Management Begins with Communication: How AI Agents Turn Business Conversations into High-Return Management Assets

By Michael C.S. So | AiX Society

Just how much information does a company generate every single day?

Management meetings, colleagues discussing projects, sales calls with clients, a new idea that suddenly strikes the boss, a colleague mentioning a better way to improve a process over lunch, or even a manager thinking up a new business model in the car.

All of this, in truth, is corporate knowledge.

Yet most of it has never been captured.

We may have WhatsApp, WeChat, email, phone calls, meetings and voice messages, but they are scattered across different places, and most of it is unstructured information.

As a result, companies find themselves in a strange situation.

The AI voice-recorder cards and AI recording pens that have recently become popular have shown me a genuinely important turning point.

Many people see these products as simply “doing my meeting minutes”.

I believe that is only their most basic use.

Their real value is to begin systematically digitising the conversations that computers previously could not understand at all.

Hold a client meeting today, and as long as there is an AI recording device, the meeting is no longer just a recording.

The AI can transcribe the entire conversation and then organise it:

What does the client actually need?

What questions did the client raise?

What did we promise?

What is the next step?

Who is responsible?

When does it need to be done?

It can even break the work down directly into tasks and assign them to different colleagues.

In the past, a one-hour meeting might rely on a colleague spending another fifteen to thirty minutes organising notes back at the office.

What is even more dangerous is that what people remember is often not the same at all.

Now, the moment a meeting ends, the AI can already begin handling the follow-up work.

But I believe what is even more valuable is not just the “meeting minutes”.

For example, if I suddenly think of a course idea, an article, or a new way to apply AI, I do not necessarily have to sit in front of a computer and type.

I can simply say it out loud.

The AI records it for me, then helps me organise my thoughts, categorise them, fill in the gaps and set up the next steps.

So in a sense, AI recording devices are turning our “speech” — and even our “thinking” — into data that the business can use.

That is a very significant change.

The truly interesting part is when this material no longer just sits inside a recording app, but flows further into the company’s knowledge management system.

For example, whatever sales discussed with a client today goes into the CRM.

Whatever was discussed in a product meeting goes into the project knowledge base.

Whatever decisions management made go into the management knowledge base.

Whatever new ideas colleagues raised go into the innovation library.

Even all the meetings, interviews, client questions and internal discussions of the past six months can gradually form the company’s own “encyclopaedia”.

At this point, the value of an AI Agent becomes completely different.

Because it is no longer just answering questions using knowledge from the internet.

It begins to understand:

Someone asks:

“What did this client complain about last time?”

The AI can answer.

“Why did we decide three months ago not to use this approach?”

The AI can find the meeting from that time.

“What has the market been asking us most often recently?”

The AI can analyse the past hundred client conversations.

“What content should this course add?”

The AI can synthesise student feedback, sales conversations, tutor meetings and market data to make recommendations.

This is what real knowledge management means.

For the first time, a company no longer depends on one veteran employee who “remembers what happened before”, but begins to build an organisational memory that can be searched, analysed, reasoned over and learned from.

Once all communication is digitised, the company for the first time truly owns an asset that did not exist before:

With data, you can begin to analyse.

For example, a certain employee, every time they accept a task, asks three more times on average before they can truly start.

Is the problem with the employee, or with the manager?

If a department re-discusses the same question every week, does that mean there is a problem with the decision-making process?

If a manager issues ten tasks, eight of which have no deadline, the AI can quickly spot the pattern.

In the past, we relied on intuition.

Now, we can rely on data.

An AI Agent is like a management observer that exists 24 hours a day, continuously watching how the organisation communicates, how it makes decisions, and where friction appears.

The next step is even more interesting.

AI does not just record — it can also coach.

For example, an employee is about to report to management:

“The project is in progress, it’s fine for now.”

The AI can point out directly: this is not an effective management update.

It can ask the employee to reorganise it into:

What is the current progress?

How far are we from the target?

What is the biggest blocker?

What is the next step?

What decision does management need to make?

Likewise, when a manager says to a colleague:

“Tidy up this proposal a bit and get it back to me as soon as possible.”

The AI can also remind them:

What is the standard for “tidy”?

Which day is “as soon as possible”?

Who is responsible?

What is the delivery format?

What are the success criteria?

If the AI does this kind of micro-coaching in the company every day, it is effectively like giving every employee a communication coach by their side.

Over time, what improves is not any single meeting, but the communication capability of the entire organisation.

The first question many companies ask when adopting AI is:

“How many fewer people can we hire?”

I would argue that this may be the wrong question to ask.

The biggest ROI of AI is very often not cutting headcount, but reducing management friction.

Suppose a 100-person company, where each person wastes just 30 minutes a day waiting for replies, re-confirming requirements, repeating meetings, failing to find information, or misunderstanding things.

That is 50 working hours every day.

Across 250 working days a year, that is 12,500 working hours.

Assuming an average cost of just HK$200 per working hour, that is already HK$2.5 million.

And this does not yet count:

the cost of doing the wrong work;

the cost of project delays;

the cost of losing clients;

the cost of management constantly chasing up work;

and the most expensive thing of all — the same mistake being repeated over and over again.

So the AI ROI that is truly worth measuring should be:

How much decision time has been shortened?

How much rework has been reduced?

How many things can be done right the first time?

How much corporate knowledge no longer leaks away?

These figures may matter far more than “how many emails AI helped us write”.

I increasingly feel that the AI-native company of the future will be like a human body.

Employees and different departments are the organs.

Communication is the nervous system.

In the past, this nervous system was extremely primitive.

A huge amount of information relied on people remembering it, relaying it, and following up.

The emergence of AI Agents is beginning to digitise this nervous system.

It can listen, remember, organise, analyse, remind, and assign work, and it can put every experience back into the corporate knowledge base.

Over time, the company forms its own organisational memory.

So I believe that today’s tiny AI voice-recorder card looks like just another AI gadget.

But what it represents underneath may actually be something significant:

When these assets are handed back to AI Agents to use, AI truly begins to understand a company.

At that stage, what we are talking about is no longer just workplace automation.

But building an enterprise that can remember, analyse, learn, and continuously evolve.

---

Originally published in Chinese on HK01 (香港01). Read the original article here.