Read web / dot-com era
1990-2003. Websites primarily published information for people to consume.
What is the Agentic Web?
Richard MacManus defines the Agentic Web as the next phase of the web: AI agents become a new class of user, and websites evolve from human-only pages into capability surfaces that agents can discover, interpret, cite, and act upon.
Four eras
The article places the Agentic Web in a longer web-history arc.
Read/write web / Web 2.0
2004-2012. Blogs, wikis, social software, APIs, smartphones, and social media expanded participation.
Platform web
2013-2021. Large platforms increasingly mediated discovery, distribution, and monetization.
Agentic web / AI era
2022-present. AI systems increasingly mediate how people discover and use websites and digital products.
From pages to capabilities
Sites become data, context, and action layers for humans and AI agents.
Agentic web definition
The agentic web is an emerging phase where AI agents discover, interpret, and act on websites and digital products on behalf of users.
AI agents are a new class of user
AI agents do not browse like humans; they extract, compare, summarize, cite, and increasingly act.
Websites still matter
Websites remain canonical places to publish ideas, explain products, establish identity, build trust, and share durable sources of truth.
Shift from pages to capabilities
The web shifts from documents and destinations toward exposed capabilities such as search, retrieve, compare, subscribe, book, buy, query, configure, and act.
Website as data and context layer
Sites become sources of underlying content, structure, metadata, and capabilities that AI systems can use.
Capability surface for humans and agents
Human-facing interfaces are no longer enough; sites need agent-readable and agent-usable surfaces.
Navigation becomes delegation
Users increasingly delegate discovery and task execution to AI agents rather than directly clicking through every interface.
Model Context Protocol milestone
The article identifies Anthropic's late-2024 Model Context Protocol as a major step toward connecting agents with external tools and data.
Why websites still matter
The article argues that websites remain canonical sources of truth even as agent workflows change how they are used.
Open agentic web milestone
The article notes Microsoft's 2025 open agentic web framing and W3C activity around AI agent protocols.
Agent Experience Assessment
The source presents an assessment service for evaluating how well a website works from an agent point of view.
Primary image
The source image visualizes the four eras of the web.
Markdown alternate
The source page advertises a Markdown representation, reinforcing machine-readable publishing patterns.
FAQ
Questions and answers are named RDF resources.
What is the agentic web?
Does the article say websites will disappear?
What changes from the page model?
What are examples of capabilities?
How does the article define agentic?
What are the four eras of the web?
Read web, read/write web, platform web, and agentic web are the four eras described.
Why is MCP important in the article?
What standards activity is mentioned?
What does AI agent as user mean?
What should websites expose?
What service does Richard MacManus offer?
Glossary
Terms and definitions link into the RDF graph.
Agentic Web
AI agent
Capability surface
The agent-usable layer through which a website exposes actions, data, and context.
Model Context Protocol
A protocol introduced by Anthropic to connect AI systems with external tools and data sources.
AI Agent Protocol
Platform web
The era when large platforms mediated discovery, distribution, and monetization.
Read/write web
The Web 2.0 era of blogs, wikis, APIs, social software, and user-generated participation.
Canonical source of truth
Agent experience
How well a website or product surface can be discovered, understood, and used by AI agents.
Metadata
HowTo
A practical workflow for making a website agent-ready.
Map human and agent users
Identify what people do directly and what agents may need to do on their behalf.
Inventory site capabilities
List actions such as search, retrieve, compare, subscribe, book, buy, query, configure, and act.
Expose structured content
Improve metadata, entity descriptions, canonical URLs, feeds, and machine-readable summaries.
Add agent-facing interfaces
Provide APIs, protocol endpoints, MCP servers, or other tool surfaces where appropriate.
Preserve human trust signals
Keep clear authorship, source provenance, editorial context, and human-readable pages.
Test with agent workflows
Evaluate how agents discover, interpret, cite, and execute tasks using the site.
Iterate toward open interoperability
Track standards and adapt the site so agents can operate without fragile scraping or platform lock-in.