Microsoft Build 2025: Supporting the Open Agentic Web

Reading Time: 9 minutes

Prefer watching instead of reading? Watch the video here. Prefer reading instead? Scroll down for the full text. Prefer listening instead? Scroll up for the audio player.

P.S. The video and audio are in sync, so you can switch between them or control playback as needed. Enjoy Greyhound Standpoint insights in the format that suits you best. Join the conversation on social media using #GreyhoundStandpoint.

---

In 2025, the open web has a new language. At Microsoft Build, the company introduced a new protocol called NLWeb, designed to bring natural language interfaces directly to websites. The ambition? Nothing less than giving the agentic web its own HTML moment.

NLWeb allows developers to embed natural language “instructions” into their websites using just a few lines of code, enabling AI agents and models to parse, understand, and interact with web content in semantically rich ways. Unlike traditional APIs, which require developers to predefine rigid endpoints, NLWeb is a declarative, model-agnostic approach that lets websites expose intent and context naturally. Any compatible model — open, proprietary, cloud-hosted, or local — can consume these interfaces. This makes the web not just readable by machines, but usable by agents.

It’s a big shift. Just as HTML turned the internet from a document archive into an interactive canvas, NLWeb could transform static web experiences into dynamic, agent-navigable systems. For enterprises, this means that user interfaces — forms, dashboards, portals — could become fluid conversation spaces. Not chatbots bolted to the corner, but structured, discoverable, and secure conversational surfaces at the protocol level.

Greyhound CIO Pulse 2025 data shows 59% of digital leaders globally are now prioritising “agent-ready architecture” as a core tenet of their next-gen web strategies. It’s not about building new apps — it’s about making current ones discoverable, traversable, and responsive to AI-native experiences. And for that, existing HTML and JavaScript paradigms fall short.

One Greyhound Fieldnote from a Head of Digital Experience at an e-commerce leader captured the shift: “We don’t want ten new copilots layered onto our site. We want our site to be legible to the agents our customers already use. NLWeb gives us a way to speak their language — without rewriting the whole frontend.”

Microsoft’s move here reflects a deeper philosophy: the agentic web shouldn’t be vendor-controlled or locked behind opaque APIs. It should be open, inspectable, and future-compatible. NLWeb, if adopted broadly, could create a common semantic layer across the web that both humans and agents can understand, making the web not just clickable, but conversational by design.

Greyhound Standpoint: At Greyhound Research, we believe NLWeb is one of the most quietly profound announcements to come out of Build 2025. Not because it demos well, but because it gives the web its missing abstraction layer for AI. This isn’t another tool. It’s a grammar for the agentic age.

From Static Pages to Semantic Interactions: How NLWeb Changes the Developer Model

If NLWeb sounds like just another markup trick, don’t be fooled. What it represents is a decisive shift in how developers think about web interaction — from hardcoded actions and rigid DOM structures to intent surfaces designed for agents.

For decades, web development has centered around optimizing for the human browser. User journeys, page load times, visual hierarchy — all tuned for eyeballs and mouse clicks. NLWeb redefines that foundation. It asks developers to build not just for users, but for conversational agents that can interpret, reason, and act — without ever rendering a pixel.

The result is a new kind of design contract. Rather than writing code for specific interfaces, developers declare what the page means — what an action does, what data is exposed, how it should be interpreted — and allow AI agents to infer workflows dynamically. It’s HTML meets intent-based computing, tuned for agent consumption.

The beauty lies in the simplicity. With only a few lines of NLWeb markup, developers can make existing websites semantically explorable by any compliant model. There’s no need to overhaul entire architectures. Just like HTML allowed any browser to read a site, NLWeb allows any agent to understand and interact, regardless of the underlying tech stack.

Greyhound Fieldnotes from multiple digital product teams point to a similar pain: the current AI integration path is too expensive and too brittle. One Head of Engineering at a global logistics firm noted: “Today, if we want agents to interact with our apps, we have to expose APIs, define schemas, get security reviews, and teach the model our domain. With NLWeb, we can let the page speak for itself.”

According to Greyhound CIO Pulse 2025, 64% of enterprises currently experimenting with GenAI agents cite integration complexity as their top barrier to scaled adoption. And 48% say they don’t have the resources to “API-ify” their entire legacy stack. NLWeb offers a lifeline — a thin, declarative layer that brings AI interaction within reach, without requiring a ground-up rebuild.

