The Internet Wasn’t Built for AI. That’s About to Change.

 

The internet was built for people. Every website, every page, every button assumes that a human is sitting behind a screen — reading, clicking, and making decisions. That assumption is about to change.

AI assistants are evolving fast. They’re moving beyond simple chatbots that answer questions into something much more powerful: agents that can actually do things for you. Browse, compare, book, purchase, schedule — all on your behalf.

But here’s the problem. The internet doesn’t know how to talk to these agents yet. Websites are designed for human eyes, not for AI. That’s going to change. And when it does, the web as we know it will look very different.

The Internet Needs a New Handshake

Today, when you want to book a flight, you open your laptop, visit three or four airline websites, manually search for flights, compare prices across browser tabs, and then go through a checkout process. It works, but it’s slow and tedious.

Now imagine telling your AI assistant: “Find me a flight to New York for tomorrow.”

In theory, the AI should be able to check every airline, compare options, and come back to you with the best choices in seconds. But right now, it can’t do that reliably — because airline websites are built for humans to click through, not for AI to query.

The fix is surprisingly simple in concept. What if every website published a small, standardized file — think of it like a menu for AI agents — that says: “Here’s what our site offers, here’s how to search, here’s how to buy, here’s how to check an order status.”

This is what I’m calling agent.json — a universal standard that lets AI agents understand what any website can do and how to interact with it, without needing to be custom-built for each one.

This agent.json file will tell the LLM agent what are the API endpoints that they can you and how to use them. This is how we are currently using MCP and Tools.

What This Looks Like in Real Life

Let’s walk through the flight example from start to finish.

You say: “Find me a flight to New York for tomorrow.”

Your AI agent checks Delta, JetBlue, American Airlines, and United. It reads each airline’s menu of capabilities(agent.json), searches for available flights using the API endpoints, and comes back with a clean comparison — departure times, prices, layovers, seat availability — all in one view. No tabs. No back-and-forth.

You say: “Book the Delta flight, the 8:15 AM departure.”

The agent connects to Delta’s booking system (API endpoints described in agent.json), uses your saved travel profile and payment method, and completes the purchase. You get confirmation.

You say: “Add the flight to my calendar and attach the confirmation email to the event.”

Done. The agent creates the calendar entry with the right times and links the booking confirmation to it.

One conversation. Three actions. No forms, no repeated data entry, no wasted time.

This is what becomes possible when the internet is built to work with AI — not just with humans.

Which Websites Survive — And Which Ones Don’t

Here’s a consequence that most people aren’t thinking about yet: this shift will be devastating for a large category of websites.

Think about why you visit most websites. It usually falls into one of two buckets. You’re either there for a task — booking a flight, checking the weather, comparing prices, looking something up — or you’re there for an experience — watching a video, scrolling through your feed, reading something interesting.

In an AI-powered world, task-oriented websites lose their reason to exist as places you visit. If your agent can search flights, check the weather, or compare product prices directly — why would you ever go to those websites yourself? You wouldn’t. The agent handles it behind the scenes.

The websites that thrive are the ones where the experience itself is the point. YouTube, Instagram, TikTok, Spotify — you don’t want an AI to watch a video for you. You don’t want a summary of your friend’s story. These platforms survive because they deliver something an agent can’t replace: the joy of being there.

Travel booking sites, comparison tools, weather apps, documentation portals — many of these become invisible background services. They still exist, but humans stop visiting them. The front door disappears.

AI Won’t Always Wait for You to Type

There’s another assumption worth challenging: the idea that the way you talk to AI is by typing a message or speaking to it.

Text and voice are great starting points. But if you look at how automation already works in the real world — smart homes, scheduling tools, business workflows — triggers are everywhere. And most of them don’t involve typing a sentence.

A physical button. Imagine a button on your desk that kicks off your morning routine — your AI pulls up your calendar, checks the weather, briefs you on today’s meetings, and pre-orders your coffee.

A time-based trigger. Your AI checks your pipeline every Friday afternoon and drafts follow-up emails to clients who haven’t responded. No reminder needed. It just runs.

A notification. A new email arrives from your biggest client. Your AI reads it, flags it as urgent, and drafts a response for your review before you even open your inbox.

A change in the real world. Rain is forecast for Saturday. Your AI reschedules your outdoor plans and suggests alternatives — without you asking.

The chat box is just the beginning. The real power of AI agents comes when they’re woven into your daily life, responding to the world around you rather than waiting for a command.

The Security Question

There’s a serious concern that comes with all of this, and it’s important to be upfront about it.

When AI agents are authorized to act on your behalf — spending money, sending messages, accessing your accounts — the stakes go up significantly. A compromised agent or a malicious website pretending to offer legitimate services could do real damage.

On top of that, one of the most exciting things about AI is that it lets anyone build automations and tools, even without a technical background. That’s incredibly empowering, but it also means a flood of new software built without deep security expertise. More code from more people means more potential vulnerabilities.

The good news is that the same AI technology creating these risks can also help manage them. AI tools that review code for security flaws, flag suspicious behavior, and enforce best practices will become essential. But the challenge is real, and it’s growing.

I’ll be writing a deeper piece on the security implications of AI-driven development soon. Stay tuned.

We’re Still at the Very Beginning

It’s exciting to imagine this future, but let’s be honest about where we stand today: we’re at the very beginning.

AI agents are getting more capable every month, but the infrastructure around them — the standards, the security frameworks, the trust models — hasn’t caught up yet. There is no universal standard like agent.json today. There's no shared agreement on how agents should identify themselves, what they're allowed to do, or how to handle things when they go wrong.

But the direction is unmistakable. The web has evolved before — it developed standards for search engines, for logins, for data sharing. It will develop standards for AI agents too. The question isn’t whether this happens. It’s how fast, and who gets to shape the rules.

The internet is going to change. Entire categories of websites will shift from places you visit into services that run quietly in the background. New ways of triggering AI will emerge that go far beyond typing into a chat window. And yes — new risks will come with it.

We’re building the agentic internet right now, whether we’re ready for it or not.