Become a member of our daily and weekly newsletters for the latest updates and exclusive content about leading AI coverage. Leather
In the first generation of the internet, in the late nineties, searching was good but not great, and it was not easy to find things. This led to the rise of syndication protocols in the early 2000s, where Atom and RSS (really simple syndication) offered a simplified way for website owners to make the headlines and other content easily available and searchable.
In the modern era of AI there is a new group of protocols on the rise to serve the same basic goal. This time, instead of making sites easier for people to find, it’s all about making websites easier for AI. Anthropic‘S Model Control Protocol (MCP), Google‘s Agent2Agent and large language models/ llms.txt are among the existing efforts.
The newest protocol is Microsoft’s Open-source NLweb (Natural Language Web) effort, which was announced during the Build 2025 conference. NLWEB is also directly linked to the first generation of web syndication -as it was conceived and made by RV Guha, who helped in making RSS, RDF (Resource Description Framework) and schedule.org.
NLWEB enables websites to easily add AI-driven conversation interfaces, making each website effectively becoming an AI app where users can request content using a natural language. NLWEB is not necessarily about competing with other protocols; It builds on it earlier. The new protocol uses existing structured data formats such as RSS and every NLweb body functions as an MCP server.
“The idea behind NLweb is that it is a way for anyone who has a website or an API to easily make their website or their API an agentic application,” said Microsoft Cto Kevin Scott during his Build 2025 Keynote. “You can really think about it a bit like HTML for the agent web.”
How NLWEB works to make the web for companies
NLWEB transforms websites into AI-driven experiences through a simple process that builds on existing web infrastructure while the use of modern AI technologies.
Result on existing data: The system starts with the use of structured data that websites already publish, including Markup, RSS feeds and other semi-structured formats that are often embedded in web pages. This means that publishers do not have to fully rebuild their content infrastructure.
Data processing and storage: NLWEB contains tools for adding this structured data to vector databases that make efficient semantic search and collection possible. The system supports all important vector database options, so that developers can choose the solution that best fits their technical requirements and scale.
AI improvement layer: LLMS then improves this stored data with external knowledge and context. For example, when a user asks about restaurants, the system automatically places geographical insights, reviews and related information by combining the vectorized content with LLM options to offer extensive, intelligent answers instead of simple data recording.
Universal interface -creation: The result is a natural language interface that serves both human users and AI agents. Visitors can ask questions in regular English and receive conversation reactions, while AI systems can have the information from the site programmatic access and questions via the MCP framework.
With this approach, every website can participate in the Emerging Agentic Web without requiring an extensive technical overhaul. It makes AI-driven search and interaction as accessible as making a basic web page in the early days of the internet.
The emerging AI protocation landscape brings many choices for companies
There are many different protocols in the AI space; Not all do the same.
Google’s Agent2AgentFor example, it is about enabling agents to talk to each other. It is about orchestrating and communicating agent AI and is not particularly aimed at AI-ABObleStende websites or AI content. Maria Gorskikh, founder and CEO of Aia and a contribution to the Project Nanda Team at MIT, Explained to Venturebeat that Google’s A2A makes structured tasks possible between agents with the help of defined schemes and life cycle models.
“Although the Protocol is open-source and model-agent by design, the current implementations and tooling are closely linked to the Gemini Stack of Google-Waardoor The lake is a backend orchestra racker than a general interface for web-based services,” she said.
Another emerging effort is LLMS.TXT. The goal is to help LLMS with better access to web content. While it is on the surface, it may sound somewhat like NLweb, it is not the same.
“NLWEB does not compete with LLMS.TXT; it is more comparable to web sheep aids that try to distract intention from a website,” Michael Ni, VP and chief analyst at Constellation Research told VentureBeat.
Krish Arvapally, co-founder and CTO van van Dappier, Explained to Venturebeat that LLMS.TXT offers a formatting style format with training rights that help LLM-Crawlers to take content in the right way. NLWEB focuses on making real -time interactions directly possible on the website of a publisher. DapsenPier has its own platform that automatically records RSS -Feeds and other structured data and then delivers brand -discussing conversation interfaces. Publishers can syndicate their content with their data market.
