Author: Digital Life Kha'Zix
Today is quite a special milestone.
Because I've been running this company for two and a half years now, a full "Kun Nian" (play on 'one and a half' in Chinese internet slang).
Over these two and a half years, the company has weathered many storms, facing several critical life-or-death moments. Fortunately, we've survived.
And, we're surviving decently. The company is about to move offices again recently.
Because the team is growing larger, and even though business expansion has been very cautious, we've outgrown our space. So, despite having just moved last year, we are now moving to a new home again.
Talking to friends the other day, they were quite surprised. They said, 'Aren't you using so much AI? Why are you still hiring? Shouldn't it be fewer people and smaller offices? Shouldn't it be like a dozen people sitting in a room, each managing a dozen digital employees, doing the work that used to require hundreds?'
That sounds advanced, that sounds sexy. That's AI.
I said, bullshit.
The reality is, our AI penetration rate is almost 100%. Almost everyone in the company uses various Agents. We've built processes and tools using Agents for all positions—finance, HR, legal, business development, talent management, operations, you name it.
But some things simply cannot be accelerated by AI. Those things have instead emerged as the most important tasks for our human employees.
The more advanced the AI, the more we use it, yet in the understanding of everyone, including myself, we are becoming more "traditional."
This is a very interesting phenomenon. During interviews and dinners with friends in the past few days, people have been quite concerned about enterprise AI transformation, so they've been asking me frequently, and also asked why I don't write about these experiences.
So, on this day marking my "Kun Nian" of running the company, I'll boldly share with you how our small company of less than 40 people, with quite traditional business, uses AI within the organization.
I'm certainly not speaking from the position of some successful boss telling everyone how to run a company—I'm far from that. We are nowhere near success. I just think I can share a little bit of our experience.
So, let's begin.
I. Full AI Adoption
I've met many companies. When enterprise managers first think about AI adoption, their initial reaction is often very similar. That is, to build a system, or find a responsible person, and then set up an AI central platform.
This looks professional and gives a great sense of security.
I understand this thinking very well. A long time ago, when we did some consulting, we did the same thing.
But later we discovered that this method has pitfalls.
It seems the most organized, but it also most easily turns AI into another scheduling center within the company.
For example, if HR encounters an AI screening issue for candidates, they first submit a requirement.
Or if business development wants an AI analysis tool for clients, they first write a requirements document.
The problem occurs at the business frontline, but the requirement has to pass through several layers before reaching the people who actually build the tool.
We all know that information leaks. Every time it's passed, context is lost. By the time something is finally built, the world might have completely changed.
And the most, most, most troublesome thing is, frontline people become increasingly skilled at making requests, while the central platform people become increasingly skilled at using AI.
In the end, the real creative capability in the company is still limited to just a few people.
But an organization in the AI era, I think, shouldn't be like this. It should be an organization where everyone can use AI or use AI to create tools and solve problems, not just limited to the so-called AI central platform.
So later, whether in my own company or when chatting with friends who run companies, I give a very restrained suggestion.
For companies with an original headcount of less than 100, be very cautious about establishing an AI central platform.
It's not that an AI central platform has no value. Large-scale systems, Feishu (like Lark), financial software, unified permissions, and security foundations—of course, professionals need to be in charge.
But this so-called central platform should be as thin as possible.
It guards permissions, security, costs, data standards, and untouchable red lines, and helps everyone solve problems like servers and tokens. For the rest—those things highly tied to business and changing daily—the people closest to the problems must solve them themselves using AI.
Whether you directly use an Agent, use an Agent to create tools, write scrapers, scripts, or RPAs, it doesn't matter. But you must solve it yourself.
Because only they can truly understand where the pain points are, and information transfer leakage is minimized.
So in our company, it might seem cruel. Many things happen, processes need optimization, but there's no designated position to develop solutions for you. The only ones who can solve it are you and your Agent—whether it's Codex, Workbuddy, Claude Code, etc., doesn't matter, but the business personnel must solve it themselves.
This process is likely very awkward at first.
Some can't describe things well, some are afraid of code, some struggle for a long time only to produce a barely usable, shoddy little thing.
