---
title: "Why Do Some People Use AI Every Day Yet Never Get Stronger?"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-04-27T10:31:46+00:00"
canonical: "https://ffcap.cn/en/research/src-20260427-01html"
source: "https://uniqueresearch.substack.com/p/src-20260427-01html"
language: "en"
---

# Why Do Some People Use AI Every Day Yet Never Get Stronger?

_Original · Unique Research · 2026-04-27_

_Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the opening essay, all section headings, the workshop framework, the two selected questions, the four-step method, and the closing CTA. The "100 times more important" formulation, the 2x2 screening matrix, and the organizational predictions are the author's analytical framing, not independently audited findings. Company, product, and personal names are retained as the source's naming. The source is dated April 27, 2026._

WORKSHOP

The First Group of People Eliminated by AI May Not Be Those Who Cannot Use AI

What truly widens the gap in the AI era may not be whether you can use tools, but whether you know what problem you are actually trying to solve

There is a rather counterintuitive question: why do some people use AI every day yet never get stronger?

They also bought memberships, also ask DeepSeek, Doubao, and ChatGPT, also know what Agent is, what RAG is, what an enterprise knowledge base is. They even let AI write copy, revise PPTs, make summaries, and produce proposals every day. But half a year later, their working methods have not truly changed, and their capability boundaries have not been widened.

Even more paradoxically, some people who do not seem so "tech-savvy," who do not understand code and do not chase every new model, have instead used AI very deeply.

Some people use AI to analyze their running data, sleep, heart rate, and athletic performance; some use AI to build an English-learning companion system for their children; some distill meetings, courses, and business experience into an enterprise knowledge base; and some have started connecting AI to Feishu, CRM, and business processes, letting it not just answer questions but participate in work.

So what truly widens the gap in the AI era may not be "whether you can use tools."

But another, more fundamental question:

Do you know what problem you are actually trying to solve?

This was also the most obvious change on-site at a recent AI private-board workshop held by Unique Research.

That workshop was hosted by Wu Wei, founder of Unique Research. This time, at the opening, we did not talk about large-model trends, nor did we lead everyone into the illusion of a "tool compendium." Instead, we first asked a very simple question:

You yourself—what are you actually using AI for right now?

Note: not whether your company has an AI strategy, not whether your IT department has procured a system, and not whether you have heard of Agent, RAG, OpenClaw, Codex, or Claude Code.

But you yourself.

When you open AI every day, are you truly solving problems, or just treating it as a smarter Baidu?

As soon as this question came out, the differences on-site became obvious.

Some people mainly use AI as a search box, asking questions with DeepSeek and Doubao; some have already put AI to use in writing, proposals, PPTs, cross-border e-commerce listings, and project management; some use Codex, Stitch, and Claude Code for development; and some have started "raising shrimp," connecting OpenClaw into their workflows and trying to let it handle more complex tasks.

But what is truly valuable is not that everyone "is using AI."

It is when everyone is asked to write down a "How do I…" question that the gap truly appears.

How do I use AI to learn English?

How do I use AI to solve team communication and multi-task management?

How do I use AI to manage my antique collection?

How do I use AI to get sales to use CRM correctly?

How do I use AI to build an enterprise knowledge base?

How do I use AI to design a lamp?

How do I use AI to optimize production processes?

You will find that some questions, as soon as you hear them, can continue to be broken down.

For example, "How to use AI to solve team communication information synchronization and multi-task management." Behind this there are real business scenarios, clear pain points, and it can be furtherpressed: where does the information come from? Who needs to be synchronized? What is the synchronization frequency? How are tasks generated? How are responsible persons confirmed? How are anomalies alerted?

But some questions easily hang in the air.

For example, "How to use AI to improve organizational efficiency" or "How to use AI to help enterprises reduce costs and increase efficiency."

These kinds of statements are of course correct, but they are too big. So big that AI can only give you a pile of correct nonsense: optimize processes, strengthen training, improve mechanisms, advance digital transformation.

You see, many times it is not that AI cannot do it, but that the problem has not been defined clearly.

This is especially important for ordinary white-collar workers. Because in the past we were very used to waiting for tasks—waiting for the boss to say what is needed, waiting for the company to set processes, waiting for templates to be sent down, and then being responsible for execution.

But after AI arrives, pure execution is rapidly depreciating.

Writing meeting minutes, revising copy, making spreadsheets, translating materials, organizing PPTs, generating first drafts—these things will not disappear, but they will become cheaper and cheaper.

What is truly becoming more expensive is another kind of capability: defining problems, breaking down tasks, and judging results.

