Will Big Tech Control the Future of AI Agents on the Web?

The history of the internet has always been shaped by a tension between open access and corporate control. On one side is the democratic vision of a network where anyone can build, connect, and share. On the other is the growing reality of a digital landscape dominated by a handful of powerful companies.

A New Scientist article by Chris Stokel-Walker (11 June 2025) explores how this struggle is entering a new phase with the rise of AI agents, autonomous systems designed to act on our behalf online. The fear is that the very protocols allowing these agents to communicate could become yet another tool for big tech to consolidate power, leaving smaller players and open innovation on the margins.

What Exactly Are AI Agents?

If you’ve used ChatGPT, Gemini, or Claude, you’ve already experienced conversational AI. An AI agent takes that idea a step further. Instead of just answering questions, it can act on your behalf, browsing websites, booking flights, comparing products, or even coordinating with other agents to complete a task.

Think of an agent as your digital personal assistant that doesn’t just talk, but also does.

Why the Internet Isn’t Ready for Them

Here’s the catch: the internet we know was built for humans, not machines. Websites, logins, and forms are designed for people clicking and typing, not for AI agents negotiating with one another. To make agents work, we need new “protocols”, shared rules that let software talk to software.

Right now, two big players are leading the charge:

• MCP (Model Context Protocol) – developed by Anthropic, it’s like giving every agent a universal phone number so it can securely connect to data sources and tools.

• A2A (Agent-to-Agent) – developed by Google, this goes further, creating a kind of group chat where agents can discover each other, exchange information, and coordinate tasks.

Both sound promising, but they come with a catch: they’re being built inside big tech companies.

The Risk of Walled Gardens

History gives us a warning here. Social networks, app stores, even email to some extent, all started open but were eventually fenced off by big players who controlled access, extracted fees, or limited interoperability.

If MCP or A2A becomes the “default” protocol, whoever controls it will effectively control how AI agents interact with the web. That could mean:

• Smaller startups struggling to compete.

• Hobbyists and independent developers locked out.

• Users pushed into ecosystems where choice is limited.

In short: the internet risks becoming less like a public library, more like a private shopping mall.

Enter ANP: The Open Alternative

This is where the Agent Network Protocol (ANP) comes in. Led by Gaowei Chang and supported by the World Wide Web Consortium (W3C), ANP is an open-source alternative designed to keep the agent internet open to everyone.

Unlike MCP or A2A, ANP doesn’t rely on a central authority or a catalogue of “approved” agents. Instead, any agent can discover and connect with another, much like the early internet when anyone could set up a website or an email address without asking permission.

In plain terms: ANP is trying to keep the internet a level playing field, where innovation doesn’t require permission from Google or Anthropic.

Why This Matters Beyond Tech Circles

You don’t have to be a developer to see why this is important. Imagine a future where:

• Healthcare apps can share data across systems freely, instead of being locked into one vendor’s ecosystem.

• Small businesses can deploy their own AI agents without paying gatekeeping fees.

• Individuals can build personal agents that truly work for them, not for a company’s bottom line.

That’s the difference between an open standard and a corporate-controlled one.

My Takeaway

As Catherine Flick, a researcher at Staffordshire University, put it: open protocols are about bringing democratisation back to the internet. If we leave the rules of AI agents solely in the hands of big tech, we risk repeating the mistakes of past digital platforms, where openness gave way to lock-in.

I see this as one of the defining infrastructure questions of the AI time. The technology is moving fast, but unless we push for open, transparent standards, the internet of agents may not belong to all of us.

So the next time you hear about “AI agents booking your holiday,” remember: the real question isn’t can they? It’s who decides how they do it—and for whose benefit?