Phones Will Disappear in 5 Years? Elon Musk's Prediction, AI Agents & MCP

I grew up in the iPhone and smartphone era. They’re part of my daily life. I never imagined that “phones” and “apps”, two things that define our generation’s digital existence, could “possibly” become history in just five years.

Elon Musk predicts: “Apps and phones will disappear within five years.” He’s not the first to make such a prediction, but hearing him say it triggered me to think about and organize recent observations. So I want to write this article to organize how I understand the AI business world, how I see the business paradigm shift in the AI era. I’ll discuss why Elon Musk made this prediction, how MCP and AI agents are changing the digital ecosystem, and after witnessing JD.com and online bookstores’ physical transformations in Taiwan, how I’m rethinking my value as a knowledge worker.

Musk’s Shocking Prediction: “In Five Years, Apps Will Be as Outdated as Blockbuster”

#

In November 2025, Tesla CEO and one of the world’s richest people, Elon Musk, made a prediction on the famous podcast The Joe Rogan Experience that media outlets went crazy over: within five to six years, traditional smartphones, operating systems, and mobile apps will all become obsolete.

Here’s exactly what he said:

“There won’t be operating systems. There won’t be apps in the future…
Apps are like Blockbuster video pretty much, and everything’s run through AI.”

More importantly, Musk further explained the new architecture that will replace them:

“The what we call a phone will really be an edge node for AI inference…
You’ll get everything through AI.”

The key is his mention of edge node. What exactly is that?

Imagine your phone is no longer an independent terminal device but becomes a “smart interface.” Its main functions are reduced to displaying visuals on a screen and playing audio. All computation and decision-making are done by AI agents. This device will process AI locally as much as possible to reduce data transmission with servers. In other words, everything you might want (whether videos, shopping, or calendars) will come directly through AI. You don’t need to operate the phone or install apps. Only AI is the real brain working behind the scenes.

I could throw away all the apps on my phone and do anything with just AI—it sounds a bit like science fiction. This makes us pay more attention to recent developments in MCP and AI agent integration, because they’re already making Musk’s “smart interface” edge node possible. We’re already on the road to this sci-fi world.

MCP + AI Agent: Reshaping How the Digital World Works

#

To understand why Musk’s prediction might come true, we need to know two key technologies: Model Context Protocol (MCP) and AI agents.

What Is an AI Agent? How Is It Different from ChatGPT?

#

Before discussing MCP, we need to understand a key concept: AI agents. I believe by late 2025, everyone’s probably heard this term so much their ears are ringing. Here’s a simplified analogy:

• ChatGPT or Gemini and other large language models are “AI tools” that do one thing at a time
• AI agents are “AI employees” that complete one task at a time by doing many things

An AI agent is an AI application that can autonomously perceive, decide, and execute tasks without human intervention. Its biggest features are autonomy and task execution capability. Here’s an example to illustrate the difference:

Regular conversational AI (like ChatGPT):

You: "I want to book a restaurant for a date"
ChatGPT: "I suggest trying Google Maps search, or checking reviews on Dcard"

> You need to open Google Maps and Dcard apps yourself and complete all the next steps

AI Agent:

You: "I want to book a restaurant for a date"
AI Agent: Auto-search nearby restaurants → Compare reviews and prices
          → Check your calendar → Find an opening → Complete reservation
          → Add info to your Google Calendar → Remind you when to leave

> You just wait for results—you don't have to do anything!

An AI agent is like the brain for your life and work, like a butler who can accomplish complex tasks for you. But for a butler to help at your home, you need to give them a key to access all the rooms and let them know what tools are available at home, right? MCP is the key that connects AI butlers to all the world’s tools.

What Is MCP? The “Universal Adapter” of the AI World

#

MCP (Model Context Protocol) is an open standard launched by Anthropic in November 2024. It’s like a “USB-C port” for AI applications—a universal connection method that lets AI models safely and standardly access external real-time data and tools. Specifically, MCP servers provide clients with three core functions: resources (including context), prompts, and tools.

Let’s continue with the USB-C port analogy to explain in plain language. Now mice, keyboards, and speakers can all connect to your computer using the same rectangular USB-C port. Imagine a world where mice have round ports, keyboards have diamond-shaped ports, and speakers have S-shaped ports—wouldn’t that be crazy with so many ports? If an AI world “didn’t have” MCP, it would be this chaotic.

