Bloomberg: How Artificial Intelligence Will Disrupt the Way Companies Are Organized?
TechFlow Selected TechFlow Selected
Bloomberg: How Artificial Intelligence Will Disrupt the Way Companies Are Organized?
Economic systems have long been built on the notion that expertise is scarce and expensive. Artificial intelligence is about to make such expertise abundant and nearly free.
Navigating Web3 tides with focused insights
Image source: Generated by Wujie AI
For most of human history, hiring a team of experts with PhDs would require a massive budget and months of preparation. Today, typing just a few keywords into a chatbot instantly grants access to the collective wisdom of these "minds."
When intelligence becomes cheaper and faster, the foundational assumption underpinning our social systems—“human insight is scarce and expensive”—will collapse. When we can summon insights from dozens of experts at any moment, how will corporate structures evolve? How will our approach to innovation change? And how should each of us think about learning and decision-making? The critical question facing individuals and organizations alike is: when intelligence itself is universally accessible and nearly free, how will you act?
The Historical Decline in the Cost of Wisdom
We have witnessed dramatic drops in the cost of knowledge and explosive expansions in its accessibility multiple times throughout history. The invention of the printing press in the mid-15th century drastically reduced the cost of disseminating written materials. Before that, texts were painstakingly copied by hand—often by monks or other specialists—making them both expensive and time-consuming.
Once this bottleneck was broken, Europe experienced profound societal transformations: the Protestant Reformation shook religious institutions; literacy rates surged (laying the foundation for universal primary education); scientific research flourished through printed publications. Commercially oriented nations like the Netherlands and Britain benefited immensely—the Dutch entered their "Golden Age," while Britain went on to play a dominant global role for centuries.
Over time, widespread literacy and public education elevated overall societal intelligence, laying the groundwork for industrialization. Factory jobs became increasingly specialized, and more complex labor divisions fueled economic growth. In the late 18th century, countries with higher male literacy rates industrialized first; by the end of the 19th century, the most technologically advanced economies were also those with the highest literacy rates. As people acquired new skills, more specialized roles emerged, creating a virtuous cycle that continues today.
The advent of the internet pushed this trend even further. As a child, researching a new topic meant spending half a day searching library catalogs with pen and paper. Back then, acquiring knowledge was costly and cumbersome.
Now, artificial intelligence has taken up the baton in this thousand-year journey of reducing the cost of intelligence, opening a new chapter for our economy and way of thinking.
My “Eureka Moment” with ChatGPT
When I first used ChatGPT in December 2022, I immediately recognized it as a milestone product. At first, I used it for trivial "number tricks," such as asking the AI to rewrite the Declaration of Independence in the style of Eminem (it came up with something like "Yo, we're speaking out loud, these people won't be brought down," and so on).
In hindsight, this was like asking a Michelin-starred chef to make you a grilled cheese sandwich—utterly wasteful. It wasn’t until an afternoon in January 2023, when my 12-year-old daughter and I spent a few hours using ChatGPT to design a brand-new board game together, that I truly grasped the power of this tool.
I started by telling the AI which board games we liked and disliked, and asked it to identify common patterns. It found that we preferred mechanics involving "path-building," "resource management," "card collection," "strategic planning," and "high uncertainty in outcomes," while disliking certain elements common in games like Risk or Monopoly.
I then asked it to generate novel, non-obvious game ideas based on these preferences, ideally with historical context. ChatGPT proposed a game called Elemental Discoveries: players take on the roles of 18th–19th century chemists who conduct experiments, gain points, trade resources, and sabotage each other.
Next, I prompted it to refine the resource system, gameplay mechanics, and playable character roles. It suggested archetypes like "Alchemist," "Saboteur," "Merchant," and "Scientist," pairing them with real historical figures such as Lavoisier, Joseph-Louis Gay-Lussac, Marie Curie, and Carl Wilhelm Scheele.
Using what was still a relatively "primitive" version of ChatGPT, we created a playable—if rough—board game within two or three hours. Eventually, I had to stop, not only due to time constraints but also sheer mental exhaustion. That experience gave me firsthand insight: an AI "collaborator" can compress development cycles that would normally take weeks into just a few hours. Imagine applying this to product development, market analysis, or even corporate strategy—what immense potential could be unlocked?