What makes Microsoft’s move here compelling is that it’s not just launching a new protocol — it’s encouraging a new developer mindset. One where websites are not just visual containers, but semantic surfaces. One where functionality is not buried in JavaScript, but exposed clearly for agents to reason over.

Greyhound Standpoint: At Greyhound Research, we believe this shift is long overdue. If AI agents are to become real web participants — not just extensions of proprietary platforms — then the web must start speaking their language. NLWeb is the first serious attempt to make that possible. It turns the web into a place where models can do more than read. They can understand.

MCP and the Case for Agent-Aware Standards Across the Stack

If NLWeb is the interface layer for the open agentic web, the Model Context Protocol (MCP) is its operating logic — the scaffolding that ensures agents understand who they are, what they’re doing, and why it matters. With first-party support for MCP now embedded across GitHub, Copilot Studio, Dynamics 365, Azure AI Foundry, Semantic Kernel, and Windows 11, Microsoft is taking a firm position: conversational agents must be context-aware by design.

In simple terms, MCP enables agents to carry and reference structured context — identity, history, task scope, permissions — across tools, clouds, and workflows. Think of it as the equivalent of browser cookies, session tokens, and user metadata — but for autonomous systems built on language models. Without it, agents become stateless, forgetful, and unpredictable. With it, they become accountable, explainable, and orchestrable.

Microsoft isn’t just implementing MCP inside its own ecosystem — it’s joining the MCP Steering Committee alongside GitHub and contributing new capabilities to the broader open standard. That matters. Because as the AI agent space accelerates, the real risk isn’t runaway models — it’s runaway fragmentation. Agents built in isolation won’t scale. But agents built on shared context protocols can reason, coordinate, and comply in structured environments.

Greyhound CIO Pulse 2025 reveals a revealing split: 51% of enterprises already using multiple agents report problems with task confusion, duplicated effort, and conflicting instructions. Why? Because each agent is working from its own frame of reference, unaware of the broader business or workflow scope. MCP offers a way out — a lightweight, pluggable standard that agents can use to “think” in context, not just generate in isolation.

A Greyhound Fieldnote from a Director of Automation at a global financial services firm said it best: “We’re not short on copilots. We’re short on context. If MCP can help us stop reinventing onboarding, approvals, and handoffs with every new agent, we’re in.”

Importantly, Microsoft is taking a platform-neutral stance. MCP is being pushed not as a proprietary framework, but as a vendor-agnostic mechanism for interoperability. It’s designed to work across clouds, LLMs, and orchestration stacks, which means enterprises won’t be forced to standardize on a single toolchain just to make their agents coherent.

Greyhound Standpoint: At Greyhound Research, we see this as a rare but critical intervention. Too much of the GenAI market is focused on power — bigger models, faster inference, more tasks. But what’s missing is discipline. MCP offers that discipline. It introduces a shared memory for agents — a fabric of understanding that lets them carry knowledge responsibly from one action to the next.

Trust, Traceability, and Windows 11: Making the Agentic Web Secure by Design

In a world rapidly filling with autonomous agents, Microsoft is making a deliberate move to bake safety into the core experience, not as an afterthought, but as a foundational design principle. With its latest updates to Windows 11, Microsoft is extending support for the Model Context Protocol (MCP) all the way down to the operating system layer, effectively making trust and traceability default attributes of the agentic web.

This isn’t just a technical update. It’s a strategic stance on how agent-driven computing should work — with transparency, governance, and auditability built in from the first prompt. According to Microsoft’s announcement on securing the protocol, agents that run on Windows 11 will now benefit from system-level protections, including sandboxing, memory isolation, and attestation hooks that ensure agents can’t overstep their intended scope without being flagged or blocked. It’s a zero-trust mindset applied to the emerging AI workforce.

What makes this move significant is that Microsoft isn’t focusing only on the cloud. By embedding these safeguards into Windows — still the most widely deployed enterprise endpoint in the world — the company is closing one of the biggest gaps in the AI lifecycle: the edge. For many enterprises, agents won’t live purely in data centers. They’ll be deployed on devices, inside apps, alongside users. And without protections at that level, all the cloud governance in the world doesn’t matter.