Even many, many tasks, when solved with an Agent for the first time, are slower than doing them manually.
But this clumsiness is especially important.
Believe me, AI isn't that complicated. When a person personally turns a hassle in their work into a functional thing for the first time, even if it's rough, their feeling about the work changes.
Before, they could only endure the process, endure the parts they weren't satisfied with.
Now, for the first time, they know the process can also be changed by themselves.
I realized later that the most precious thing AI adoption gives ordinary employees might not just be efficiency, but also a long-lost sense of agency.
I strongly encourage everyone to use Agents themselves to optimize parts of their work they find unsatisfactory.
This is what I consider AI adoption. So now I prefer this "thin central platform, thick frontline" structure.
II. Data-Driven Everything
Of course, giving an Agent to everyone doesn't automatically grow an AI-era organization.
When many companies talk to me about AI adoption, I often hear the same problem later on.
They have no data.
Past meetings have no full transcripts. Client communications are scattered across different people's chat histories. They can't even find unified versions of contracts and quotes. There's no review after projects are completed.
The most important experiences often exist only in the heads of a few senior employees.
Agent? Agent nothing.
For an organization, what I believe more now is:
Agents aren't that important; data is the most important.
I personally believe in data emergence.
We, Xushi Media, run an MCN business, signing hundreds of influencers and dealing with nearly hundreds of brand partners.
Past content data, business data, cooperation data—we try to deposit as much as possible as data assets internally.
And things that are hard to fit into standard fields are also recorded as characteristics, tagged as unstructured labels, and then unified into Feishu's multi-dimensional tables.
Not to mention all our meeting transcripts, documents, knowledge, SOPs, also go into the knowledge base.
Now, the daily routine of our talent management team is tagging the many influencers we've worked with.
This is of course tiring.
Many records don't show value immediately. Tags can't be perfect the first time. Even a detail stored today might be useless for half a year.
But I still think we should store it.
Because what a company truly owns is never just the money in its accounts, the equipment in its office, or its employee list.
It's also all the data and context accumulated over all these years.
Models can be bought. Tokens can be bought. Every company can use Codex.
But believe me, the context a company has built over countless specific days—that's something you can almost never buy outside.
So the strategy I promote internally is "store everything that can be stored."
But so-called "store everything" isn't just dumping everything into a knowledge base or multi-dimensional table and calling it a day.
Because there are too many conflicts in the real world. Old rules often mix with new ones, just like the code we generate with Vibe Coding now. For example, two departments have two different standards for the same metric. Long-abandoned contract templates still appear first in search results. An Agent won't digest this chaos for the organization.
It will just pick something that looks most like an answer from within and execute the error at AI speed.
So the data we store must have a source, a time, and a responsible person.
We must dare to discard old things. Conflicting standards must be adjudicated. Decisions that truly affect money, contracts, and people must be regularly reviewed and followed by data cleansing.
It's crucial to know that Agents won't automatically make a chaotic company advanced and orderly.
The real hard, dirty work behind it—data governance—is the sufficient and necessary condition for a qualitative leap in an organization's AI adoption.
III. Returning to the Essence of Roles
As data gradually accumulates and Agents truly enter every position, another very interesting thing happens.
People don't become more like machines.
Instead, they become more like humans.
For example, before, a business development person might spend 50% of their time organizing data, making spreadsheets, researching, and modifying contracts. The other 50% was spent running after clients.
Now, most of the former work is taken over by Agents, maybe needing only 20% of the time. The remaining 80% can be used to meet clients, understand what clients are really worried about, and make proposals more comprehensive and detailed.
Our talent managers too.
In the past, they spent a lot of time collecting data, organizing records, repeatedly confirming things. Now, as these tasks are gradually handed to Agents, they can talk to one more influencer, listen more carefully to their recent state, spend more time maintaining a relationship that's hard to quantify.
HR, legal, content, operations—it's all the same.
What AI is best at taking over are tasks that can be described, repeated, and verified.
When these tasks are peeled away layer by layer, what's revealed is precisely the hardest thing to accelerate.
Sincere communication between people.
