Join our daily and weekly newsletters for the latest updates and special content related to the industry’s leading AI coverage. Get more information
In the first generation of the web, in the late 1990s, the search was fine but not great, and things were not easy to find. This resulted in the increase in syndicating protocols in the early 2000s, which provides owners of the atom and RSS (really easy synchronization) website to make headlines and other materials easily available and easy to find.
In the modern age of AI, a new group of protocols is emerging to fulfill the same basic purpose. This time, instead of finding sites for humans, it’s all about making the website easy for AI. AnthropicModel Control Protocol (MCP), GoogleCurrent efforts include Agent 2 Agent and Large Language Model/ LLMS.Txt.
The latest protocol is an attempt by Microsoft’s open source NLWEB (Natural Language Web), which was announced during the Blood 2025 conference. The NLWEB has also been directly linked to the first generation of the standards of the web synthesis, as it was conducted by RV Goha, which helped create an RSS, RDF (Resources Detail Framework). Skima Dot Org.
NLWEB enables websites to easily add AI -powered interactions interface, effectively converting any website into an AI app where users can question the content using natural language. NLWEB is not necessarily about competing with other protocols. Rather, it is built on them. The new protocol uses existing structural data formats such as RSS, and every NLWEE example as an MCP server.
Microsoft’s CTO Kevin Scott said during his 2025 key note, “The idea behind the NLWEB is that it is a way for everyone to have a website or API that already makes its website or their API an agent application.” “You can really think about it as an agent for the web for HTML.”
NLWEB business for businesses how to make the web AI-enabled
The NLWEB transforms websites through a straightforward process in AI -powered experiments that develop the existing web infrastructure, taking advantage of modern AI technologies.
Build on existing data: This system begins with structural data that websites are already published, including markup, RSS feeds and other semi -made formats that are usually embedded in the web pages. This means that publishers do not need to be completely rebuilt their content infrastructure.
Data processing and storage: The NLWEB contains tools to add this structural data to the vector database, which enable efficient cement search and recovery. This system supports all the major options for the vector database, which allows developers to choose a solution that fits their technical needs and scale best.
AI EVERY Layer: LLMs then increase this stored data with external knowledge and context. For example, when the user inquires about the restaurant, the system automatically lays to provide a comprehensive, intelligent response rather than a simple data retrieval instead of recovering simple data by linking the vector -related content to geographical insights, studies and related information.
Universal interface created: The result is a natural language interface that serves both human consumers and AI agents. Visitors can ask questions in simple English and receive conversation response, while the AI ​​system can access site information through the MCP framework and inquire about them.
This approach allows any website to participate in the emerging agent web without the need for extensive technical boundaries. It makes the AI ​​-powered search and interaction as much accessible as it was to create a basic web page in the early days of the Internet.
Emerging AI protocol brings many choices to landscape businesses
There are a lot of protocols emerging in the AI’s place. Not all do the same thing.
Google Agent 2 AgentFor example, agents have to be able to talk to each other. This agent is about AI’s archerystatry and communication and is not focused on existing websites or AI content in particular e-billing existing websites. Maria Gorski, Founder and CEO AIA And auxiliat in it Project Nanda Team in MIT, Venture Bat explained that Google’s A2A enables a structural task to pass between agents using schemes and life cycle models.
He said, “Although it is a model-ugonostic by the protocol open source and design, its existing implementation and tooling are closely tied to Google’s Gemini stack.
There is another emerging effort llms.txt. The purpose is to help LLMS access better web content. While living on the surface, it may look like NLWeb, it is not the same thing.
“Nlweb does not compete with LLMS.TxtXT. It is more comparable to the web scraping tools that try to reduce the intention from a website,” said Burj Research VP and Principal Analyst told Venture Bet.
Crush Umply, co -founder and CTO Deepier, Venture Bet explained that LLMS.Txt provides Mark Down Style format with training permission that helps LLM crawlers properly ignite content. NLWEB focuses on a publisher’s website directly to activate the real -time conversation. DAPPierce has its own platform that automatically consumes RSS feeds and other structural data, then provides an interface of the branded, embedded conversation. Publishers can syndicate their content in their data market.
