While everyone is still treating AI as “advanced Word”, some people have already used it to open “unmanned companies”. The author of this article proposes that real AI thinking is not to learn a few more models, but to write “intelligence first” into the DNA of the organization: first use the world model to “rehearse the future” in the cloud at low cost, and then let the agent conduct trial and error in parallel, hedge physical costs with computing power, and finally realize the business closed loop of “controlling the real with the virtual”.
A lot of understanding about AI is basically a mess now. Some people say that pattern recognition is not intelligent native, and some people say that pattern recognition is a typical AI algorithm, why not intelligent native. There are similar things about unmanned companies, one-person companies, and so on.
One of the more unique points here is AI thinking, which is very different from Internet thinking, but now no one says it.
Does mastering AI require a new way of thinking, and if so, how to define it?
Application level of AI
We can always use various large models like more advanced Word, in this case, AI is a better tool, at this time there is really no need for AI thinking, just use it often.
But AI is obviously more than just a tool, and multi-agent systems can encapsulate complete business into their own systems. At this time, AI is no longer a simple tool, but has become the main body of value creation.
Of course, there are different levels between the two of the main body of the tool, which is roughly as follows:
The further you go, the more you need a new way of thinking. Otherwise, just like Genghis Khan’s style of play can’t control light infantry, the more you want to advance, the more likely you are to hurt yourself, and if you want to be fast, you won’t be able to reach it.
Smart first
The first and new principle to be followed in “Unmanned Company” is intelligence first.
Pay attention not to the boss, the status quo, etc., but to intelligence.
This is actually the same as AI becoming the main body of value creation, and it is also the premise that AI can truly be effective.
Some people may ask, what if you can’t give priority to intelligence?
Then use AI as a tool and don’t let it be the main body. Otherwise, even if the multi-agent system runs in a short period of time, it will have a certain effect, and it will gradually die.
After 10 years of interaction design, why did I transfer to product manager?
After the real job transfer, I found that many jobs were still beyond my imagination. The work of a product manager is indeed more complicated. Theoretically, the work of a product manager includes all aspects of the product, from market research, user research, data analysis…
View details >
Because the cost of using AI tools is very low (it should be easier than learning Office), the difficulty in AI is not to use tools, but to encapsulate the business based on the basic characteristics of AI.
The most critical combination of the above three and business is AI thinking.
AI thinking
AI thinking is a new problem-solving methodology that must be adopted when we apply the principle of AI First to the organizational process of production and service.
It does not refer to having individuals learn to write code or use AI tools, but rather to thinking and acting on a model that is endogenous to computation and simulation at the strategic and execution level. Its core essence can be summarized in three points: virtual first, large-scale trial and error, and computing power hedging.
1. Virtual-First Simulation: Rehearse everything before you act
Traditional business models follow a linear process of “plan-do-feedback” (PDCA), where every step takes place in the physical world and is extremely costly for trial and error. The first principle of AI thinking is “virtual first”, that is, before investing real resources, create a “world model” in the digital world that highly corresponds to the real environment, and conduct large-scale simulations in it.
This world model, as has been hotly discussed in the academic circles recently, is an algorithmic agent for the real-world environment. It can be narrowed down to being relevant only to one’s own business, and the core goal is not to generate realistic video for entertainment, but to simulate all actionable possibilities in the real world to support purposeful reasoning and action.”
This ability is called “hypothetical thinking” in psychology, and in practice it is what we often call “thought experiments”.
AI has made the cost of thought experiments in this vertical field extremely low.
Whether it is AlphaGo exploring chess paths that humans have never thought of through self-play, or the autonomous driving system predicting the future trajectory of all vehicles and pedestrians on the street, its essence is to deduce countless “possibilities” in a low-cost virtual world to find the optimal solution.
This is the first superpower given to us by AI thinking: to see the future before acting.
2. Scalable Trial and Error (Rapid): Explore the optimal path with parallel computing
Human trial and error is serial, expensive, and limited by personal energy and experience. AI can perform millions, tens of millions, or even hundreds of millions of parallel trials and errors in world models at a marginal cost close to zero.
It can take a marketing team a week to design and evaluate three ad strategies. An AI Agent can generate a thousand combinations of copy and images within an hour, test the click-through rate and conversion rate in a virtual user group, and iterate in real time based on feedback, and finally select the best solutions to put into the real market.
This large-scale, automated trial-and-error cycle increases the speed of innovation by several orders of magnitude. This is equivalent to changing the timeline.
The basis of this ability is that virtual preemption can generate countless “hypothetical trajectories” in which agents can make full use of all “imagined experiences” through reinforcement learning or imitation learning.
In particular, if the cost of trial and error is low enough, it is already in the digital space, so it can also go beyond the virtual first.
3. Computational Hedging: Replacing physical costs with computational costs
The economic basis of “virtual first” and “large-scale trial and error” is “computing power hedging”. This means that we can use relatively cheap computing resources (CPU/GPU time, electricity) to hedge and replace extremely expensive physical world resources (e.g., time, raw materials, human capital, market opportunity costs).
In the past, validating a new drug required years of clinical trials and billions of dollars in investment.
Today, AI can simulate drug-protein interactions in molecular-level models of the world, screening out a large number of ineffective or toxic candidates in advance, narrowing down physical trials to a few of the most likely options for success. Here, millions of dollars in computing power costs hedge against the risk of hundreds of millions of dollars in R&D failure.