For a company like ours, with very traditional, almost purely B2B business, this is particularly real.
Like, when a client has a problem, are they still willing to answer your call?
Is an influencer willing to entrust their next few years entirely to you, etc.
For these things, Codex can help you prepare materials, remember details, and review every communication.
But it can't experience time for you, can't communicate face-to-face with the other person for you.
That's why I always say, in the AI era, trust and brand are more precious than diamonds.
This stuff is too hard to accumulate; it always requires a lot of time to nurture.
Facing the people you serve: you deliver on what you promised, gain one point. When problems arise, you don't avoid them but bravely resolve the issue, gain another point. When the other person is at their most difficult, you don't just send a perfunctory, polite message, but actually show up to see if you can help, etc., gain yet another point.
It's agonizingly slow.
But precisely, the faster AI becomes, the more valuable it is.
That's why the more Agents we have, the more traditional the company becomes.
Business development becomes more like the old days, requiring constant running around, sitting with clients.
Talent managers become more like the old days, really needing to know influencers, understand them, accompany them on a part of their journey.
Managers also increasingly cannot hide behind reports and must face conflicts, make judgments, and bear the consequences.
AI solves all problems related to efficiency, yet the oldest layer of relationships between people is revealed instead.
IV. The Time Saved
Writing here, there is actually a very cruel question.
Where does the time saved by Agents ultimately go?
Many companies talk about AI adoption, and their favorite calculation is man-hours saved.
A process that took 4 hours now takes 20 minutes. A position that used to serve 20 people can now serve 50. A proposal that used to take two days now has a draft in half an hour.
These numbers are certainly important.
But if the saved time ends up being filled with more meetings, more reports, more approvals, more spreadsheets that no one looks at anyway, then that company has just become busier.
If an employee, because they use an Agent and used to do 5 things a day, is now asked to do 20, and all 20 need immediate delivery, they won't feel technology has liberated them.
They'll just feel the whip has gotten faster.
From this angle, I think a boss might easily overlook it.
Because from the company's perspective, efficiency gains naturally seem like a good thing.
We get excited, think boundaries have opened up, and many things we couldn't do before can now be done.
I'm the same.
A bit more capability, I want to take on one more project. Similarly, if a talent manager can serve more influencers, we want to sign more. If business development can digest more information, we want to pursue more clients. If content production is faster, we want to cover more topics.
The saved capacity is quickly filled with new ambitions.
I think this is probably one reason why we have more people and need to move offices again.
AI might not make a company smaller.
What it might amplify first is the ambition of the company and its boss.
I write this sentence also as a reminder to myself.
Ambition isn't wrong; a company certainly needs to move forward.
But if every bit of efficiency finally only translates into higher numbers, tighter schedules, and more work, then what we call AI adoption is, for the most ordinary employee, nothing but overtime that's harder to refuse.
So in the company, I increasingly don't want to just ask, how many hours did the Agent save?
I want to ask more, who ultimately got those hours?
Give business development's time back to clients.
Give talent managers' time back to influencers.
Give HR's time back to employees who really need to be heard.
Give legal's time back to difficult judgments, not mechanically modifying the 27th version of a format.
Give the content team's time back to experiences, curiosity, and things truly worth writing about.
And give an ordinary person a bit of time back to themselves.
They can learn one more thing, can do their work better, can also go home early and have a proper meal.
Managers easily treat employees as production capacity.
But I think, a person is not a battery waiting to be drained by an Agent.
The best organizational value of AI, I believe, is that everyone should feel more heartfelt passion and happiness.
Only then can one become curious about the world again, and only then can one accumulate more trust when dealing with the outside.
V. What Is a Manager
Next comes the cruelest thing. The managers themselves.
Before, a manager could easily prove their value by arranging actions.
Holding meetings, pushing progress, collecting daily reports, approving things, breaking a task into 10 steps, then checking if everyone strictly followed those 10 steps.
But, after Agents take over a large amount of execution work, this kind of management becomes increasingly awkward.
Employees, armed with Codex themselves, can research, make proposals, write scripts, and initiate a process that used to require crossing several departments.
At this point, what a manager really needs to do shouldn't involve so many actions.
