After Two and a Half Years of Startup, the Company Becomes More 'Traditional' After Full AI Integration

Author: Digital Life Kazk

Today marks a rather special point in time.

I have been running my company for two and a half years, and it has officially been a year since we made a significant transition.

Over these two and a half years, the company has faced numerous challenges and critical moments of survival, but fortunately, we have managed to stay afloat.

Moreover, things are going relatively well, and we are about to move again.

As our team and business continue to expand cautiously, we have reached a point where we simply cannot accommodate everyone in our current office, which we just moved into at the end of last year.

A few days ago, while chatting with friends, they expressed surprise, saying, "Aren't you using so much AI? Why are you still hiring? Theoretically, there should be fewer people and smaller offices, right? Shouldn't there be just a dozen people in one room, each with a dozen digital employees, handling the work of hundreds of people from the past?"

This is what AI is about, it sounds so advanced and appealing.

I said, "That's nonsense."

The reality is that our AI penetration rate is almost 100%. Almost everyone in the company is using various agents, and all positions, including finance, HR, legal, business, brokerage, and operations, are utilizing agents to streamline various processes and tools.

However, there are some things that simply cannot be accelerated by AI, and 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 everyone's understanding, including mine, we have become increasingly 'traditional'.

This is a fascinating phenomenon. During interviews a few days ago and at dinner gatherings with friends, many were quite interested in the AI transformation of businesses, so they frequently asked me why I don't write down these experiences.

Therefore, on this significant milestone of my company being one year old, I dare to share how a small, relatively traditional business with fewer than 40 people utilizes AI.

I certainly do not stand in the position of a so-called successful boss telling everyone how to manage a company; we are far from successful. I just think I can share a bit of my experience.

So, let’s get started.

1. Full AI Integration

I have encountered many companies, and when managers talk about AI integration, their first reaction is often very similar: to set up a system, find a responsible person, and establish an AI middle platform.

This seems professional and provides a sense of security.

I completely understand this idea; we have done the same in the past when I was a consultant.

However, we later discovered that this approach has its pitfalls.

It appears to be the most organized but also the easiest way for AI to become just another scheduling center within the company.

For example, if HR encounters an issue with candidate AI screening, they first raise a demand.

Or if the business wants to create an AI analysis tool for clients, they first write a requirements document.

The problem occurs at the frontline of the business, but the demand has to pass through several layers of people before reaching those who actually create the tools.

We all know that information can be lost; with each transmission, context is lost, and by the time the product is finally created, the world may have changed significantly.

The most troublesome aspect is that frontline employees become increasingly skilled at raising demands, while those in the middle become adept at using AI.

In the end, the only people in the company who truly possess creative capabilities are still just a few.

However, I believe that an organization in the AI era should not be like this; it should be one where everyone can use AI or create tools to solve problems, rather than being limited to a so-called AI middle platform.

So later, whether in my own company or while chatting with friends who run their own businesses, I always give a very restrained suggestion.

For companies with fewer than 100 employees, establishing an AI middle platform should be approached with caution.

It’s not that an AI middle platform lacks value. Large systems, Feishu, financial software, unified permissions, and security foundations certainly require professionals to manage.

But this so-called middle platform should be kept as thin as possible.

It should safeguard permissions, security, costs, data standards, and non-negotiable red lines, while solving issues related to servers, tokens, and so on. The remaining highly business-related and constantly changing matters must be addressed by those closest to the problems using AI themselves.

Whether you use agents directly, create tools with agents, write scripts, or do RPA, it doesn’t matter; what matters is that you solve it yourself.

Because only they can truly understand where the pain points lie, and the loss of information during interaction will be minimized.

Thus, in our company, it may be quite harsh; many issues and processes needing optimization arise, but there is no specific position to help you develop solutions. The only ones who can solve it are you and the agents, whether it’s Codex, Workbuddy, Claude Code, or others; it doesn’t matter, but it must be the business personnel who resolve it.

This process is likely to look very unappealing at first.

Some people may struggle to describe their needs, some may fear coding, and some may spend a long time producing something barely usable.