Greyhound CIO Pulse 2025 shows a growing concern among IT leaders: 67 percent now believe agent execution at the edge — whether on mobile, desktop, or browser — represents the greatest blind spot in their AI risk posture. Many have already seen agents go rogue, run recursive loops, or access files they were never intended to. What they’re missing is enforcement that travels with the agent.

A Greyhound Fieldnote from a CISO at a large public sector agency captured this tension: We’ve spent years locking down the cloud, only to find that GenAI is now leaking risk back into the endpoint. We need the OS to act like a co-governor, not just a runtime.

By aligning Windows 11 with MCP and broader agent safety protocols, Microsoft is stepping into that co-governor role. It’s building an architecture where agents aren’t just powerful — they’re visible, verifiable, and stoppable. That’s what enterprises need now more than ever: control without compromise.

Greyhound Standpoint: At Greyhound Research, we see this not just as an operating system enhancement, but as a redefinition of what a modern OS needs to be in the age of ambient AI. It’s no longer just a platform for applications. It’s a perimeter — one that must be intelligent, policy-aware, and unafraid to say no when agents overreach.

A New Web Contract: Why Enterprise AI Needs Open, Composable, Governable Agents

It’s tempting to frame all these announcements — NLWeb, MCP, Windows 11 integration — as technical progress. But to stop there would be to miss the larger story. Microsoft isn’t just shipping protocols and features. It’s proposing a new web contract. One where conversational agents are not just extensions of proprietary platforms, but first-class participants in the open digital economy.

For decades, the web has evolved around the human user: interfaces designed for clicks, forms, visuals, and sessions. But as agents begin to play a more central role in how people and systems interact, that architecture starts to look increasingly brittle. Static endpoints, closed ecosystems, and siloed logic no longer cut it. Enterprises need agents that can compose workflows on the fly, navigate services autonomously, and respect governance boundaries without manual babysitting.

The only way that vision scales is with open protocols — a lingua franca for AI systems that aren’t controlled by one vendor, or one model, or one deployment topology. This is why NLWeb and MCP matter. They’re not just Microsoft’s attempt to own the next wave of standards. They’re a deliberate step toward making the agentic web composable, observable, and safe by default.

Greyhound CIO Pulse 2025 shows that 73 percent of enterprise technology leaders now view closed AI ecosystems as a top-three adoption blocker. Not because the technology isn’t good, but because the lock-in is too high, the context-sharing too fragile, and the integrations too costly. What these leaders want isn’t more copilots. It’s portable, interpretable, reusable intelligence. And for that, standards aren’t nice to have — they’re existential.

A Greyhound Fieldnote from the Chief Digital Officer at a global manufacturing company summed it up: We’re already drowning in copilots. What we need is a way to make them interoperable, auditable, and accountable, without starting from scratch every time we onboard a new use case.

Microsoft’s moves at Build 2025 mark a conscious shift in that direction. Rather than keeping the agent economy trapped inside its own walls, it’s seeding the foundations for a wider ecosystem. One where websites don’t just surface data — they expose intent. One where models don’t just guess — they inherit context. One where the OS doesn’t just execute — it enforces trust.

Greyhound Standpoint: At Greyhound Research, we believe this open, agentic web won’t just be a technical achievement. It will be a strategic differentiator for enterprises navigating the next decade of digital transformation. Because the question isn’t whether AI will reshape the web. The question is whether it will be built to scale safely, across real-world complexity. And for that, open beats closed — every time.

Copyright Policy. All content contained on the Greyhound Research website is protected by copyright law and may not be reproduced, distributed, transmitted, displayed, published, or broadcast without the prior written permission of Greyhound Research or, in the case of third-party materials, the prior written consent of the copyright owner of that content. You may not alter, delete, obscure, or conceal any trademark, copyright, or other notice appearing in any Greyhound Research content. We request our readers not to copy Greyhound Research content and not republish or redistribute them (in whole or partially) via emails or republishing them in any media, including websites, newsletters, or intranets. We understand that you may want to share this content with others, so we’ve added tools under each content piece that allow you to share the content. If you have any questions, please get in touch with our Community Relations Team at connect@thofgr.com.