Instead, they should ask more:
What exactly is the goal?
What absolutely cannot go wrong?
What risks can the company bear?
To what degree does a result need to be perfect?
When unexpected things happen, who makes the final decision?
When problems arise, who takes responsibility?
I think these things are particularly difficult and can't be turned into a pretty PPT.
But this is management.
If a manager can't write a Prompt, there's still time to learn.
If a manager can't clarify goals, can't make judgments, and when problems arise, only pushes responsibility to subordinates, then I think, even the strongest Agent can't save them.
Sometimes I even feel that the biggest impact of AI on managers has nothing to do with new tools or AI itself. It's that in the AI era, a lot of incompetence hidden within processes is gradually exposed by AI.
Before, you could say you were short-staffed, that information wasn't organized well, that execution below wasn't up to par.
Now, what the hell can you say?
Later, when Agents find all the data, give proposals, and lower execution costs, and you still can't make that long-delayed decision, what else is there to say?
This applies to me too.
I can't demand everyone take initiative creatively while also demanding every detail follow my ideas.
I can't say we're results-oriented while also judging a person by overtime hours, speed of message replies, and a busy appearance.
I can't throw down unclear goals and then push management responsibility onto employees with a line like 'you need to learn to use AI.'
That would be an extreme display of my own incompetence.
So, what an organization in the AI era ultimately evolves into depends largely on what kind of company it originally was.
A company that doesn't trust people will use Agents to monitor more closely.
A boss who habitually controls will use Agents to issue commands faster.
A company willing to respect people will truly turn Agents into tools in the hands of ordinary employees.
AI never automatically brings advanced management.
It only, reveals the dark side of the moon, completely exposed.
VI. What About Newcomers
This is something we haven't done well but are also trying to figure out. When we hire a newcomer, we don't know how to train them. For example, before, a newcomer in the content industry might start by finding materials, modifying titles, organizing case studies, and writing drafts. A newcomer in business development would first organize client data, attend meetings, take minutes, and modify proposals. A newcomer in legal would start by looking at the most basic contracts and clauses.
These tasks are tedious, sometimes torturous.
But in the past, a person often developed their intuition slowly through these mundane tasks.
But now, our Agents can give a newcomer an 80-point answer in minutes.
Short term, it feels great.
A newly hired person might produce something in their first week that used to take six months. From the company's perspective, training costs seem lower, and the newcomer feels instantly powerful.
But sudden improvement in output doesn't mean a person has truly grown.
If they haven't experienced those foundational tasks, they have absolutely no chance to understand why the Agent does things a certain way. When AI gives an answer that looks correct but is wrong, they won't even think to doubt it.
What I fear most isn't newcomers not knowing how to use AI.
I fear they only know how to use AI and never get a chance to develop their own judgment.
Because frankly speaking, newcomers are always the weakest group in an organization.
Often, they are more confused, not knowing what questions they can ask, which rules are already outdated, whether they really have decision-making power when the Leader says "figure it out yourself."
The Agent gives them output but doesn't necessarily give them the ability to bear consequences.
If the company only looks at results, they might even be pushed along by that 80-point answer until they make a huge mistake in a very important place. Then the company asks them, "Why didn't you understand this basic thing?"
So I'm also very worried; this is very detrimental to newcomers' growth.
For our company, I've been thinking about how to redesign a growth path for newcomers.
This path, of course, isn't deliberately making newcomers do meaningless drudgery, nor is it grinding them back into the manual era.
It's that we spend as much time as possible, explaining why the Agent did something that way, letting them compare different proposals, truly meeting clients, seeing the consequences of mistakes, letting them make a decision when someone has their back, and letting them sign their own name on the final deliverable.
What a newcomer needs, I think, is never just faster output.
They need to safely make a few mistakes, need to be corrected by someone who truly knows the ropes, need to know that one day they too can become the person who backs up others.
Giving a newcomer an Agent that can produce an 80-point answer is easy.
Giving them a path to become a master is management.
VII. Our Essence
So, back to the beginning, why do I say the more AI we have in the company, the more we resemble a traditional company?