In fact, for many tasks, using agents for the first time may be slower than doing it manually.

But this clumsiness is particularly important.

Trust me, AI is not that complicated. When a person first creates something operational from the troubles in their work, even if it’s rough, their perception of work changes.

In the past, they could only endure processes and tolerate the aspects they were dissatisfied with.

Now, for the first time, they realize that processes can also be changed by themselves.

I later realized that the most precious thing AI integration offers ordinary employees may not just be efficiency but a long-lost sense of initiative.

I highly encourage everyone to use agents to optimize the aspects of their work that they find unsatisfactory.

This is what I believe AI integration is about, which is why I now endorse this structure of 'thin middle platform, thick frontline'.

2. Data-Driven Everything

Of course, simply distributing agents to everyone will not automatically create a so-called organization of the AI era.

Many companies that come to me to discuss AI integration often encounter the same issue.

They lack data.

Past meetings have no transcripts, client communications are scattered across different people's chat records, and they can't even find a unified version of contracts and quotes. After projects are completed, there is no review.

Many of the most important experiences exist only in the minds of a few senior employees.

What good is an agent in this case?

For an organization, I now believe more than ever that:

Agents are not that important; data is the most important.

I firmly believe in the emergence of data.

In our virtual and real media company, we do MCN business, signing hundreds of influencers and connecting with nearly hundreds of brands.

We strive to treat past content data, business data, and collaboration data as data assets to be retained internally.

Moreover, those things that are difficult to fit into standard fields will be tagged as features and placed into multi-dimensional tables in Feishu as unstructured labels.

Not to mention that all our meeting transcripts, documents, knowledge, and SOPs will also be stored in the knowledge base.

Our brokerage team now spends their daily routine tagging many influencers we have collaborated with.

This approach is certainly exhausting.

Many records may not show value at the moment, and tags cannot be perfected in one go. In fact, a detail saved today may not be useful for half a year.

But I still believe it should be stored.

Because what a company truly possesses is never just the money in its accounts, the equipment in its office, and the list of employees.

It also includes all the data and context accumulated over the years.

Models can be bought, tokens can be purchased, and Codex can be used by every company.

But believe me, the context that a company leaves behind over countless specific days is something that can almost never be bought from outside.

Therefore, the strategy I promote internally is to store as much as possible.

However, this so-called storage should not just be about dumping everything into a knowledge base or multi-dimensional table.

Because the conflicts in the real world are too many; old rules have long been mixed with new rules, just like the code we now generate with Vibe Coding. For example, two departments may have two sets of standards for the same metric, and outdated contract templates may still appear at the top of search results. Agents will not digest this chaos for the organization.

They will only pick out something that looks most like an answer and execute it with AI speed, even if it’s incorrect.

Thus, the data we store must have sources, timestamps, and responsible parties.

Old things must be discarded, conflicting standards must have someone to adjudicate, and decisions that truly affect money, contracts, and people must be regularly reviewed and cleaned.

It is essential to understand that agents will not automatically make a chaotic company advanced and orderly.

The data governance behind this is the truly hard, tedious, and dirty work, yet it is also a sufficient and necessary condition for the organization’s AI integration to leap forward.

3. Returning to the Essence of Positions

As data gradually accumulates and agents truly enter every position, another interesting thing happens.

People do not become more machine-like.

Instead, they become more human.

For example, a business person used to spend 50% of their time organizing data, creating tables, researching, and modifying contracts, while the other 50% was spent meeting clients.

Now, with agents handling much of the preliminary work, they may only need 20% of their time for that. The remaining 80% can be spent meeting clients, understanding their concerns, and making proposals more comprehensive and detailed.

Our brokers are the same.

In the past, they spent a lot of time collecting data, organizing records, and confirming details repeatedly. Now, these tasks can gradually be handed over to agents, allowing them to engage more with influencers, listen carefully to their recent situations, and spend more time maintaining relationships that are difficult to quantify.

HR, legal, content, and operations are all similar.

AI excels at handling tasks that can be described, repeated, and verified.