MCP is the other large protocol and is increasingly becoming a de facto standard and a fundamental element from NLweb. Fundamentally, MCP is an open standard for connecting AI systems with data sources. Ni explained that according to MCP, MCP is the transport layer, where MCP and NLWeb together offer the HTML and TCP/IP from the Open Agentic Web.
Forrester Senior Analyst Will Mckeon-White sees a number of benefits for NLweb compared to other options.
“The most important advantage of NLweb is a better control over how AI systems ‘see’ the pieces that are websites, making better navigation and better understanding of tooling possible,” McKeon-White told Venturebeat. “This could reduce both errors of systems that misunderstand what they see on websites and reduce the interface.”
Early Adopters already see the promise of NLweb for Enterprise Agentic AI
Microsoft did not only throw NLweb about the proverbial wall and hoped that someone would use it.
Microsoft has already involved several organizations and uses NLweb, including public media from Chicago, Allrecipes, Eventbrite, Hearst (Delish), O’Reilly Media, TripAdvisor and Shopify.
Andrew Odewahn, Chief Technology Officer at O’Reilly Media is one of the early adopters and sees real promise for NLweb.
“NLweb uses the best practices and standards that have been developed on the open web in the last decade and makes them available for LLMS,” Odewahn told Venturebeat. “Companies have spent a long time optimizing this type of metadata for SEO and other marketing purposes, but now they can take advantage of this wealth of data to make their own internal AI Smarter and more capable with NLweb.”
In his opinion, NLWeb is valuable for companies, both as consumers of public information and publishers of private information. He noted that almost every company has sales and marketing efforts where they might have to ask: “What does this company do?” Or “What is this product about?”
“NLweb offers a great way to open this information for your internal LLMS, so you don’t have to hunt and pick it to find it,” said Odewahn. “As a publisher you can add your own metadata using Schema.org Standard and NLWEB internally as a MCP server to make it available for internal use.”
The use of NLweb is also not necessarily a heavy lift. Odewahn noted that many organizations probably already use many of the standards that NLweb trusts.
“There is no disadvantage to try it out now, because NLweb can walk completely within your infrastructure,” he said. “It is open source software meeting best in open source data, so you have nothing to lose and to win a lot by trying it now.”
Do companies now have to jump on NLweb, or wait?
Constellation Research Analyst Michael NI has a somewhat positive position on NLweb. However, that does not mean that companies must take over immediately.
Ni noted that NLweb is in the very early stages of the due date and that companies have to expect 2-3 years for substantial acceptance. He suggests that leading companies with specific needs, such as active market places, can try to take pilots with the possibility to involve the standard and help shape.
“It is a visionary specification with a clear potential, but it needs ecosystem validation, implementation tools and reference integrations before the regular Enterprise pilots can reach,” Ni said.
Others have a slightly more aggressive point of view about adoption. Gorskikh proposes to follow an accelerated approach to ensure that your company is not lagging behind.
“If you are a company with a large area of contents, internal knowledge base or structured data, it is now a smart and necessary step to stay ahead,” she said. “This is not a awaiting moment it looks more like the early acceptance of APIs or mobile apps.”
That said, she noted that regulated industries should take care of them. Sectors such as insurance, banking and health care must postpone production use until there is a neutral, decentralized verification and discovery system. There are already efforts at an early stage that tackle this-as the Nanda project at MIT in which Gorskikh participates, that builds an open, decentralized register and reputation system for agentic services.
What does this all mean for AI leaders from Enterprise?
For AI leaders of companies, NLWEB is a turning point and a technology that should not be ignored.
AI will communicate with your site and you must switch it on. NLweb is a way that will be particularly attractive for publishers, just as RSS became a must-have for all websites in the early 2000s. In a few years, users will simply expect it to be there; They will expect to be able to search and find things, while Agentic AI systems must also have access to the content.
That is the promise of NLweb.