Vertex AI uses the open-source Google ADK project, but you can use any open-source agent framework and deploy agents using Agent Runtime.
5. How can agents on Azure AI Foundry and Google Vertex AI use the Atlan MCP server?
Both platforms can use Atlan’s MCP server to get enterprise context to agents. The mechanism differs slightly, but the end goal is the same: a Foundry agent adds Atlan’s MCP server as a remote MCP tool, and a Vertex AI agent built with the ADK has built-in constructs to connect to Atlan’s MCP server as a client.
Sources
- What’s New in Microsoft Foundry, October and November 2025, Microsoft Foundry Blog (December 2025)
- Introducing Gemini Enterprise Agent Platform, Powering the Next Wave of Agents, Google Cloud Blog (April 2026)
- Gemini Enterprise Agent Platform Product Page, Google Cloud (2026)
- The New Gemini Enterprise: One Platform for Agent Development, Google Cloud Blog (2026)
- The Next Wave of Azure Innovation: Azure AI Foundry, Intelligent Data and More, Microsoft Azure Blog (November 2024)
- Google Cloud Unveils Vertex AI, One Platform, Every ML Tool You Need, Google Cloud Blog (May 2021)
- What Is Microsoft Foundry Control Plane?, Microsoft Learn
- What Is Microsoft Foundry Agent Service?, Microsoft Learn
- Agent Platform Overview, Gemini Enterprise Agent Platform, Google Cloud
- Microsoft Entra Agent ID Documentation, Microsoft Learn
- Agents Overview, Gemini Enterprise Agent Platform, Google Cloud
- Choose How to Build With Microsoft Foundry, Microsoft Learn
- What Are Hosted Agents?, Microsoft Learn
- Agent Studio Overview, Gemini Enterprise Agent Platform, Google Cloud
- OpenAI Agents SDK, OpenAI GitHub
- What Is Foundry IQ?, Microsoft Learn
- Agent Gateway Overview, Gemini Enterprise Agent Platform, Google Cloud
- The Model Hub, Hugging Face
- Enable an Agent-to-Agent Endpoint on a Foundry Agent, Microsoft Learn
- Model Garden on Gemini Enterprise Agent Platform, Google Cloud
- Agent Runtime, Gemini Enterprise Agent Platform, Google Cloud
- Vertex AI Name Changes, Gemini Enterprise Agent Platform, Google Cloud
- Connect Agents to Microsoft Fabric with Fabric IQ (Preview), Microsoft Learn
- Knowledge Catalog, Google Cloud
Azure AI Foundry vs. Google Vertex AI: Related reads
- What is an Agent Context Layer? Architecture Guide
- Agent Context Layer Design: Principles and Decisions
- Context Repository for AI Agents: A Practical Guide for 2026
- Context Engineering for Multi-Agent Systems: A 2026 Guide
- What Snowflake Cortex Needs from a Context Layer in 2026
- Genie Ontology and the Atlan Context Layer: Better Together
- Context Engineering 101: How to Build a Deep Context Layer for AI Analysts
- What is MCP (Model Context Protocol): A 2026 Guide
- What OpenAI’s Data Agent Reveals About Enterprise AI
- Gartner on Context Graphs: Top Insights, Capabilities & Implementation Recommendations for 2026
- Context Engineering for AI Governance: Complete 2026 Guide
- Semantic Layers: The Complete Guide for 2026
- Who Should Own the Context Layer: Data Teams vs. AI Teams? | A 2026 Guide
- Context Layer vs. Semantic Layer: What’s the Difference & Which Layer Do You Need for AI Success?
- What is Context Engineering: Why Does Enterprise AI Depend On It in 2026?
- Context Graph vs Knowledge Graph: Key Differences for AI
- Context Graph: Definition, Architecture, and Implementation Guide
- Context Graph vs Ontology: Key Differences for AI
- What Is Ontology in AI? Key Components and Applications
- Context Layer 101: Why It’s Crucial for AI
- Combining Knowledge Graphs With LLMs: Complete Guide
- What Is an AI Analyst? Definition, Architecture, Use Cases, ROI
- Ontology vs Semantic Layer: Understanding the Difference for AI-Ready Data
- What Is Conversational Analytics for Business Intelligence?
- Google Agentspace (Now Gemini Enterprise): What It Is, How It Works
