Artificial General Intelligence (AGI) is already here.

Recently, I've spent a lot of time debugging the product. While doing so, an idea popped into my head:

AGI is not something that will happen in 3 - 5 years; it's already here.

This is a recursive process, so both the depth and scope will continuously increase during the development. However, this doesn't change the fact that it's right here, not something in the future but present now.

When AI can fully cover all the required functions in any role (such as programming), it is actually AGI because any role requires a large amount of comprehensive judgment.

Then why don't we perceive it clearly?

It's roughly similar to the situation after the tank was invented. It was taken to Siberia, and after a while, people concluded that a tank was inferior to a dog - drawn sled.

This will obviously trigger a series of chain reactions.

From AI - native to the Unmanned Company

Recently, the "Opinions of the State Council on Deeply Implementing the 'Artificial Intelligence +' Initiative" mentioned AI - native enterprises in a prominent position:

AI - native is not just a simple technology. It's like a thinking mode after the matching of technology and organizational models, which transforms the production process of products and services.

(This picture was originally drawn by Ericsson and then optimized. It's okay)

This comprehensiveness leads to an awkward fact: it's difficult to achieve two - way understanding. Those who understand technology may not understand organizational operations, and those who understand organizational operations may not understand what's happening in the AI world.

If we believe that AGI is already here, then the above point will undoubtedly become a key obstacle to implementation.

How to build a system that seamlessly connects human knowledge with local AGI should be the biggest challenge at this moment. Of course, this is also the biggest challenge for realizing AI - native.

And the ultimate goal of all these efforts is the "Unmanned Company" or Level 5 of AI.

Everything Can Be Reconstructed

A philosophical way to describe change is: Everything is a number, and everything is in ultimate flux.

Limited by the short lifespan of humans in history, we don't often experience significant changes.

But recently, history seems to have accelerated, and major technologies like the Internet and AI have emerged one after another.

Especially recent AI has evolved at a much faster pace. From an application perspective, today's AI is completely different from the AI in 2022.

This speed is the essential characteristic of AI - driven change: It evolves faster than you.

Human evolution is difficult, so training in companies often yields little result.

Process re - engineering is relatively faster. However, since it involves many interest groups and it's difficult to accurately calculate the adjustments, process re - engineering is slow.

But AI is different. AI has risen at an extremely high speed and completely reshaped another way to achieve goals.

This is the core reason why AI - native is bound to come, and its result will almost certainly be that everything can be reconstructed.

(It's a schematic diagram)

Take the relatively mature field of programming as an example. Traditional software development is a highly specialized and collaborative process. The birth of a project usually requires product managers to conduct requirement analysis, architects to design the system blueprint, front - end engineers to handle the interface, back - end engineers to process the logic, and test engineers to ensure the quality... It's like a "symphony" completed through the collaboration of different roles and functions.

In the "AI - native" model, this symphony is being simplified into a "solo performance". In this new model, we can input relatively detailed requirements in natural language to large - model coding assistants like Claude Code or others and then tell it where to make corrections.

A few days ago, a friend who works on algorithms told me: This thing is really damn useful. I completed a delivery without writing a single line of code and accomplished in one day what used to take a team weeks to do.

Although the forms of production are different, the results of the above two paths are the same, and the latter may even be better.

For programming, the latter is AI - native, while the former is not, even if everyone in the former also uses AI tools.

AI - native is a value - creation system with AI as the main body and the principle of intelligence first.

AI takes on the role, like a flying kite, and humans only need to move the string in their hands at most.

In the AI - native form, AI becomes the main body of value creation. The collaboration between AI and AI replaces the past complex organizational process. The organization is internalized into the relationship between intelligent agents.

Recursive Process and the Unmanned Company

The collapsible organization and business improve as the intelligence level rises, so this is a recursive process. It could start with a programming team, then recursively move to operations. Deploying a service on the cloud may no longer require a large - scale operations team. Then it will cover many past functions and finally the entire company.

(Schematic diagram: recursion)

Coincidentally, while I was writing this paragraph, a former colleague who was in charge of operations sent me a message. He said:

Now I can complete the services of k8s, log elk, monitoring prometheus, mysql, redis, and mongodb in a week. Moreover, I've not only solved the PaaS but also the IaaS by using AI programming. Now I can deploy the desired services on machines at will between self - built IDCs and public clouds. If I connect the network in advance, it's a hybrid cloud.

This was one of our goals when we worked together. It was extremely difficult back then, and we felt it would take a year.

