Everybody knows that Microsoft bought OpenAI a long time ago and now Microsoft should have the most advanced infrastructure for deploying AI solutions and SOTA models, right?

Well, why don’t we hear about them that much? What do they have in that space?

Let’s find out.

First of all Microsoft released a whole bunch of things, with very confusing naming as usual.

There are: Microsoft Semantic Kernel, Azure AI Foundry, Azure AI studio, Azure OpenAI Service, Azure AI Services, Phi series of models, Copilot, Power Virtual Agents, Copilot Studio, Autogen and so on. Some of them kinda do the same thing, some are rebrandings of the other things, some of them are open source and not actively supported. In short, it is a complete zoo and a mess of names, apparently so that potential adversaries could not figure out what exactly each product is for and could use them. That worked very well, congratulations Microsoft! Indeed you wouldn’t hear much about that stuff, who would want to dig through that pile.

Microsoft and OpenAI: The Real Relationship

Microsoft’s partnership with OpenAI began with initial investments and has grown substantially, now reportedly totaling around $13 billion. While this significant financial backing gives Microsoft considerable influence, it doesn’t mean ownership. The partnership has been described as having “revenue sharing agreements that flow both ways, ensuring that both companies benefit from increased use of new and advanced AI offerings,” according to Microsoft’s official statements.

However, recent reports suggest this relationship may be showing signs of strain. As noted by various tech publications, tensions have been building between the two companies since at least mid-2023, with some describing the partnership as “turning sour.” This shift appears to coincide with OpenAI’s development of competing products and Microsoft’s efforts to build its own in-house AI models. That doesn’t surprise me at all.

Azure AI Platform Components

Microsoft’s Azure AI ecosystem consists of several key components:

• Azure AI Foundry — A unified platform for enterprise AI operations, model builders, and application development. Launched in late 2024, it provides everything needed to customize, host, run, and manage AI-driven applications with seamless integration to GitHub, Visual Studio, and Copilot Studio.
• Azure OpenAI Service — Offers industry-leading AI models from OpenAI within the secure, enterprise-grade Azure environment. This service provides access to models like GPT-4, GPT-3, Codex, and DALL-E with added security features, regional compliance, and private networking.
• Azure AI Model Catalog — A repository of foundation models from various creators including Microsoft, OpenAI, Mistral, Meta, and Stability AI, enabling organizations to discover, evaluate, customize, and deploy AI models.
• Azure AI Services — A suite of cloud-based artificial intelligence services including Vision AI, Translator, Speech recognition, and other cognitive services that help developers create intelligent applications.
• Microsoft’s Homegrown Models — Microsoft has been developing its own AI models, most notably the Phi series.
• Autogen — an open-source framework to build agentic systems (similar to CrewAI). Its purpose is experimentation and creating prototypes quickly, rather than anything serious.
• Semantic Kernel — also a multiagent open source framework, but Semantic Kernel is designed to integrate large language models into enterprise applications. It is backed by Microsoft’s customer support services.

What is Azure AI Foundry?

Azure AI Foundry is Microsoft’s comprehensive platform for AI development, deployment, and management, designed to streamline the creation of intelligent applications and agents. Formerly known as Azure AI Studio, it was rebranded in late 2024 to reflect its expanded capabilities and Microsoft’s vision for a more robust AI development ecosystem.

The platform serves as a unified hub where developers, data scientists, and business stakeholders can collaborate to build, deploy, and manage AI solutions using Microsoft’s extensive AI infrastructure and a wide range of models from Microsoft and third-party providers.

Key Components and Features

1. Development Environment

Azure AI Foundry provides a cohesive environment for AI development through:

• Web Portal: A browser-based interface for exploring models, building projects, and managing AI resources.
• SDK: A comprehensive software development kit that allows developers to build and deploy AI applications programmatically.
• API Integration: Standardized APIs that enable seamless integration with existing applications and workflows.
• Evaluation Framework: Built-in tools for assessing model performance and application quality.
• Deployment Options: Streamlined processes for deploying applications to various environments.

This unified approach eliminates the need to navigate multiple tools and services, significantly streamlining the AI development process.

2. Model Catalog

• Microsoft Models: Including the Phi series (Microsoft’s efficient small language models), Azure OpenAI models, and specialized models for specific tasks.
• Third-Party Models: Models from leading AI companies such as OpenAI, Mistral, Meta and so on.
• Model Collections: Curated sets of models organized by provider, task, or industry to simplify model selection.

The catalog includes models for various tasks such as text generation, code generation, image creation, multimodal understanding, and specialized industry applications.