Imagine the old internet world worked like this: if there are 10,000 AI tools and 10,000 external services (like Gmail, weather apps, shopping sites), to let them communicate with each other, you’d theoretically need to establish 100 million (10,000 × 10,000) independent connections. Like every appliance having its own unique plug specification with different shapes and sizes, leading to terrible interoperability.

MCP’s emergence is to solve this fragmentation problem. Through this standard protocol, each AI tool and each external service only needs to implement the MCP interface once—the example above only needs to be reduced to 20,000 connections. Like the entire AI world suddenly using the same socket standard—suddenly, all devices can easily connect.

Going back to what we mentioned earlier, AI agents’ biggest feature is autonomous decision-making and execution. But if AI agents can’t easily connect to external tools, they’re like robots with brains but no hands and feet: they can think but can’t do. MCP allows AI models to transcend their pre-trained knowledge limitations and connect to various tools and real-time data sources, letting AI agents really flex their muscles.

When AI agents can do anything for you, isn’t that what Musk said: we only need phones as “smart interfaces”, where apps and operating systems don’t matter because everything is decided and executed by AI agents.

This isn’t theory. Courts are witnessing real battles showing AI agents have real influence.

Amazon vs. Perplexity: A War About “Who’s Shopping”

#

In 2025, e-commerce giant Amazon sued AI startup Perplexity. The core dispute is over Perplexity’s AI browser Comet, whose AI agent can help users auto-login and shop on Amazon’s platform.

AI automatically compares prices, reviews, and brands for you, decides what to buy—amazing, right!

What Experience Does Perplexity Provide?

When you want to buy a new coffee maker, the traditional way is: open Amazon app → search → browse 50 products → compare prices and reviews → feel indecisive → close app → think about it tomorrow.

Perplexity’s Comet smart assistant changes this flow. You just say “help me find a coffee maker suitable for a small household, budget $150,” and AI will automatically search, compare, and even directly place an order, saving you tons of time.

Why Is Amazon Alarmed?

Amazon isn’t simply a platform for selling things. Its real intent is to show users ads or display sponsored products, and influence purchase decisions through upselling (making you buy more expensive) and confusing promotions (making you buy more). But when shopping decisions are all made by AI agents, AI won’t be attracted to ads and won’t impulse buy! This directly threatens Amazon’s most profitable advertising business, so of course it resists.

Perplexity positions this dispute as innovators fighting monopolists, seeing it as “big companies using legal threats (bullying) to stop innovation.” Regardless of the lawsuit’s outcome, I cite this case because it presents a clear signal:

When AI Commoditizes Everything Online, Experience Economy Becomes the Moat

#

If AI really replaces phones and apps, what can’t it replace? My observations this year make me think “hands-on experience” might be the answer.

Expert Perspective: “Experience” Can’t Be Fully Digitized

#

Stella & Amy’s Podcast discussed how retail giant Walmart is also investing resources in e-commerce and digitization. Analyst Iris offered a key insight:

Industries where “customer experience is greater than the product itself” and products that need “physical experience to make decisions”
are hard to fully e-commercialize

This insight echoes many of my recent experiences. Over the past decade, many knowledge workers frantically pursued scale. People record online courses, write e-books, build automated sales funnels. Knowledge workers were taught that “zero marginal cost” is the holy grail of knowledge monetization.

But now, when AI can generate a professional-looking online course in seconds, everyone suddenly realizes: Things that can be scaled and digitized are often the easiest for AI to replace.

Big companies are already having the same realization, seeking experience economies that AI can’t easily replace:

Case 1: JD.com Physical Stores Aren’t Just Stores—They’re Experience Centers

#

This year when my fiancée and I visited family in China, one of the most impressive observations was seeing JD.com’s appliance physical stores.

The appliance stores we commonly see in Taiwan first established massive physical stores, then later started developing e-commerce. JD.com did the opposite. Already being one of China’s largest private e-commerce companies, they later developed physical stores.

Of course, JD.com’s appliance physical stores aren’t just selling products. They’re selling experiences. JD.com adds home-themed model rooms, recreating real family living atmosphere, evoking scenario-based consumption experiences. When you sit in the “home” they’ve created, watching TV shows you like, this kind of immersion is something no online store can provide. AI can’t give you the see-it, touch-it emotion.

