The internet has always been a glorious mess, open, useful, frustrating, creative and occasionally weird enough to make you wonder who designed it, then remember that everyone did.

That messy openness has been part of its magic. For decades, the web has worked because we could search, browse, compare, click, question and decide. We had too many tabs, too many ads and far too many cookie pop-ups, but we also had agency.

AI changes that relationship.

Increasingly, we will ask an AI agent rather than search the web in the familiar way. Instead of receiving a page of links, we will receive an answer, a recommendation or, eventually, an action taken on our behalf. That could make the web dramatically more useful, but it could also make it smaller, more controlled and less transparent.

In a recent conversation with Mozilla CEO Anthony Enzor-DeMeo, he put the issue very simply: “When you’re going with an agent, an agent is providing you the answer.” That is the key shift. Once AI becomes the interface between people and the internet, the critical question becomes: why did it choose that answer?

The New Front Door To The Internet

Search engines already shape what we see online. Social media platforms do the same. AI agents could take that power further because they may remove the visible selection process altogether.

When we use a search engine, we can still scan the options, see a sponsored result, ignore it, open several tabs and make a choice. The process is imperfect, but it remains visible enough for us to exercise judgment.

With AI agents, that process can disappear. The agent may summarize the web, rank the choices, discard what it considers irrelevant and present one answer as if it were obvious. Convenience improves, yet visibility declines.

Enzor-DeMeo raised the concern directly: “Is this information that is given to me because the bot thinks it’s the best, or is it because it might be an ad?”

That question will define the next phase of the web. If an AI recommends a pair of shoes, a mortgage provider, a news source or a medical article, we need to know what shaped that recommendation. Was it selected because it genuinely fits our needs? Was it influenced by a commercial deal? Was it shaped by the platform’s incentives?

This is where AI could damage the internet. It could reduce discovery, narrow choice and place more power in the hands of a few dominant interfaces. Yet AI could also improve the internet by helping us find better information faster, avoiding low-quality content and organizing complex research. The outcome depends on design choices being made right now.

Why The Browser Suddenly Looks Strategic Again

For years, browsers have felt almost invisible. They are the window through which we access the internet, but most people rarely think about them unless something breaks.

Enzor-DeMeo argues that the browser may become one of the most important AI layers because it holds so much context. A browser can understand what we search for, what we read, which services we use and how we move across the web. That context could make AI assistants much more useful. It also makes the browser a highly sensitive point of control.

“The browser has an immense amount of context on the user,” he told me, referring to browsing history, search patterns, preferences and stored information. For AI companies, that context can create more personalized answers. For users, it is deeply personal information that deserves careful handling.

Mozilla’s position is interesting because Firefox is taking a different path from many larger technology companies. It is not trying to build the biggest AI model. It wants to provide an independent layer where users can choose which AI tools they use.

Enzor-DeMeo said, “We’re not beholden to one AI model,” and described Firefox’s advantage as “optionality.” That is powerful in a market where browsers, search engines and AI assistants are being bundled together.

Microsoft has Copilot. Google has Gemini. AI companies are experimenting with browser-style interfaces of their own. The commercial logic is obvious. Control the browser, and you control the AI interface. Control the AI interface, and you control the user relationship.

Choice, Consent And The Problem With Creepy Technology

The next phase of AI will test whether consent is treated as a genuine product principle or a line in a privacy policy.

Enzor-DeMeo was clear about Mozilla’s approach. Browsing information in Firefox is stored on the user’s device. “It’s your information, it’s not ours,” he said. If users want to use that information for a more personalized AI experience, they should be able to make that choice.

That sounds simple, but it cuts against much of the modern digital economy, which has become very good at collecting, inferring and monetizing user behavior. People have grown used to a certain level of online creepiness. We talk about a product, search for something vaguely related and then watch ads for it follow us around the internet like an overenthusiastic salesperson.

Enzor-DeMeo summed up Mozilla’s ad philosophy in two words: “nothing creepy.”

He also made an important point about advertising. “I actually think ads keep the internet free,” he said. A web without advertising would almost certainly become more paywalled, which would limit access for billions of people.

The real problem is hidden influence. If AI assistants become the new interface for search, shopping and decision-making, advertising must be transparent. Users should know when a recommendation is sponsored, when their data is being used and when they are being nudged toward a commercial outcome.

This becomes especially important as AI agents begin to act on our behalf. A traditional ad may influence what we click. An AI agent could influence what we buy before we have seen the alternatives. If your digital assistant starts quietly inserting paid recommendations into its advice, the personal assistant becomes a sales channel wearing a friendly mask.

Human-Centered AI Needs Real Controls

Many companies talk about human-centered AI. The phrase is easy to use and harder to prove.

For Enzor-DeMeo, it means giving users control over how AI appears in the browser. “I think AI is valuable,” he said, “but I want the human at the center of it.”

Mozilla has introduced AI controls in Firefox so users can turn AI-related features on or off. Interestingly, Enzor-DeMeo said, only a small percentage of users switched everything off. That suggests most people are not rejecting AI. They want control, clarity and the ability to decide when AI helps.

This is a useful lesson for every business leader. People rarely object to genuinely useful technology. They object to technology that is forced on them, hidden from them or designed in ways they do not understand.

For companies building AI into products, trust has to be treated as part of the product itself. It shows up in settings, defaults, explanations and the ability to opt out. It also comes from culture. Enzor-DeMeo said leaders need to understand what their product stands for and what users expect from it. That advice applies far beyond browsers.

The Better Internet AI Could Build

One of the biggest risks is that AI makes the internet less open.

If the most capable AI services sit behind subscriptions, access becomes unequal. If AI agents answer questions without sending traffic back to websites, publishers and creators may find it harder to fund high-quality content. If a handful of companies control the dominant models, browsers, ad systems and shopping interfaces, the open web could gradually become a set of private corridors.

Enzor-DeMeo warned that the internet could become “less open” if AI moves too heavily toward subscription models. He also argued that open source models, user choice and more efficient AI systems will be crucial for global access.

Despite the risks, I remain optimistic about what AI can do for the web. The internet has become too cluttered, too noisy and too difficult to navigate. AI can help us make sense of that complexity. It can organize research, summarize long documents, explain difficult topics and make digital tools easier for people who do not think like software engineers.

Enzor-DeMeo sees real upside, too. He said he hopes AI will bring “a lot more entrepreneurship” because access to information and the ability to do more with small teams will expand. That feels right. AI gives individuals and small businesses capabilities that once required a much larger organization. Used well, it can unlock creativity, experimentation and learning.

So, will AI break the internet or make it better?

The honest answer is that it can do either. AI can create a web that is more closed, opaque and manipulative. It can also make the internet more useful, more personal and more accessible. The difference will come from the values embedded into the systems we build, the choices we give users and the transparency we demand from the companies shaping this next chapter.

The internet has always worked best when it gives people room to explore. AI should help us explore with more clarity, more control and a little less digital weirdness along the way.