As these tasks are stripped away layer by layer, what remains is precisely the most difficult aspect to accelerate.

The genuine communication between people.

For us, a company with a very traditional B2B business model, this is particularly realistic.

For instance, whether a client is willing to take your call after a problem arises.

Whether an influencer is willing to entrust you with the next few years of their time, and so on.

These things can be prepared with Codex, can help you remember details, and can assist in reviewing each communication.

But it cannot replace the experience of time or face-to-face communication with the other party.

That’s why I always say that in the AI era, trust and brand are more precious than diamonds.

This is incredibly difficult to accumulate and requires a significant amount of time to cultivate.

When dealing with your clients, fulfilling your promises earns you a point. When issues arise, facing them courageously without avoidance earns you another point. When the other party is in their most difficult times, not just sending perfunctory responses but genuinely checking to see if you can help earns you yet another point.

It’s painfully slow.

Yet, paradoxically, the faster AI becomes, the more valuable this becomes.

This is also why, as we integrate more agents, our company becomes increasingly traditional.

Business roles resemble those of the past, constantly going out to meet clients and sitting down with them.

Brokers also resemble those who truly need to understand influencers, accompany them, and walk alongside them for a while.

Managers can no longer hide behind reports; they must confront conflicts, make judgments, and bear the consequences.

AI has solved all efficiency-related issues, yet the most ancient relationships between people have been brought to the forefront.

4. Where the Saved Time Goes

As I write this, there is a rather harsh question.

Where does the time saved by agents ultimately go?

Many companies discussing AI integration love to calculate the hours saved.

A process that used to take four hours now only takes twenty minutes. A position that used to serve twenty people can now serve fifty. A proposal that used to take two days to draft can now produce a first draft in half an hour.

These numbers are certainly important.

But if the time saved is ultimately filled with more meetings, more reports, more approvals, and more forms that no one really looks at, then the company has just become busier.

If an employee, because of using an Agent, originally did five tasks a day but is now required to do twenty tasks, all of which must be submitted immediately, they will not feel liberated by technology.

They will only feel that the whip has been cracked faster.

From this perspective, I think a boss might easily overlook this.

Because from the company's standpoint, improving efficiency naturally seems like a good thing.

We get excited and feel that boundaries have opened up, and many things that we couldn't do in the past can now be done.

I feel the same way.

With a bit more capability, I want to take on another project; similarly, if agents can serve more influencers, I want to sign more influencers; if business can digest more information, I want to pursue more clients; if content production is faster, I want to cover more topics.

The capabilities saved are quickly filled with new ambitions.

I think this is one of the reasons why we have more people and need to move offices.

AI may not necessarily make the company smaller.

What it first amplifies may be the ambitions of the company and the boss.

I write this down as a reminder to myself.

Ambition is not wrong; a company must move forward.

But if every efficiency only turns into higher numbers, denser schedules, and more work, then what we call AI transformation, for the average employee, is just a more difficult overtime to refuse.

So in the company, I increasingly don't want to just ask how many hours the Agent saved.

I want to ask, who ultimately benefited from those hours.

Return business time to clients.

Return agents' time to influencers.

Return HR's time to those employees who truly need to be heard.

Return legal's time to difficult judgments, rather than mechanically revising the 27th version of a format.

Return the content team's time to experiences, curiosity, and truly worthy topics to write about.

Also, return a little time to an ordinary person.

They can learn something new, do their work better, or go home early to enjoy a proper meal.

Managers easily treat employees as mere productivity.

But I believe people are not batteries waiting to be drained by Agents.

The best organizational value of AI, I think, is that everyone should have a deeper love and joy for their work.

Only then will they be filled with curiosity about the world, and when they engage with others, they can accumulate more trust.

V. What is a Manager

Next comes the most brutal aspect: the managers themselves.

In the past, a manager could easily prove their value by arranging actions.

Meetings, pushing progress, collecting daily reports, making approvals, breaking tasks into ten steps, and checking if everyone strictly followed those ten steps.

However, after Agents consume a large amount of execution work, this type of management will become increasingly awkward.

