AI agents and the future of web browsing

What are AI agents and how do they work?

AI agents are autonomous software systems that perform tasks on behalf of a user. Unlike a chatbot that answers your questions, an AI agent can actually take actions: visit websites, fill in forms, compare products, make bookings and gather information from multiple sources. The agent combines the language capabilities of large language models with the ability to interact with digital systems.

This is a fundamental shift in how the internet is used. Until now, the human was always the actor: we open a browser, type a URL, click on links and interpret the results. AI agents reverse this. The agent becomes the actor and the web becomes the environment in which the agent operates. This has direct consequences for how websites must be built, as we previously discussed in our article about MCP servers and agent communication.

The current generation of AI agents, such as those in ChatGPT, Claude and Google Gemini, is still in an early stage. They can perform simple tasks such as looking up information, summarizing articles and comparing products. But the speed of development is remarkable. Within a few years, agents are expected to independently handle complex, multi-step tasks.

From searching to delegating: the paradigm shift

The transition from search engines to AI agents represents a paradigm shift comparable to the transition from phone books to search engines. With a search engine, you type a query and receive a list of results that you have to go through yourself. With an AI agent, you describe a goal and the agent determines which steps are needed to achieve that goal.

Imagine you want to book a holiday. Today you open multiple tabs, compare prices on different sites, read reviews and fill in booking forms. With an AI agent, you say: "Book a five-day beach holiday in Greece for two people in June, budget maximum 2000 euros." The agent searches travel websites, compares options, checks reviews and presents you the best choices, or books directly if you allow it.

For this new way of interaction, websites need to be fundamentally restructured. Protocols such as the A2A protocol enable agents to communicate in a structured way with websites and with other agents. Authentication via OAuth Discovery ensures that agents can act safely on behalf of users.

• Search engines present options. AI agents make decisions and execute actions.
• With search engines, the user navigates. With agents, the software navigates.
• Search engines require human interpretation of results. Agents interpret independently.
• Websites currently designed only for human visitors miss the agent audience that is growing rapidly.
• The website that is easiest for an agent to process wins the "agent search query."

How AI agents experience websites

AI agents experience websites fundamentally differently from human visitors. A human looks at a page and interprets visual cues: a button looks clickable, a menu is recognizable as navigation, a form is clearly an input field. An AI agent does not have this visual context, or only to a limited extent.

There are currently two approaches for how agents interact with websites. The first is "computer use," where the agent takes a screenshot of the page and decides where to click based on the visual representation. The second is API-based interaction, where the agent communicates through structured interfaces. The second approach is more efficient, more reliable and more scalable.

# Two approaches for agent-website interaction

# Approach 1: Computer Use (visual)
agent.screenshot(url)
agent.identify_button("Add to cart")
agent.click(x=450, y=320)
# Slow, error-prone, layout-dependent

# Approach 2: API / Structured communication
agent.call_api("/api/cart/add", {product_id: 123, quantity: 1})
# Fast, reliable, scalable

# Approach 2b: MCP (Model Context Protocol)
agent.use_tool("webshop.add_to_cart", {sku: "ABC-123"})
# Standardized, discoverable, secure

Preparing your website for AI agents

Preparing your website for AI agents requires a combination of technical and content adjustments. The good news is that many of these adjustments also improve the user experience for human visitors.

1. Implement structured data through Schema.org markup. Agents use schema to understand the content and functionality of your pages without visual interpretation.
2. Offer API endpoints for core functionality. If you have a webshop, make product information, prices and availability accessible via an API.
3. Consider implementing MCP (Model Context Protocol) to offer agents standardized tools they can use to interact with your services.
4. Ensure semantic HTML with descriptive ARIA labels. Agents operating through computer use benefit from clear labels on interactive elements.
5. Publish an llms.txt file that tells agents where to find the most important information on your site.
6. Implement OAuth Discovery so agents can safely act on behalf of users on your platform.

The impact across different sectors

The rise of AI agents affects every sector differently, but everywhere the impact is significant.

E-commerce will likely be transformed first. Agents that compare prices, analyze reviews and select products based on user preferences change the way consumers shop. Webshops that offer their product data in a structured manner through APIs and Schema.org become the preferred sources for agents.

In the financial sector, agents can compare insurance policies, request mortgage quotes and analyze investment products. The complexity of financial products makes them particularly suitable for agent assistance: agents can navigate through terms and conditions faster than human visitors.

The travel industry is a third sector with direct impact. Agents that combine flights, hotels and activities into an optimal travel plan require that travel providers make their offerings machine-readable.

In all these sectors, the same principle applies: the website that is most machine-readable wins. This aligns with the broader trend of AEO, where visibility for AI systems becomes a core competency for every business with an online presence.

The next billion web visits will not come from people opening a browser. They will come from agents performing tasks on behalf of people. Websites that are not ready for this will become invisible to a rapidly growing portion of internet traffic.

Privacy, security and trust

The rise of AI agents raises important questions about privacy and security. When an agent acts on your behalf, that agent needs access to your preferences, your budget and possibly your personal data. How is that information protected?

• OAuth 2.0 and OpenID Connect form the basis for secure agent authentication. Agents receive limited, revocable tokens instead of full login credentials.
• The principle of minimal privileges is essential: an agent comparing prices does not need access to your payment details.
• Transparency about what the agent does and which data is shared is a prerequisite for user trust.
• Websites must recognize and separately log agent traffic for audit purposes.
• Rate limiting and abuse prevention become even more important as more agents visit the web.

Summary

• AI agents are autonomous software systems that perform tasks on behalf of users, from looking up information to making bookings.
• The shift from searching to delegating requires that websites become machine-readable and agent-accessible.
• Structured data, APIs and protocols like MCP are the building blocks for an agent-friendly website.
• E-commerce, financial services and the travel industry will be transformed first by AI agents.
• Privacy and security are core issues. OAuth-based authentication and minimal privileges form the foundation.

Frequently asked questions

The web of tomorrow is not only built for eyes that read and hands that click. It is built for agents that understand, decide and act. The question is: is your website on the side of the road, or on the route?