3. Advanced Agentic Capabilities

• AI Agent Service: A fully managed service for building, deploying, and scaling AI agents that can perform complex tasks.
• Computer-Using Agent (CUA): A groundbreaking capability announced in March 2025 that allows AI agents to interact directly with software interfaces, enabling them to navigate websites and use desktop applications.
• Multi-Agent Systems: Support for creating systems of specialized agents that can collaborate to solve complex problems.

4. Integration with Data Sources

For AI applications to be effective, they need access to relevant data. Azure AI Foundry facilitates this through:

• Data Connections: Secure connections to various data sources, including Azure Storage, SharePoint, OneDrive, and more.
• Knowledge Bases: Tools for creating and managing knowledge repositories that can ground AI responses in enterprise data.
• Semantic Indexing: Advanced indexing capabilities for efficient retrieval of relevant information.

While our previous discussion focused on Microsoft’s AI infrastructure and models, one critical component of Microsoft’s AI ecosystem deserves special attention: Microsoft Copilot Studio.

What is Microsoft Copilot Studio?

Microsoft Copilot Studio is an end-to-end conversational AI platform that empowers organizations to create, customize, and manage AI agents using either natural language instructions or a graphical interface. It evolved from Microsoft’s earlier Power Virtual Agents product, and I have to agree — Power Virtual Agents was a crappy name.

Copilot Studio serves as both a creation platform for building custom AI agents and a management system for deploying and monitoring these agents across an organization. It represents Microsoft’s vision for democratizing AI development, allowing non-technical users to create sophisticated AI solutions while providing advanced capabilities for professional developers.

Key Capabilities of Copilot Studio

1. Low-Code Development Environment

Copilot Studio offers a graphical development interface that makes it more accessible to business users, not just professional developers.

2. Autonomous Agent Capabilities

One of the most significant recent advancements in Copilot Studio is its autonomous agent functionality, which became generally available in March 2025. These autonomous agents can:

• Automatically respond to signals and events across business systems
• Initiate tasks and workflows without human intervention
• Learn from interactions to improve performance over time
• Escalate complex issues to human operators when necessary

As Microsoft explained in their announcement: “Autonomous agents dramatically save time for individuals, teams, and entire organizations by automating business processes and getting tasks done with minimal human intervention.”. Typical marketing bullshit of course as no agentic system right now is adequate enough to trust it with mission critical or complex tasks.

3. Integration with Microsoft Ecosystem

Copilot Studio is deeply integrated with Microsoft’s broader ecosystem, allowing agents to connect with:

• Microsoft 365 applications (Outlook, Teams, SharePoint, etc.)
• Dynamics 365 for CRM and ERP functionality
• Power Platform for advanced automation and analytics
• Azure services for enhanced capabilities and integrations
• Third-party applications through connectors and APIs

4. Enterprise-Grade Security and Governance

For enterprise deployments, Copilot Studio provides robust security and governance features:

• Role-based access control
• Data loss prevention policies
• Compliance with industry regulations
• Audit logging and monitoring
• Integration with Microsoft’s security frameworks

The Latest Innovation: Computer Use Feature

Not long ago, Microsoft announced a groundbreaking new capability for Copilot Studio called “Computer Use.”, allowing agents to directly interact with user interfaces across desktop and browser applications.

With Computer Use, Copilot Studio agents can:

• Navigate websites and web applications
• Input data into forms and fields
• Extract information from on-screen elements
• Interact with desktop applications
• Perform sequential tasks across multiple applications

As Microsoft explained in their announcement: “With computer use in Copilot Studio, makers can build agents that automate tasks on user interfaces across both desktop and browser applications, enabling end-to-end automation scenarios.”. It works for very simple things only though, but the idea is to improve them over time.

What is Semantic Kernel?

Microsoft Semantic Kernel is an open-source SDK that allows to build, orchestrate, and deploy AI agents and multi-agent systems. It serves as a lightweight, flexible framework that enables seamless integration of modern AI models and services into custom applications. So it is, basically, Microsoft’s version of LangChain.

Semantic Kernel provides a specialized AzureAIAgent component that directly integrates with Azure AI Agent Service, allowing developers to leverage Azure AI Foundry’s agent capabilities through Semantic Kernel code, and in general it is integrated into Azure AI infrastructure natively, although it doesn’t save it from lots of bugs and inconsistent behavior so even if you use it — it won’t completely revive you from integration headache. Still, if you intend to use Microsoft stack — it is an obvious choice.