Employees, equipped with Codex, can research materials, create proposals, write scripts, and initiate processes that previously required crossing multiple departments.

At this point, what managers really need to do should not involve so many actions.

Instead, they should ask:

What is the ultimate goal?

What absolutely cannot go wrong?

What risks can the company afford?

To what extent does the result need to be perfect?

In the event of an unexpected situation, who will make the final decision?

If something goes wrong, who will take responsibility?

I find these questions particularly difficult and cannot be neatly presented in a PowerPoint.

But these are what management is about.

A manager who cannot write a Prompt still has time to learn.

A manager who cannot clearly articulate goals, make judgments, and only shifts responsibility to subordinates when problems arise, I believe, will not be saved by even the strongest Agent.

Sometimes I feel that the greatest impact of AI on managers is not related to whether the tools are new or AI-related; it is that in the AI era, much of the incompetence that was previously hidden in processes is gradually exposed by AI.

In the past, one could say there weren't enough hands, the information wasn't organized well, or that execution was inadequate.

But now, what can you say?

In the future, when Agents have found all the materials, provided proposals, and reduced execution costs, if the remaining decision that cannot be made is still delayed, what can you say?

This applies to me as well.

I cannot demand that everyone actively create while also expecting every detail to align with my vision.

I cannot say I am results-oriented while judging someone based on overtime hours, message response speed, and busy expressions.

I also cannot leave unclear goals for others and then use the phrase "you need to learn to use AI" to shift management responsibilities onto employees.

These are, in fact, manifestations of my extreme incompetence.

So, what will organizations in the AI era evolve into? Often, it will depend on what the company was like originally.

A company that does not trust its people will use Agents to enhance monitoring.

A controlling boss will use Agents to issue commands more quickly.

A company that respects its people will truly turn Agents into tools in the hands of ordinary employees.

AI will never automatically bring advanced management.

It will only reveal the hidden aspects of the organization.

VI. What Should Newcomers Do

This is an area where we have not done well but are actively exploring. When we hire a newcomer, we often do not know how to guide them. For example, a newcomer in the content industry might start by finding materials, revising titles, organizing cases, and writing drafts. A newcomer in business might begin by organizing client information, attending meetings, writing minutes, and revising proposals. A newcomer in legal might start by reviewing the most basic contracts and clauses.

These tasks are tedious and can sometimes be quite torturous.

But in the past, many people grew their instincts through these basic tasks.

However, now our Agents can provide an 80-point answer to newcomers in just a few minutes.

In the short term, this is particularly gratifying.

A newly hired person might produce something in their first week that would have taken six months to complete in the past. From the company's perspective, training costs seem reduced, and newcomers feel they have suddenly become capable.

But a sudden improvement in output does not mean that a person has truly grown.

If they have not gone through those foundational tasks, they will have no opportunity to understand why the Agent does what it does. When AI provides an answer that looks correct but is actually wrong, they won't even think to question it.

What I fear most is not that newcomers do not know how to use AI.

I fear that they only know how to use AI but will never have the opportunity to develop their own judgment.

Because frankly speaking, newcomers are always the weakest group in the organization.

Often, they are quite confused, not knowing what questions to ask, which rules are outdated, or whether they truly have decision-making power when a leader says, "You decide."

Agents may provide them with output but do not necessarily equip them with the ability to bear the consequences.

If the company only looks at results, they might be pushed along by that 80-point answer until they make a huge mistake in a critical situation, and then the company asks them why they don't understand this.

So I am also very concerned that this is extremely detrimental to the growth of newcomers.

For our company, I have been thinking about how to redesign a growth path for newcomers.

This path is certainly not about deliberately making newcomers continue to do meaningless grunt work, nor is it about forcing them back to a manual era for the sake of training.

Rather, it is about spending as much time as possible helping them understand why the Agent does what it does, comparing different proposals, genuinely meeting clients, seeing the consequences of mistakes, making judgments when someone has their back, and signing their name on the final output.

What a newcomer needs, I believe, is never just faster output.

