The First AI App Boom Won't Be on Your Phone — It's Already Happening in the Back Office

Recently, Meta told a ghost story about excess computing power, and AI stocks plummeted in response. Everyone is anxious.

Everyone is asking whether this AI game can continue, when AI applications will finally explode, when the AI equivalent of TikTok, WeChat, or Xiaohongshu will appear, and whether AI will repeat the dot-com bubble of 2000.

Let me make a prediction: Within six months, the AI infrastructure bubble is very likely to burst; one year from now, enterprise AI applications will begin to explode; two years from now, personal AI applications will truly take off.

Why do I say this?

Because today's AI is remarkably similar to the personal computer from forty years ago.

Today's AI is indeed strong at chatting and assisting with tasks, but it's still far from what ordinary people imagine as "open and use, get hooked instantly." When you want it to write a weekly report or plan a trip, your first reaction is often not "how convenient," but "how should I phrase this to it?"

Back then, computers were the same: big, clunky machines with no games, no short videos, no social networks, and not even a graphical interface in the early days. Users faced a black box with a blinking cursor. You had to type commands yourself, learn the operations yourself, and endure all sorts of inexplicable errors.

In 1981, IBM launched its first PC, priced at $1,565, equivalent to over five thousand dollars today. If you took this thing to an average consumer and asked "Do you want to buy it?", the answer you'd likely get is that you're crazy.

But here's the question — why would anyone pay for such a black box?

Enterprises.

Enterprises don't need a computer to be fun; they need it to work. Financial accounting, inventory management, order tracking — these tasks were previously done entirely with paper and people, which was inefficient and error-prone.

No matter how difficult a computer was to use, as long as it could improve efficiency, reduce errors, and save labor, enterprises would pay for it.

The IBM PC sold 250,000 units in its first year, and the vast majority of buyers were companies. Throughout the 1980s, the growth in PC installations was almost entirely driven by enterprise procurement.

It wasn't until hardware became cheap enough and the software ecosystem mature enough that the home market took off in the 1990s — first with games, then with the internet.

Today, when many people think of the internet, they think of Google, Facebook, YouTube, Taobao, but none of these existed at the very beginning. What the early internet truly changed were things so plain they were almost boring: corporate email, internal communications, online forms... The companies that first made money from the internet were all doing the hard work of helping enterprises manage customers, orders, and collaboration.

Enterprise demand propped up the installed base first; only after the user base and habits were established did consumer-facing applications grow out of it.

---

Back to today's AI.

Three years in, ChatGPT, Claude, Codex, various Agents, and generative tools all exist, but ordinary people indeed haven't seen many "must-use" new applications. Everyone is waiting for a C-end super entry point, something like WeChat or TikTok that people open every day.

But we might be looking in the wrong direction.

Because the first wave of explosion will happen in offices, not on personal phones or PCs — on the B-end, not the C-end.

Why will enterprise scenarios come first? Because enterprises calculate ROI. A customer service representative's annual salary is over a hundred thousand yuan. If AI replaces 30% of the workload, saving tens of thousands a year, the math works out, and they will pay. Ordinary consumers aren't like this — free is best, and spending twenty yuan on a subscription requires hesitation. The gap in willingness to pay determines which side matures first.

So enterprise AI will penetrate just like enterprise informatization did back then: first cutting into peripheral processes (knowledge base Q&A, meeting minutes, code assistance), then entering core processes (ticket processing, contract review, financial reconciliation), and finally, cross-system Agents will truly run complete processes on behalf of people.

As for personal AI applications, they will come, but they need enterprises to pave the way first.

The C-end has three structural difficulties: weak willingness to pay — individuals don't open their wallets as readily as enterprises;

tasks are too fragmented — enterprise processes, however complex, at least have fixed goals and fixed systems, whereas individual lives have vastly different needs that are hard to cover with a single standardized product;

habits haven't formed yet — most people are still figuring out how to ask AI questions. Enterprises can push adoption through training and systems, but consumers can only wait until the product is so good they naturally start using it.

Once enterprise scenarios drive down model costs and polish stability, personal AI assistants can transform from "interesting toys" into "daily indispensable entry points."

---

So back to the opening question. Three years in, why haven't AI applications exploded yet?

Because we've been staring at the phone's home screen, but the real first wave of explosion is happening quietly in offices. Code repositories, customer service backends, ticketing systems, financial processes — these places won't trend on social media, but they have budgets, ROI, cost-reduction pressure, and people willing to pay for results.

The first step of a true technological revolution is never about getting people addicted; it's about saving them money.