We attended MS Build 2025 and learned a lot about what's most important to Microsoft. In this blog post we outline the highlights. We also wrote more indepth posts about the more strategic announcements which you can read here:

Three part series on Azure AI Foundry

• Azure AI Foundry: What is it and Why Should You Care
• Azure AI Foundry – Part II:  A Developer’s Deep Dive into the Foundry
• Azure AI Foundry Part 3: Governing AI at Scale – Trust, Safety, and Turbocharged Growth

Model Context Protocol is Important and We Explain Why

Model Context Protocol: A Bridge Between Legacy Desktop Applications and Modern AI

Keynote Highlights

1. Azure AI Foundry

You may have heard of Azure AI Foundry and thinking to yourself what it is. Microsoft calls it their "Agent Factory" but it is essentially Microsoft's answer to building AI applications in the enterprise, without the complexity involved in security and regulation controls around using AI. Microsoft is investing heavily into Azure AI foundry and they are integrating all of your favorite LLM models to use (Llama, GPT4o, DeepSeek V3). 

2. There is no limit to Copilot

We just watched copilot create a solution for a user story in Github and all within the span of an hour. And all they had to do was assign the issue to their copilot agent in Github and everything was done. You can create filters so that it doesn't merge without human PR approval and safety measures that fit your needs. 

In addition, for those using VS Code, Microsoft is now open sourcing the Github Copilot Extension and moving key AI features from the extension into the core product of VS Code. This gives more transparency into how AI works and build on top of AI tools in VS Code.

3. Agentic AI is Here to Stay

MS Build is Agent, Agent, Agent all the time now. If you haven't read it yet, we have a great blog post explaining Agentic AI and how to get started. Microsoft is going all in with Agents. They shared the concept of the open agentic web, where the goal is to have every website embedded with an AI agent to complete complex tasks.

NLWeb is one great example, making it easy to effectively turn any website into an AI app to enhance your websites functionality and interactivity. 

• Azure Migrate now does full code-level assessment - GA today. Useful for scoping legacy app moves without the yak shave.
• GitHub Copilot gets “agent mode.” Refactor a multi-project solution, heal its own compile errors, and run tests - all from one prompt in VS / VSCode. Preview starts now.
• PostgreSQL extension for VS Code (public preview). First-party object explorer, IntelliSense and query history – looks like a decent PG admin inside the IDE.
• SQL Server 2025 preview + SSMS 21 (with Copilot). Native JSON, REST API, semantic search, OneLake mirroring.
• Azure AI Foundry security upgrades. Prompt Shields, PII filters, Defender for Cloud hooks – paranoid people about LLMs touching client data should be happy.
• Entra Agent ID (preview). Unique identity + access control for every Copilot/Foundry agent.
• Windows AI Foundry. Local model catalog & auto-optimized runtimes for Copilot-class PCs
• Bottom line: Microsoft is doubling-down on “agentic” everything – from DevOps to data – and Copilot is going into every tool we touch. Plenty here to accelerate our migration with AI story with customers.

Me & Cheyenne fan-girling with Anders Hejlsberg and Mads Kristensen (the "Fathers" of C# and Typescript)

Day 2: Digging In

Today, we sat in on a keynote session that unpacked the tech that was announced in the keynote on day 1. We saw some really amazing demos and watched as copilot was able to be assigned an issue in github and pull files from external sources (such as figma) to be able to update a website interface. The capabilities seem endless here. 

Some more cool features for AI Foundry:

Azure AI Foundry (aka the Agent Factory)

• You can get same time access to OpenAI models as they launch
• Model routers - selecting the best model for your prompt in real time
• Azure API management: can access APIs via an agent or app with MCP
• Model customization
• Distillation -  break larger models into smaller models and optimize for cost/performance
• A complete stack of agent capabilities and services
  • Agent Service - create, deploy and monitor agents
  • Agent Catalog - utilize a starter set of agent templates
  • Agent Knowledge - bring over data to train your agent on specific tasks

We know that keeping up with all the AI innovations these days is extremely time consuming. But one thing we've learned at MS Build is that Microsoft seems to be trying to make it easier for developers in their ecosystem and using Azure to create and optimize models for their company's needs and fast track their enterprise AI solutions. 

Got any questions about anything we've learned at MS Build? Just reach out to us or comment below!