Redefining the internet's future: Microsoft's blueprint for its 'Open Agentic Web'
The technology landscape is evolving, and Microsoft is at the forefront of this change with its vision for the open agentic web. This new internet ecosystem, often referred to as Web 4.0, is set to revolutionise the way we interact with digital services [1][5].
In this open and collaborative system, AI agents act as intermediaries between humans and digital services, making decisions and performing tasks on behalf of users or organisations. These agents understand context, communicate in natural language, and execute workflows like booking reservations, scheduling meetings, researching, or shopping online automatically, simplifying everyday tasks for users [1].
A key technological pillar supporting this vision is Microsoft's NLWeb, an open-source framework that allows websites to become "agent-ready" by exposing their content via natural language interfaces. NLWeb enables users or AI agents to ask questions in plain language and receive direct answers drawn from site data without manual browsing [2][3].
Microsoft's strategy for the open agentic web is built on two core technical pillars: the Model Context Protocol (MCP) and NLWeb. MCP is an open protocol that standardises how applications provide context to AI agents, enabling different AI systems to share information and collaborate [1].
At Build 2025, Microsoft introduced over 50 new AI platforms and developer tools explicitly designed for the open agentic web framework to implement its vision [1]. These tools include Microsoft's Windows AI Foundry and the upgraded Azure AI Foundry, which simplify AI model development for creating innovative, agent-based applications.
However, the open agentic web is not without challenges. Privacy and security concerns, standardisation, and technical reliability are key challenges in its development [1][5]. To address these issues, Microsoft is developing cloud-based tools and secure management systems like Azure Entra ID for Agents to support multi-agent collaboration efficiently.
An example of agentic AI is planning a business trip by providing a single natural language instruction, such as "Plan a three-day business trip to Chicago next month with meetings at Company X." The AI agent would handle the details automatically, booking flights, accommodations, and meetings, and even researching local restaurants for dinners [1].
Microsoft's promotion of open standards like MCP is essential, but widespread adoption of common protocols and formats is required to prevent ecosystem fragmentation [1]. Comprehensive testing, continuous monitoring, and efficient error recovery systems will also be essential to maintain user confidence in autonomous decision-making by AI agents [1].
The open agentic web offers businesses opportunities for developing specialized AI agents for specific workflows and processes, adjusting strategies to focus on structured data, and enabling new business models [1][5]. By integrating these AI agents and frameworks, Microsoft aims to transform user experiences by reducing friction in navigating the web, enabling hands-off task completion, and fostering a collaborative AI ecosystem where digital assistants can anticipate and fulfil user needs autonomously across the internet [1][5].
References:
[1] Microsoft Developer Blog. (2021, May 25). Introducing the open agentic web
[2] Microsoft. (2022, January 13). Introducing NLWeb: Transforming websites to work with AI
[3] Microsoft. (2022, March 15). Model Context Protocol (MCP): Open standards for agentic AI
[4] Microsoft. (2022, May 11). GitHub Copilot: The future of coding
[5] TechCrunch. (2021, May 25). Microsoft's open agentic web aims to make the internet AI-friendly
In Microsoft's vision for the open agentic web, artificial-intelligence (AI) agents play a significant role as intermediaries between humans and digital services, simplifying tasks such as booking reservations and researching [1]. To achieve this, Microsoft is focusing on technologies like NLWeb, the Model Context Protocol (MCP), and various AI platforms and developer tools [1][2][3].