Similarly, when a company decides whether to enter a new market, it no longer needs to spend months conducting expensive market research, but can operate a “virtual branch” in a world model that simulates consumer behavior, competitive landscape, and social culture in that market, observe its virtual earnings reports for several quarters, and make final decisions.
If we must find a unified example of the above points, we can review the story of making apps in the mobile Internet era:
You can grind carefully and choose a direction to make an App;
It can also be like a certain company directly is the App array, and the number is good and the one is kept.
Obviously, the key to the latter is not the idea, but the cost of trial and error and the success rate. AI thinking can undoubtedly greatly improve the universality of the latter, no longer limited to making apps.
Unmanned Company: The ultimate organizational carrier for AI thinking
When the above three AI ways of thinking are systematically applied to a business organization, their final form will inevitably evolve into an “unmanned company”.
“Unmanned company” does not mean that there is no one in the physical space, but that its core value creation chain is dominated by AI agents rather than human employees. The role of human beings has changed from a hands-on executor to a designer of goals, rule-setters and value givers.
In such organizations, AI thinking is no longer the “icing on the cake” tool, but the “operating system” on which it survives.
Its technical core can borrow the blueprint depicted by the latest paper: PAN (Physical, Agentic, and Nested) general world model architecture.
- Physical: Unmanned companies need to simulate real-world physical dynamics. For example, an unmanned e-commerce company whose world model needs to understand the complete logistics process of a package from the warehouse to the user’s hands.
- Agent: The core of the company is the agent that acts autonomously. Unmanned companies must support simulations of multi-agent behavior, such as how an agent in charge of marketing and an agent in charge of customer service work together. Its future development direction is to expand from a single agent to the simulation of the behavior of the entire business or social collective.
- Nested: The world model of an unmanned company is layered and nested. It can use LLM-like structures for strategic planning and conceptual reasoning at a high level, and fine physical or sensory details at a low level, such as diffusion models.
Of course, as mentioned earlier, the physical and the agent may not overlap, but in any case, in general, this is destined to be a system that relies on inversion: the virtual controls the real.
To give the simplest example, a typical unmanned company workflow is as follows:
The human founder sets a business goal (e.g., “Increase the ROAS of a product to 2 this quarter”). This goal is fed into the company’s “brain”. Then, multiple AI agents (market analysis agents, advertising creative agents, budget allocation agents, etc.) conducted a large number of simulated delivery experiments in this model sandbox. They “pre-compute and cache the various possible world states, the feasible actions in those states, and their simulation results.” (It is not necessary to simulate, and a small amount of real environment operation can also be done)
Ultimately, an action plan with the highest expected return is selected and automatically executed on real-world advertising platforms such as Google Ads.
From theory to reality: a new wave of business shaped by AI thinking
Although the universal and fully mature “unmanned company” is still the vision of the future, the principles of AI thinking have penetrated into the current business hotspots and shown great power.
Industry and manufacturing: digital twins and virtual factories
Nvidia’s Omniverse platform is a prime example. Automakers create a 1:1 digital twin factory in Omniverse before building a new production line. In this virtual factory, engineers can simulate every movement of the robot arm, test the beat of the production line, optimize logistics routes, and even simulate the operational safety of workers. This is the perfect embodiment of the ideas of “virtual first” and “computing power hedging”, replacing expensive physical installation and rework with virtual debugging.
Content and Marketing: AIGC and Automation Growth
Traditional marketing models are being disrupted. Today, a “one-person team” can use GPT to generate marketing copy, Midjourney and Sora to generate advertising images and videos, and then use automation tools for omnichannel distribution and A/B testing. Behind this is the AI mindset: large-scale content idea generation and effect testing at a very low cost, which used to require a large team to complete. While current video generation world models such as Sora do not yet support interactive reasoning, they have shown the immense potential of AI in idea generation.
Science and R&D: AI-driven hypotheses and validation
The essence of scientific research is the cycle of “putting forward hypotheses, conducting experiments, and verifying conclusions”. AI is accelerating this cycle like never before. For example, AlphaGeometry2, mentioned in the paper, has been able to solve Olympic-level geometric puzzles, which is essentially an efficient “thought experiment” in a purely mathematical world model. Systems like ReasonerAgent can automate literature research and information integration on the web, assisting human researchers in forming and validating hypotheses faster. The Chai 2 mentioned above is also a good example of this.
The future belongs to enterprises that master the “simulation-action” flywheel
We are moving from an “experience-driven” business world to an “analog-driven” business world.
The core competitiveness of an enterprise is no longer just how much capital, how many talents, or how much successful experience it has accumulated in the past. The core competencies of the future will depend on: How fidelity is your world model to the simulation of the real world? How fast is your “simulation-action” flywheel?
Mastering AI thinking means mastering the ability to “foresee the future” and “choose the future” at the lowest cost. An “unmanned company” built on this foundation will have agility, efficiency, and scalability that traditional organizations cannot match. They can adapt more flexibly to market changes, capture the potential needs of users more accurately, and ultimately gain a structural advantage in the competition.
Our ultimate goal is to build AI systems that are “adaptable, resilient, and autonomous unique to human intelligence.” The road is long but full of opportunities, and those companies and individuals who are the first to embrace AI thinking and begin to build their own “world models” and “unmanned company” prototypes will undoubtedly become pioneers in defining the next business era.
Of course, before you can do it, you need to first reduce the incision of the business, and then determine how many levels of simulation are needed.
You can observe and learn first, don’t worry, for example, you can read “Unmanned Company” first…
There is an element of imagination on it, but obviously it will build a real intelligent civilization, which will be a very different world.
