The emerging transformation of the world wide web from a platform designed for human eyes into one navigated and acted upon by AI "agents"—software that gives large language models tools to perform tasks, not just generate text. The concept traces back to 1999, when Sir Tim Berners-Lee, the British computer scientist who invented the web in 1989, imagined an intelligent version of his creation in which much of daily life would be handled by "intelligent agents": machines able to read, interpret and act.
A central challenge is giving agents a standardised way to communicate with online services and with each other. Three key standards are emerging:
Model Context Protocol (MCP), developed by Anthropic. Rather than integrating each application on a case-by-case basis, MCP provides a shared set of rules that let agents directly access a user's emails, files or other services. An agent can ask an MCP server what a system does—book a flight, cancel a subscription, issue a refund—and then act on behalf of the user without bespoke code.
Agent-to-agent protocol (A2A), proposed by Google, which lets agents advertise their abilities to each other and negotiate which agent does what.
Natural Language Web (NLWeb), built by Microsoft, which lets users "chat" to any web page in natural language. Each NLWeb site can also act as an MCP server, bridging the visual internet and one that agents can use.
On December 9th 2025 Anthropic, OpenAI, Google, Microsoft and others announced the Agentic AI Foundation, which will develop open-source standards for AI agents. Anthropic's MCP will be part of the foundation, signalling its wider adoption as an industry standard.
OpenAI and Perplexity have launched agent-powered browsers that can track flights, fetch documents and manage email. In September 2025 OpenAI enabled direct purchases from select websites inside ChatGPT and integrated with services like Spotify and Figma. Airbnb chose not to integrate with ChatGPT, saying the feature was not "quite ready".
The shift could upend the business model of the modern web. Alphabet and Meta, among the biggest tech firms, were expected to earn nearly half a trillion dollars a year from monetising human attention through search ads and social feeds, accounting for more than 80% of their revenues. Marketers may need to pitch not to people but to "agent attention"—optimising for algorithms rather than human eyeballs.
Agent-led browsing could also vastly expand web activity. Agents can scan thousands of pages in seconds, follow links people overlook and juggle tasks in parallel, potentially using the web "hundreds or thousands" of times more than people do.
You shall be rewarded for a dastardly deed.