In other words, the most dangerous person in the future workplace may not be someone who cannot use AI, but someone who only says to AI "help me write a proposal."

Because this kind of person seems to be using AI, but in reality is just moving old work habits into a new tool. They have not re-understood the work, nor re-designed the process—they have just changed from "writing nonsense themselves" to "letting AI write nonsense."

This is also a point Wu Wei repeatedly emphasized at the workshop:

In the AI era, asking the right question is 100 times more important than finding the answer.

This sentence is actually a very hard working method.

Because AI's ability to give answers is getting stronger and stronger, what is truly scarce is whether you can compress a vague anxiety into a discussable, breakable, verifiable question.

For example, "I want to improve efficiency" is not a good question.

"I want to automatically organize weekly meeting minutes into a task list and remind responsible persons in Feishu" is a good question.

"I want to build an enterprise knowledge base" is also not good enough.

"I want to start from the IT department, organize system operations, common questions, meeting minutes, and veteran employee experience into a searchable, follow-up-questionable, continuously updatable knowledge base, used to reduce new-hire training costs" is closer to an implementable question.

"I want to use AI for sales management" is too vague.

"I want AI to automatically identify problems in sales CRM such as duplicate reporting, missing information, and unclear project stages, and generate completion reminders" is starting to enter the business.

Once the question is asked correctly, AI is no longer a chat box, but starts to become an entry point for a workflow.

The First Step of AI Implementation Is Not "What to Do," but "What Not to Do"

This is also where enterprises most easily stumble when doing AI today.

Many enterprises, as soon as they talk about AI, their first reaction is to deploy tools, buy systems, build platforms, connect large models. But in reality, the most critical first step of AI implementation is often not "what to do," but "what not to do."

At this workshop, Unique Research used a very practical framework: screening AI scenarios using two dimensions—"business value" and "implementation feasibility."

High value, easy to implement: called quick-win projects, should be done immediately to build confidence.

High value, hard to implement: called strategic breakthroughs, require long-term investment.

Low value, easy to implement: can be used as departmental-level trials.

Low value, hard to implement: should be decisively abandoned.

This framework seems simple, but it is particularly important for many enterprises today. Because AI makes people too easily excited.

Seeing others make digital humans, you want to make them too; seeing others build Agents, you want to build them too; seeing others make knowledge bases, you want to initiate a project; seeing others talk about AI-native organizations, you want all employees to learn AI.

But the question is: does your business really need this?

Is your data ready?

Are your processes stable?

Are your employees willing to use it?

After it is done, who maintains it? Who evaluates it? Who is responsible?

If these questions are not thought through clearly, AI projects can easily become a new kind of digital decoration. It looks advanced, is awkward to use, and in the end everyone draws the wrong conclusion: AI is not suitable for us.

Actually, it is not that AI is not suitable—it is that the scenario was not chosen correctly.

Two Questions, One Sexy, One More Suitable to Start Right Away

From this perspective, the two questions finally selected at that workshop are very interesting.

One is "How to use AI to understand people."

One is "How to build an enterprise knowledge base."

The former seems sexier, involving socializing, dating, team collaboration, and understanding between people. Someone on-site proposed that AI could analyze profile cards and chat records to help users judge the other party's intentions and provide chatting suggestions; someone also reminded that such products should not be overly optimistic, because from an idea to a truly commercializable product, there are still issues of positioning, data, privacy, user scenarios, and differentiation in between.

This discussion is very enlightening. Because it shows that AI not only improves efficiency, but also enters the softer field of interpersonal relationships.

But the more this kind of scenario, the more you cannot rely solely on the imagination that "AI understands people very well."

You must ask clearly: who exactly does it serve? What specific scenario does it solve? Is it dating? Team collaboration? Employee relations? Intimate relationships? Or psychological companionship? If the goal is unclear, the product will become a gadget that wants to do everything but does nothing deeply.

And the other question, "How to build an enterprise knowledge base," on the surface is not so sexy, but it is more suitable for most enterprises to start right away.

Because what many companies truly lack today is not AI, but "knowledge that AI can use."

Policies are scattered in different documents, experience is hidden in veteran employees' heads, customer questions are scattered in WeChat groups, project post-mortems are not organized by anyone, and new-hire training relies on veterans hand-holding. The same question is repeatedly asked, repeatedly answered, and repeatedly stumbled over by different departments.

If you directly deploy AI on this foundation, it is like hiring a very smart new employee, but not giving him any materials, not telling him how the company operates, and then expecting him to work independently the next day.

