The Interface Is Changing Again

The way we interact with the internet has changed before.

We used to open websites to get things done. Then mobile applications became the primary interface for many of those activities. Instead of visiting a website, we opened an app and did the same thing there.

The next shift in how we interact with software is already underway.

1. Websites are becoming optional¹

We're starting to ask AI assistants like ChatGPT, Claude, or Grok directly, instead of searching Google, opening Stack Overflow, and navigating through multiple pages to find an answer.

Even when we do use Google, the search experience itself is changing. AI Overviews now summarize some search results before we visit the websites behind them. There's also a prompt to continue the conversation in AI Mode without leaving the search experience.

I've caught myself following that path several times. And I'm noticing more and more people around me doing the same.

Sometimes, we get what we came looking for without ever visiting the website.

2. Agents start taking the role of applications.

We're already beginning to rely on purpose-built agents for increasingly complex tasks.

Coding agents are an interesting example. Instead of manually navigating between files, editors, terminals, documentation, and other tools, developers are increasingly delegating tasks to agents that can inspect a codebase, make changes, run tests, and iterate.

The vision being presented by many frontier AI labs goes even further: developers may eventually spend less time looking at and writing code directly, and more time describing what they want built and reviewing the results.

Whether we get there remains to be seen, but the direction is interesting.

Travel planning is another example. Many frontier AI demonstrations follow a similar pattern: ask an agent to plan a trip, and it researches destinations, compares flights and hotels, and puts together an itinerary.

These experiences aren't seamless yet. But they illustrate a different way of interacting with services that we currently access through multiple applications.

3. Agents start working with other agents.

We're already experimenting with this. We're connecting agents to tools through protocols such as MCP, building orchestration layers, and splitting tasks across subagents.

But much of this is still being assembled on top of infrastructure that wasn't originally designed for agents to coordinate with one another. We try to make a single agent handle increasingly complex workflows, or divide the work across subagents and stitch the results together. It works in some cases, but coordination and execution can still be brittle.

Perhaps the next step isn't simply building more capable agents or adding more subagents. It might be building systems designed for agents from the ground up: environments where they can discover capabilities, exchange context, delegate work, and coordinate execution.

We might end up building agents on top of systems built specifically for agents.

This transition won't necessarily be frictionless. Existing platforms may not always welcome external agents interacting with their services.

Amazon blocked Meta's Muse AI agent from shopping on its website ². It had also taken Perplexity to court over its Comet shopping agent, in a dispute that saw a temporary injunction overturned on appeal ³.

An agent might be capable of completing a task, but that doesn't mean the services it depends on will let it.

---

The browser may not disappear. Applications may not disappear either. But the way we interact with software is already changing, and the systems underneath those interactions may change along with it.

Whatever comes next, I think we're beginning to see the contours of that shift.

¹ Optional doesn't mean disappearing. It means websites may no longer be the default destination for every interaction.

² Amazon blocks Meta’s Muse AI assistant in new standoff over agentic shopping. geekwire

³ Amazon loses US court ban on Perplexity's AI shopping tools. reuters