They need to safely make a few mistakes, be corrected by someone who truly understands the field, and know that one day they can also be the person who supports others.

It is easy to give a newcomer an Agent that can produce an 80-point answer.

But providing them with a path to grow into an expert is true management.

VII. Our Essence

So returning to the beginning, why do I say that the more AI there is in a company, the more we resemble a traditional company?

Because when AI accelerates everything that can be accelerated, the truly difficult parts of the organization, those that cannot be accelerated by AI, finally surface from beneath the ice.

Data can be automatically organized, but trust cannot.

Contracts can be quickly generated, but responsibility cannot.

Genuine communication between people can never be automated. Just like client information can certainly be analyzed, but whether they will still trust you when bad news arises cannot be calculated purely through analysis.

These matters are very old, very slow, and very unsexy.

Yet the fate of a small company often hinges on these things.

For our company, we are not one of those companies with technological barriers; we are just a small company trying to carve out a little business in this era to support our companions and find our own way to survive.

We survive because there are still clients willing to entrust us with their budgets, influencers willing to entrust their careers to us, and leaders willing to collaborate with us on offline events and even variety shows. We also have a group of colleagues who believe that we can work together to "link everything in the AI era."

So I increasingly feel that whether it is data, Agents, or automation, their true value in the organization should not be to remove people from the equation.

We create so many influencer tags not so that one day we can stop knowing influencers.

On the contrary, it is to help agents understand what influencers have experienced in the past and what they need now when they meet them.

Business uses Agents to organize client information not to avoid seeing clients.

On the contrary, it is so that when sitting in front of clients, we do not waste time on tasks that should have been completed in advance.

We store meetings, documents, and SOPs not so that the organization relies solely on the system.

But to ensure that every newcomer does not have to lower their head and seek help everywhere, nor do they have to step into all the pitfalls that everyone else has already encountered.

AI is responsible for removing unnecessary friction between people.

And then, allowing a person to have more time to genuinely connect with another person.

This is our essence.

We are not some super company that can operate itself in the cloud with just a few dozen digital employees.

We are a group of very ordinary people, using the most advanced technology of this era to strive to do some very traditional things well.

Serve a client well, support an influencer, guide a newcomer, keep a promise, and help those we work with live a little better.

So ironically, the more we use AI, the more traditional we become, but I believe this is not a regression.

It is just that as technology gradually strips away the outer shell of efficiency, we finally see clearly what the most essential things between people are.

In Conclusion

As I write this, looking back over the past two and a half years, I feel quite emotional.

We have gone through many life-and-death crises and made many wrong decisions.

Now, although we are still alive and even moving to a bigger office, I cannot say that we have found the right answers.

Whether full AI integration is suitable for every company, I do not know.

Whether a thin middle layer and a thick frontline are a good organizational model in the AI era, I also do not know.

I am even less sure whether the growth path we are designing for newcomers will ultimately work.

But at least one thing I am increasingly certain of.

The organizational changes in the AI era superficially involve tools, processes, data, and efficiency, but ultimately, it still tests how a company treats its people.

Are you willing to hand over the power of creation to the frontline?

Are you willing to return the time saved by Agents to clients, bloggers, employees, and life itself?

Are you willing to give a newcomer the space to make mistakes and grow, even when they are still immature?

And are you willing to stand up and take responsibility for the results when problems arise, rather than hiding behind processes and reports?

These matters ultimately determine what a company will become.

In conclusion, I would like to end this article with a quote from Antoine de Saint-Exupéry in "Terre des hommes" (Wind, Sand and Stars).

He said:

"The greatness of a profession may lie first of all in the connection it creates between people. There is only one true luxury in the world, and that is the relationship between people."

When I first read such words, I might have found them quite romantic.

But after running a company for two and a half years and experiencing moments when it seemed like the company might not survive, I now see this quote as truth, especially as I watch the people around me gradually increase.

In the future, we may also create some special columns to share how our colleagues in various positions, such as finance, legal, HR, operations, and business, are using AI. I personally find it quite inspiring to see how they apply it.