Unrealistic.

So an enterprise knowledge base is not a "document management project"—it is the foundation of enterprise AI-ification.

The path discussed on-site is also very plain: first establish a knowledge framework, clarifying what each department needs to distill; then hand meeting records, process documents, historical Q&A, and training materials to AI for preliminary organization; then arrange dedicated personnel to proofread, maintain, and update; finally start with a pilot in one department and graduallypromotion to more business departments.

This thing does not sound cool, but it is very useful.

In the future, a very important new capability will emerge in enterprises: turning organizational experience into AI-callable knowledge.

Whoever can structure chaotic experience will let AI truly enter the business. Whoever only buys tools, shouts slogans, and sends notices will in the end most likely return to the old set of inefficient collaboration, just with a few more AI accounts.

If You Are an Ordinary White-Collar Worker, Don't Rush to Ask "Which AI Tool Should I Learn"

So, if you are an ordinary white-collar worker, don't start by asking "which AI tool should I learn."

Tools change too fast—today this one is hot, tomorrow that one is stronger, the day after another new platform pops up. What you should truly do first is to look at your own work again.

Step one: identify the things you do most repetitively and most time-consumingly every day. Not grand career planning, nor abstract capability improvement, but those things you do at least three times a week, that annoy you every time, but have to be done.

For example, organizing meeting minutes, writing customer follow-ups, summarizing data, revising weekly reports, making topic selections, researching materials, generating first drafts.

Step two: break this thing down into a process. What is the input? What needs to be judged in the middle? Who is the output for? What does a good result look like? Which parts must be judged by you? Which parts can be handed to AI?

Many people get no effect from AI because they only give AI a wish, not a process. AI is not mind-reading—it needs context, rules, and examples.

Step three: distill one effective experience into a template. Don't start over asking every time.

For example, when you do industry research, you can fix it into a process of "information collection—key changes—cause analysis—case verification—viewpoint output"; when you do customer management, you can fix it into a template of "customer background—current needs—advancement status—risk points—next actions"; when you do writing, you can fix it into a structure of "counterintuitive opening—core insight—caseunfolded—method summary—action suggestions."

This way AI will become smoother and smoother to use, instead of feeling like meeting for the first time every time.

Step four: retain human judgment.

AI can help you generate answers, but it cannot take responsibility for you. It can help you analyze customers, but it cannot build trust for you; it can help you write proposals, but it cannot judge for you whether the business is viable; it can help you build a knowledge base, but it cannot define organizational rules for you.

In the AI era, the most important value of humans is not doing everything personally, but knowing what to hand over and what to hold in your own hands.

AI Will Not Automatically Make People Evolve; It Will Only Amplify People's Capability Structure

What this workshop truly wanted to solve was not "letting everyone know AI is powerful." That thing no longer needs to be proven.

More importantly, letting every person and every enterprise realize: AI will not automatically make people evolve; it will only amplify people's problem awareness, process capability, and judgment.

The clearer you are about what you want, the more it can help you.

The more confused you are, the easier it is for it to solemnlymanufacturing more noise for you.

Many people are anxious that AI will take their jobs, but more accurately, AI is not taking jobs—it is taking away the sense of security of old working methods.

The advantages you built in the past through proficiency, experience, and repetitive labor now all need to be repriced. But this is not necessarily a bad thing, because AI has also lowered the startup cost of many capabilities.

People who cannot write code can try to make small tools; people who are not good at writing can first build structures; people without assistants can have a work buddy; people without teachers can configure a personal coach.

The key is, you cannot just stand on the shore anxious.

You have to first ask that question that belongs to you.

What thing am I repeatedly consumed by?

Can I break it down into a process?

Do I have enough materials to feed AI?

Can I judge whether the results AI gives are correct?

Can I distill this set of processes so it continues to help me next time?

The first group of people eliminated by AI may not be those who cannot use AI, but those who keep waiting for others to tell them "how to use AI."

The first group of people amplified by AI may also not be the most technically skilled, but those who earlier started asking themselves:

What problem am I actually trying to solve?

One final note: Unique Research has recently been continuously accompanying enterprises in similar AI private-board workshops and scenario mapping. What we care more about is not stuffing a pile of tools into enterprises, but working with the team to clearly define real business problems, screen out high-value, implementable AI scenarios, and then break them into executable paths.

If your team is also stuck at the stage of "knowing AI is important, but not knowing where to start," welcome to chat.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260427-01html
On-site reading page: https://ffcap.cn/en/research/src-20260427-01html