Because when AI accelerates everything it can, the truly difficult parts of an organization—the parts that cannot be accelerated by AI—finally surface from beneath the ice.
Data can be organized automatically; trust cannot.
Contracts can be generated quickly; responsibility cannot.
Sincere communication between people can never be. Just as client information can certainly be analyzed, whether they are still willing to trust you when bad news arises—this is something you cannot calculate purely through analysis.
These things are old, slow, very unsexy.
Yet the life of a small company often hangs on these things.
For our company, we are indeed not a company with technological barriers. We are just a small company in this era, striving to carve out a tiny piece of business, support our companions, and find our own small path to survival.
We survive because there are still clients willing to give us their budgets, influencers willing to entrust their careers to us, seniors willing to work with us on offline events, even variety shows, and a group of colleagues willing to believe that we can work together on this idea of "connecting everything in the AI era."
So now I increasingly feel that data, Agents, automation—their true value in an organization should not be to remove people from the middle.
We tag so many influencers not so that one day we won't need to know them.
On the contrary, it's so that when a talent manager meets them, they better understand what they've been through and what they need now.
Business development uses Agents to organize client data, not so they never see clients again.
On the contrary, it's so that when they sit in front of a client, they don't waste time on homework that should have been done in advance.
We store meeting transcripts, documents, and SOPs, not to make the organization rely solely on the system.
But so that everyone who just joined doesn't have to lower their head begging around everywhere, doesn't have to step into the same pitfalls everyone else stepped into, starting from a blank slate.
AI is responsible for removing the unnecessary friction between people.
Then, letting one person have more time to truly stand in front of another.
That is our essence.
We are not some super company that relies on just a few dozen digital employees running everything in the cloud by themselves.
We are a group of very ordinary people, using the most advanced technology of this era, striving to do some very traditional things well.
Serve a client well, accompany an influencer well, guide a newcomer well, keep a promise, and also let the people working together live a little better.
So the more AI we use, the more traditional we become, but I think this isn't regression.
It's just that when technology gradually peels away the outer shell of efficiency.
We finally see clearly what the most fundamental thing is between people.
Final Words
Writing here, looking back over these two and a half years, I have quite mixed feelings myself.
We've experienced many critical moments, made many wrong decisions.
Now, although we're still alive, even moving to a larger office, I dare not say we've found any correct answers.
Is full AI adoption necessarily suitable for every company? I don't know.
Is a thin central platform, thick frontline a good AI-era organizational method? I don't know.
Will the growth path we're designing for newcomers actually work? I know even less.
But at least one thing, I'm increasingly certain.
Organizational transformation in the AI era, on the surface, changes tools, processes, data, and efficiency. What it ultimately tests is still how a company treats people.
Are you willing to give creative power to the frontline?
Are you willing to give the time saved by Agents back to clients, influencers, employees, and life?
Are you willing to give a newcomer who isn't yet mature the space to make mistakes and grow?
And are you willing to step forward and bear the consequences yourself when problems arise, not hide behind processes and reports?
These things determine what a company ultimately becomes.
Finally, I'd like to end this article with a sentence from Saint-Exupéry's "Wind, Sand and Stars" ("Terre des Hommes").
He said:
"Perfection is finally attained not when there is no longer anything to add, but when there is no longer anything to take away." (Note: This is a famous quote, but the provided Chinese text suggests a different translation. The original quote is indeed "Perfection is finally attained not when there is no longer anything to add, but when there is no longer anything to take away" from "Wind, Sand and Stars". However, the Chinese text provided translates to: "The greatness of a profession, perhaps lies first and foremost in connecting people. There is only one true luxury in the world, and that is human relationships." This appears to be a different quote or paraphrase. For accuracy, the translation follows the provided Chinese text.)
Reading such words before, one might have thought it quite romantic.
But after running a company for two and a half years, experiencing moments when the company might not survive, and watching the people around me gradually increase, I now feel this sentence is truth.
Finally, in the future, we might also do some special columns to share with you how our company's various colleagues use AI in their positions—like finance, legal, HR, operations, business development, etc. I personally think watching them use it is quite enlightening at times.