Now, this has also changed.

This is also the built - in logic of OpenAI's five - level model. An agent and an organization are essentially the same thing, just facing different scopes and having different complexities.

OpenAI probably had other thoughts back then. Its organization has great social significance, not the kind of organization we usually understand.

The most crucial point here is, of course, the evolution speed mentioned earlier.

The evolution speed of AI affects the number and depth of recursions and, of course, the choice of value - creation points.

In a process of ultimate flux, where the value model is rapidly deconstructed and reconstructed, there are no eternal products and values, but the time windows do vary.

This is why, after working on tools for nearly 8 years, I no longer do it. The sustainable time window has been significantly shortened.

Your business cycle and commercialization cycle may be longer than the technology replacement cycle, so this is a dead - end path.

The core method to break through is to keep running fast, but the financial environment makes the possibility of running fast almost zero.

What Needs to Be Mastered Is Not Technology but the Value - Creation Paradigm

In the context of rapid folding, what really needs to be mastered is not a particular technology. AI has made the cost of using technology extremely low.

What needs to be mastered is the value - creation model. What kind of model can maximize the power of AI?

Although ChatGPT seemed like a tool when it first came out, after GPT4, the shackles of being just a tool were actually broken.

One day, my classmate Zhou Wei sent me a screenshot in a group:

As soon as I saw it, I thought: The more AI develops, the less valuable execution becomes!

At this time, the paradigm undergoes a fundamental shift. The key is no longer how to use AI technology but how to encapsulate business with AI.

The key to encapsulating business is to find the real - world boundaries of AI (which are often composed of data and tools).

Then, put them into a changing model, which requires constantly breaking down the barriers to AI application and paving the way for its power to be fully exerted.

From a small - to - large perspective, the smallest things are the various tools we see now. Starting from these fragmented tools, we can't stop. Stopping means death because there is the powerful folding force of large models behind us.

From a large - to - small perspective, it's the unmanned company, which directly handles the final business in an AI - native way, focusing on sales and cash flow. It's more difficult to start because we need to deal with all the aspects that AI can't reach, such as lack of data, knowledge, and tools. We need a system to complement all these.

There is no absolute right or wrong, but what needs to be mastered has indeed changed. If we can't raise our cognitive perspective, it will be very fatal. It's like trying to find a sword by marking the boat, and it's easy to end up like digging wells in the desert with no water in any well.

To Master the New Paradigm, AI Thinking Is Required

The principle of intelligence first can be refined into a series of more specific thinking modes.

For example, intelligence first almost certainly corresponds to virtual first. The way of virtual first is large - scale trial - and - error. The fundamental reason why all these work is that we can use computing power to hedge all uncertainties.

Intelligence first will also inevitably lead to the re - definition of the role boundaries between AI and humans.

AI assisting humans and humans assisting AI may seem similar in many cases, but they are fundamentally different.

It's almost certain that we will enter a world of dependency inversion. We live in the physical world, but everything in the virtual world guides this physical world.

This may be the true nature of the world, but it has been obscured by various barriers.

After all, in the eyes of philosophers, the real world is diverse, and the so - called true existence comes from the space of universals.

So, AI thinking is a thinking mode that prioritizes the digital and intelligent space.

Reflexivity and Heat Death

Since the publication of "The Unmanned Company", there have been many feedbacks and discussions. The most interesting question is: What will happen if every company becomes an unmanned company?

In the pure economic world, the basic principle is that the more intelligent one wins.

But if the intelligence levels are the same, the information gap and capabilities will also be the same. At this time, reflexivity will cause the world to be calculated to change at any time, with large - scale calculations occurring at any time, but the returns won't keep up.

With only costs and no returns, the existing system will enter a state of heat death and then collapse.

If it rises in a spiral, it will enter a new civilization state. There is a little mention of this at the end of "The Unmanned Company".

This can be regarded as another meaning of the unmanned company. Technology makes the business form move towards the end of the existing form and start something else.

Summary

It's not difficult to verify that AGI is already here. You just need to deeply apply the model. The difficult part is really changing the world. If the world doesn't change, the capabilities of AI can't be fully exerted, and it will be like cutting the feet to fit the shoes.

What's the next step?

Do you have the power to re - define roles and processes within a certain scope? Only when combined with AI thinking can AI produce results.

This article is from the WeChat official account "Thinking about Things", author: Lao Li Talks One, Two, Three. Republished by 36Kr with authorization.