Case 2: Online Bookstore’s Physical Experiment

#

Taiwan’s largest online bookstore, books.com.tw, after dominating the online book-buying market for years, also went physical this year.

The online bookstore reported losses in Q1 2023. The previously effective “daily 66% off” promotion mechanism is now seeing declining conversion rates. Meanwhile, physical bookstores’ sales have shown an upward trend in recent years. Price wars in the book-buying market have become hard to fight in recent years, and in the AI era, they’ll only become more futile, because AI can help you compare prices endlessly. Now bookstores are rethinking new opportunities to retain customers: experience.

The physical bookstore isn’t just about selling books. It has added all kinds of physical experiences. It plans digital bookshelves and AI interaction zones, connecting online resources; builds immersive experience zones for e-books and audiobooks; the store also has art brand galleries, achieving coexistence of books and artworks in the same space, managing reading like an exhibition.

Their strategy is to make physical stores nodes for managing reading behavior, not simply pursuing profit and loss. Revenue can’t be limited to simple book transactions cuz transactional behavior will also quickly be replaced by AI agents anyway. In the future, you might buy a reasonably priced book through an AI agent, but you won’t even know or care whether the book came from the online bookstore, Momo, or an e-commerce platform.

Slightly Off-Topic: Knowledge Workers’ Struggle

#

This article talks about Musk’s prediction that apps, phones, and operating systems will disappear, likely realized through AI agents + MCP. But writing (mostly in Mandarin actually) this far, I’m also thinking about which of my skills as a knowledge worker will quickly become worthless, especially the scaling strategies I thought were advantages that have already been undermined by AI.

I’ve always believed: “If you’ve explained a concept to others many times, you should write a blog post.” This is the scaling I believed in and practiced: write one article, reach thousands or tens of thousands of people.

But in the last year or two, things started changing. When even my own learning process relies heavily on AI conversations, I wonder “what’s the difference between my blog post and what ChatGPT generates?” I was initially unwilling to face reality: how can an article I spent an entire weekend writing be compared to AI-generated content? But when I calmed down and tested with Claude, like “write an A/B testing beginner tutorial,” I found: If judging only by content “completeness” and “correctness,” AI-generated articles have indeed reached over 80 points.

Not to mention AI can do more: when I tried having AI create a “Python data analysis online course outline,” the structure it gave in 30 seconds would exceed what I spent three days planning.

I have to admit, my “scaled knowledge sharing” model I’ve been proud of for the past five years—the logic of “write once, read by countless people”—is being thoroughly commoditized by AI. If anyone can generate professional-looking tutorial content in minutes through AI, where’s the value in my blog writing?

My conclusion after reflection is as written here: seek things that can’t be scaled or digitized—that is, “experience.”

In 2025, I’ve tried more intensively giving talks and workshops at colleges, not just sharing data science and AI knowledge but also hands-on practice with students, directly listening, discussing, and solving students’ learning problems on the front line. This year I also started writing a (Mandarin) newsletter. The content isn’t as polished as blog posts, but I can ramble and write whatever I want, presenting the real me and showing “human flavor.”

AI is good at organizing and conveying information, but it can’t replace deep companionship, personalized feedback, real-time interaction, and most importantly: trust and connection between people. I think while embracing AI, I’ll continue deliberately honing these interpersonal skills AI can’t do ("People Skills").

Conclusion

#

Musk’s prediction, MCP protocol’s rise and integration with AI agents, the Amazon vs. Perplexity lawsuit—these seemingly independent phenomena actually point to the same conclusion: The digital world’s efficiency is being pushed to extremes by AI.

From my personal experience as a knowledge worker, plus seeing JD.com and online bookstores’ physical transformations, I believe what we humans can continue contributing (after losing to AI in production efficiency) are experiences that require real human presence and create emotional resonance.

At least, this is how I interpret today’s business paradigm shift.

References:

• Norah Sakal: What is Model Context Protocol (MCP)?
• Bandi Revanth: APIs don’t make good MCP tools
• A Deep Dive Into MCP and the Future of AI Tooling, By a16Z Partner