Microsoft's Latest Technical Vision for AI and Agents

Last week's Microsoft Build conference—Microsoft's annual developer gathering—provided some fascinating insights into their strategic direction with AI and AI agents.

This is a bit of a technical post on the plumbing or infrastructure of AI Agents, but it was a developer conference after all, so technical language is to be expected :) I found spending a bit of time with the technical concepts helped me understand the opportunities and challenges with agents, which is always a good thing.

The Big Picture: An Open Agentic Web

Microsoft's latest vision for AI and agents is ambitious:

We envision a world in which agents operate across individual, organizational, team and end-to-end business contexts. This emerging vision of the internet is an open agentic web, where AI agents make decisions and perform tasks on behalf of users or organizations.
— The age of AI agents and building the open agentic web - The Official Microsoft Blog

Side note: I'm having a visceral reaction to the word 'agentic' and now 'agentic web' as a term. Frankly, it makes me want to vomit. I think this matters because the language we use to describe these technologies shapes how we think about them. "Agentic web" sounds like something happening to us rather than for us. It suggests a world where faceless agents operate according to their own logic, not necessarily ours. I hope this doesn't take off as a broadly adopted term, but that ship might have sailed. Sigh.

The Foundation: Building the Plumbing

At Build, Microsoft focused heavily on infrastructure and the "plumbing" that enables AI models and services to communicate effectively. They're advancing two critical projects:

1. Model Context Protocol (MCP)

Microsoft is going all-in on MCP support:

Microsoft is delivering broad first-party support for Model Context Protocol (MCP) across its agent platform and frameworks, spanning GitHub, Copilot Studio, Dynamics 365, Azure AI Foundry, Semantic Kernel and Windows 11. In addition, Microsoft and GitHub have joined the MCP Steering Committee to help advance secure, at-scale adoption of the open protocol and announced two new contributions to the MCP ecosystem.

What is MCP?

Think of it as a connector between Large Language Models and the broader digital world. MCP allows models to break free from their training constraints by connecting to websites, APIs, and executing tasks across systems like email, chat platforms, and calendars. This connectivity is what makes truly autonomous agents possible.

It seems really simple...and it is! It's a language layer that lets data flow across systems, just like HTTP does on the web.

This protocol will make it easier for all the AI agents to vacuum up context it needs, and update other systems when it’s told to. Cool…but also a little scary…😨

2. NLWeb: The HTML of the Agentic Web

Microsoft's second major open project is NLWeb. They believe NLWeb will play a similar role as HTML does on the web. They say:

NLWeb makes it easy for websites to provide a conversational interface for their users with the model of their choice and their own data, allowing users to interact directly with web content in a rich, semantic manner.
— Introducing NLWeb: Bringing conversational interfaces directly to the web - Microsoft News

The implications for web publishers are significant:

Just like the introduction of HTML made it easy for almost anyone to create a website, we want NLWeb to make it easy for any web publisher to create an intelligent, natural language experience for their site. And as the agentic web (and economy) continue to grow, NLWeb will empower web publishers to participate on their terms, ensuring their website is ready to interact, transact and be discovered by other agents if they choose.

This means your organization’s website, your personal website, or Substack may share its content to AI models and systems with NLWeb. The website will just support it.

Major websites and players are already contributing to building this out including Shopify, Eventbrite, AllRecipes, TripAdvisor and O'Reilly. Let’s see if this becomes the standard of choice!

Microsoft's Strategic Play

Microsoft is making a clear bet on an open AI ecosystem. They're supporting open protocols in their developer tools to build an agent marketplace that anyone can plug into with any model, which is a stark contrast to the walled gardens favored by other AI companies.

Consider the iOS App Store model: developers must play by Apple's rules and surrender whatever cut Apple demands. Microsoft appears to be deliberately avoiding such closed technology ecosystems where benefits concentrate among a few players.

Why Microsoft Cares About AI Infrastructure

Their near-term strategy involves encouraging startups, founders, and developers to adopt Microsoft's tooling such as Windows, Azure, and Azure AI Foundry for building agentic solutions.

In addition, better protocols will lead to better agents to embed into our lives. Remember, AI models and agents are only as powerful as the knowledge and tools they can access.

As Kevin Scott, Microsoft's CTO, explains:

If agents are going to be useful, they have to take action on your behalf. They have to be able to use tools and make changes in systems and consult information sources that are diverse and rich...You've gotta figure out how to plumb things through where agents can talk to those things.
— Kevin Scott on The Future of Programming, AI Agents, and Microsoft's Big Bet on the Agentic Web (YouTube)

This infrastructure investment enables more sophisticated agents and facilitates a fundamental shift in how we interact with AI.

From Synchronous to Asynchronous AI Interaction

Today's AI interactions are primarily synchronous: you submit a prompt, wait for a response, then act on the output. Microsoft and others envision an asynchronous future where agents perform complex tasks independently over extended periods while you focus on other activities.

Imagine setting an agent to work on a project before you sleep, then reviewing its progress and results the next morning. This shift requires robust infrastructure which is the very plumbing Microsoft is building.

How does all this hit me?

I'm cautiously optimistic about Microsoft's open model approach. Common infrastructure could indeed boost competitiveness and innovation, potentially spreading AI benefits more broadly.

However, I also remain skeptical. Openness doesn’t mean what it used to. The internet was designed to be open, yet it evolved into a landscape of advertising noise, paywalls, and pervasive tracking. Security and privacy concerns persist, requiring tools like VPNs and ad blockers to navigate safely. This infrastructure for agents could also lead to tracking, security and privacy issues too.

Still, the potential for increased innovation through shared infrastructure makes this an fascinating development to watch. Microsoft's bet on openness could reshape the AI landscape...and it could also repeat the internet's mixed legacy of promise and problems.