The Agentic Web: How AI Agents Are Shaping the Web’s Future

In his new memoir, This Is For Everyone, Sir Tim Berners-Lee makes it clear he thinks AI agents are the future of the web. “In the near future, our lives will be transformed by AI ‘agents’ which will interact with the web and take actions to achieve specific goals,” he writes in the introduction. “I’ve been envisioning these types of web agents since the mid-1990s,” he adds, “I just couldn’t predict what form they would take!”

Berners-Lee created the web in 1990 and in 1994 founded its governing body, the World Wide Web Consortium (W3C). Although he’s stepped back from leadership at the W3C (he’s now listed as “Emeritus Director”), he’s still contributing ideas. In a design note published on W3C’s website and updated in January, Berners-Lee describes a personal AI assistant his company Inrupt is building, called “Charlie.” In an op-ed for the Financial Times in March, Berners-Lee describes Charlie as “my vision for a new interface between users and the web — the evolution from ChatGPT, Gemini, Pi and DeepSeek.”

“…our lives will be transformed by AI ‘agents’ which will interact with the web and take actions to achieve specific goals.”
– Sir Tim Berners-Lee, the web’s inventor

Berners-Lee is just one of many developers building AI agents with web technology. Much of this innovation is being fueled by Model Context Protocol (MCP), the now default way to bridge agents to external data sources. MCP has been the catalyst for a few fascinating projects, all released this year: MCP-UI (bringing web UIs to agents), NLWeb (turning your website into an AI chatbot), and WebMCP (giving developers a way to guide agents that visit their web apps).

The Agentic Web at W3C

Together, these three open source projects — MCP-UI, NLWeb, WebMCP — might fall under the umbrella of the “open agentic web.” That’s how Microsoft described MCP and NLWeb (the latter was created by Microsoft) back in May. The W3C appears to have adopted “agentic web” as a term, too. It has a working group, called AI Agent Protocol Community Group, that aims to “establish the technical foundations for the emerging Agentic Web.”

AI Agent Protocol Community Group homepage.

The group is co-chaired by two Chinese citizens: Gaowei Chang, creator of the Agent Network Protocol (ANP) and Song Xu from China Mobile. Aside: Chang’s ANP appears to be similar to Google’s Agent2Agent (A2A) open protocol, but it’s unclear at this point if W3C favors one over the other.

In an early draft of a report the AI Agent Protocol Community Group is working on, dated Aug. 29, 2025, an explicit comparison is drawn to a previous W3C vision of the Web: the Semantic Web. Here’s how the draft white paper starts off:

“This paper explores the evolution from the vision of the Semantic Web to the emerging Agentic Web, and analyzes the necessity of establishing standardized agent network protocols. Despite the forward-thinking concept of the Semantic Web proposed twenty years ago, it was not fully realized due to the limitations of artificial intelligence capabilities at that time. With the rapid development of modern AI technologies such as Large Language Models (LLMs), agents now possess the ability to autonomously execute tasks, perform complex reasoning, and solve multi-step problems, thus giving rise to the Agentic Web.”

In other words, the W3C sees the Agentic Web (ok, I’ll switch to capital letters) as an evolutionary shift from the mostly unrealized vision of the Semantic Web.

“AI needs to leverage the web to be useful.”
– Philippe Le Hégaret, VP of Technical Strategy at W3C

As it happens, there was a W3C event just last week that included discussion about the Agentic Web. The “WebEvolve 2025 Annual Event” was run on Sept. 5-6, at the Hangzhou Hua Jia Shan Resort in China. The host was Beihang University (a W3C Partner) and it was sponsored by Huawei.

Day 2 of the event focused on the Agentic Web and the opening presentation was by Philippe Le Hégaret, a VP of Technical Strategy at W3C. Le Hégaret, who has been with the W3C since 2001, talked about “Web AI standardization in W3C.” He noted that “AI needs to leverage the web to be useful,” but that it can also connect to applications to provide value back.

He then talked about a few W3C projects or experiments that are already bridging the Web and AI systems:

• Web Neural Network API (the current draft was published just last week), which Le Hégaret described as “a high level API […] for high performance, to be able to run LLMs on the client side.”
• Built-In AI APIs; for example for translation, summarization, and proofreading.
• WebMCP, which comes from W3C’s Machine Learning Working Group.

Later that day at WebEvolve, the co-chairs of W3C’s AI Agent Protocol Community Group — Gaowei Chang and Song Xu — delivered presentations.

Chang talked about his protocol, ANP. One slide made the claim that agents will replace software and that “personal assistants will replace the way people access the Internet.”

Gaowei Chang presentation about the ANP protocol.

Chang and Xu then jointly presented an update on their W3C community group’s progress. The presentation itself was in Chinese, but the slides include some English comments. Judging from the slides, the project is still at an early, high-level stage. The scope of the project is as follows:

• Inter-agent Communication Protocol
• Agent Identity/Authorization Model
• Standardized Metadata Format

MCP wasn’t mentioned in the slides of Chang and Xu, but it was a headline in at least two other presentations at the WebEvolve event. The description of a talk by Huawei’s Chunhui Mo made it plain how important MCP is to the Agentic Web: “Generative UI and MCP services represent the development direction of web standards in the AI era.”

Co-chairs of W3C’s AI Agent Protocol Community Group, Song Xu (left) and Gaowei Chang (right), at WebEvolve.

Conclusion

I would say the “agentic web” is still a work-in-progress at both the W3C level and in projects such as MCP-UI. The W3C does have guidance documents on its site about how AI should intersect with the web — such as a Web AI impact report published last August. But its other agentic web projects are still in their early stages; including the AI Agent Protocol Community Group.

As for open web projects built on top of MCP — such as MCP-UI, NLWeb and WebMCP — they’re still very early too. With that said, some very large companies, including Cloudflare and Shopify, are already using these technologies in their customer-facing products.

To misquote that old Web 2.0 trope (which was in turn borrowed from William Gibson), the Agentic Web is already here — it’s just not evenly distributed!

Richard MacManus is a Senior Editor at The New Stack and writes about web and application development trends. Previously he founded ReadWriteWeb in 2003 and built it into one of the world’s most influential technology news sites. From the early…

Read more from Richard MacManus