The internet is about to undergo its most significant transformation since the smartphone made it mobile.

A new kind of user

For three decades, the web has been designed with a single user in mind: you. Every button, every form, every scroll interaction exists because a human needs to see it, understand it, and click it. But a fundamental shift is underway. AI agents—autonomous software systems powered by large language models—are becoming primary users of the internet.

These aren’t the simple bots of the past that scraped data or filled forms. Today’s agents can browse websites, interpret content, make decisions, and take actions on behalf of humans. They can book flights, compare products, negotiate prices, and complete transactions—all without a person ever opening a browser.

As dawn song, professor of computer science at UC berkeley, puts it: the web today is designed for how humans consume information and take actions. But agents don’t operate like humans and don’t share our limitations—so the web can be designed very differently.

From browsing to doing

The agentic web represents a paradigm shift from browsing to doing. Traditional search works like this: you type a query, scan a list of links, click through pages, compare information across tabs, and eventually take action. It’s labor-intensive, and you’re the bottleneck.

In the agentic model, you express intent—”find me a flight to Barcelona next month under €400 and book the best option”—and an agent handles the rest. It searches multiple airlines, compares prices, checks your calendar, considers your preferences, and completes the booking. The web becomes infrastructure rather than interface.

This is already happening. Openai launched chatgpt atlas in late 2025, a browser where AI doesn’t just assist—it acts. Perplexity’s comet browser automates multi-step tasks across websites. Fellou, genspark, and others are building what they call “self-driving browsers” where agents navigate, click, fill forms, and transact autonomously.

The infrastructure shift

The implications run deep. If machines become the primary consumers of web content, the infrastructure needs to change. Human-readable layouts matter less than machine-readable data. Visual design becomes secondary to semantic structure. APIs become more important than interfaces.

We’re already seeing this with the model context protocol (MCP), an open standard introduced by anthropic and now adopted by openai, Google, and Microsoft. MCP provides a universal way for AI agents to connect with tools, data sources, and services—essentially USB-C for AI applications. Instead of building custom integrations for every service, developers implement MCP once and unlock an ecosystem of connections.

Microsoft has introduced NLWeb, which they believe could play a role similar to HTML for the agentic web—making it easy for websites to provide conversational interfaces and become discoverable by AI agents.

What this means

The agentic web isn’t replacing the human web—at least not yet. But it’s creating a parallel layer where machines interact with machines on our behalf. The websites, platforms, and services that adapt to serve both audiences will thrive. Those that don’t may find themselves invisible to an increasingly significant portion of internet traffic.

For anyone building digital products, This raises urgent questions: how do you design for users you can’t see? How do you optimize for agents that don’t click ads? How do you build trust with software that acts autonomously?

The answers are still being written. But the questions are no longer theoretical.