MCP is the second largest protocol, and it is rapidly becoming the main factor of de facto standard and NLWEB. Basically, the MCPAI system is an open standard for connecting with data sources. Ni explained that Microsoft believes, the MCP is a transport layer, where the MCP and NLWEB Open Agent Web provide HTML and TCP/IP of the web.
The forester senior analyst Will McCain White will see many benefits for NLWEB than other options.
“The main advantage of NLWEB is better control over how the AI ​​system builds a website, which allows better navigation and more complete understanding about tooling,” McCain White told Venture Bat. “This can reduce the misunderstandings that they are watching on the websites, as well as reduce the interface restoration.”
Early adoptions are already watching NLWEB promise for Enterprise Agent AI
Microsoft just did not throw NLWeb on the wall of the idiom and hopefully someone will use it.
Microsoft already has several organizations and uses NLWEBs, including Chicago Public Media, Allerosis, Event Bright, Hurst (Delsh), and Rally Media, Trip Advisor and Shophaif.
Andrew Odohan, Chief Technology Officer of the Rally Media, is among the early adoptions and views the real promise for the NLWEB.
“The NLWEB takes advantage of the excellent methods and standards developed over the past decade on the Open Web and makes them available to LLM,” Odohan told Venture Bat. “Companies have long spent time improving such meta data for SEO and other marketing purposes, but now they can benefit from this wealth to improve their internal AI and make it more capable with NLWeb.”
In his view, NLWEB is valuable as both public information users and private information publishers for businesses. He said that almost every company has sales and marketing efforts where they may need to ask, “What does this company do?” Or “What is this product about?”
“Nlweb gives your internal LLMS a great way to open this information so you don’t need to hunt and nasty to find it,” Odohan said. “As a publisher, you can add your metal data using Skima Dot Origin Standard and use NLWEB as a MCP server to make it available for internal use.”
The use of NLWEB is not necessary to have a heavy lift. Odohan noted that many organizations are probably using many standards that NLWEB depends on.
He said, “There is no shortage of trying it now because NLWEB can run your infrastructure completely.” “This is an open source software that matches the best in open source data, so you have nothing to lose and now get a lot from trying it.”
Should the enterprises jump on NLWEB right now, or should wait?
Michael Naye, an analyst, has a positive point of view on NLWEB. However, this does not mean that businesses need to adopt it immediately.
Ni noted that the NLWEB is in the early stages of maturity and that businesses should expect 2-3 years to adopt anything. He recommends that leading companies with specific requirements, such as in active markets, can find a pilot with the ability to create and help quality.
“This is an insightful specification with a clear ability, but before reaching the mainstream enterprise pilots, it needs environmental system verification, implementation tolling, and reference integration.”
There is a much more aggressive idea about adoption of others. Gurkskh suggests that your enterprise not be left behind to adopt a speedy approach.
“If you are an enterprise with a large number of content levels, internal knowledge base, or structural data, NLWeb is now a smart and necessary step to pilot it,” he said. “This is not a waiting and watching moment-it is mostly like adopting APIS or mobile apps.”
He said that regular industries needed to run carefully. Sectors such as insurance, banking and health care should stop the use of production unless there is a neutral, decentralized certification and discovery system. Earlier, efforts are being made for an early phase-such as the Nanda project in the MIT, which has participated in the Gorask, which is creating an open, decent registry and credibility system for the services of agents.
What does all this mean for Enterprise AI leaders?
For Enterprise AI leaders, NLWEB is a water shedding moment and a technology that should not be ignored.
AI is going to communicate with your site, and you need AI. NLWEB is a method that will be especially attractive to publishers, as must be for all websites like RSS in the early 2000s. In a few years, consumers will only expect it. They will expect them to find and find things, while the Agent AI system will also need to be able to access the content.
This is the promise of NLWEB.