In that process, I didn’t see ChatGPT merely repeating facts or stitching together information. Instead, it demonstrated analogical and conceptual thinking—connecting ideas and real-world references to deliver genuinely creative solutions tailored to our needs.
From “Stochastic Parrots” to “Deep Thinkers”
A trillion is already an astonishing number. The large language models behind ChatGPT often contain billions, hundreds of billions, or even trillions of parameters—an unfathomable level of complexity.
We still don’t fully understand why or how these models work. As they repeatedly broke new ground over the past seven years, some theoretical researchers insisted they couldn’t produce anything truly novel. In 2021, some even coined the derogatory term “stochastic parrots,” arguing that large language models simply predict text based on statistical patterns in training data—like parrots randomly repeating phrases.
Yet for those of us who have continuously used and marveled at these tools, it’s hard to believe they’re merely regurgitating content. Especially in the past six months, that skepticism has become increasingly untenable.
Early versions of large language models were more like “intuitive speakers”—lacking self-reflection or any sense of self-awareness. Using Nobel laureate Daniel Kahneman’s framework, humans typically rely on System 1 thinking (fast, intuitive, automatic), but switch to System 2 (slow, deliberate, less error-prone) for deeper reasoning. Early ChatGPT and similar models mostly exhibited System 1-like behavior, without true System 2 reasoning capabilities.
This began to change in September 2024, when OpenAI released a reasoning model called o1. This model can break down multi-step logical problems, verify intermediate conclusions (and backtrack or correct if necessary), and arrive at more reliable final answers. Unlike traditional large language models that rely solely on memory or surface-level pattern matching, these new reasoning models are gradually developing the ability to deconstruct problems and reason carefully. Some tests show they now match—or even surpass—PhD-level experts in specialized domains.
Since the release of o1, AI has advanced at an astonishing pace in just six months. The hottest topic now is turning these reasoning models into autonomous research assistants—and their performance is truly impressive.
Recently, I tasked a research agent with analyzing the environmental impact of large-scale operations such as Formula 1 races, Coachella music festivals, Disneyland, Las Vegas casinos, hospitals, and major zoos. The AI spent 73 minutes consulting 29 independent sources and delivered a detailed table along with a 1,916-word explanatory report. While there’s still room for improvement—roughly equivalent to a graduate student’s multi-day effort—it saved me days of work.
Just 18 months ago, my AI systems could only handle small tasks taking half an hour or less. Now, they’re capable of tackling far more complex, time-intensive research assignments.
The Emergence of Cognitive “Production Lines”
We’ve been witnessing an ongoing evolution in how we use knowledge and perform cognitive labor. From ancient temples and scholars monopolizing wisdom, to the printing press enabling mass dissemination, to the internet making information instantly accessible—the challenge has progressively shifted toward “how to interpret information.” Now, tasks once considered rare and complex are becoming readily available and inexpensive.
Yet when I speak with executives at large corporations, I find most are only using AI in narrow, low-value areas—such as automating customer service to cut costs. The CEO of Salesforce noted last December that 86% of their 36,000 weekly customer support queries are now handled by AI. Swedish fintech company Klarna claims two-thirds of its customer conversations are managed by AI, generating $40 million in profit from this single initiative. But saving 10% on costs via customer service alone isn’t enough to transform a business. No great company has ever risen to prominence purely through cost-cutting.
Most companies start with lower-tier tasks—using AI to automate $50-per-hour jobs like customer chats. Useful, yes—but far from transformative. The reality is that AI is equally capable of performing tasks worth $5,000 per hour—such as R&D, strategic planning, or expert consulting. So why are only a few companies deploying AI in these high-impact areas?
One reason is the difficulty in imagining that work traditionally reserved for senior managers or top-tier experts could be (at least partially) performed by machines. Because exceptional talent is scarce, high-value tasks have long been seen as inherently valuable. Our organizational designs are built upon the assumption that “genuinely intelligent people are in limited supply.”
Take the pharmaceutical industry: a single blockbuster drug can determine a company’s fate. The bottleneck lies in advancing compounds through a costly, lengthy approval process—typically taking 10–15 years and over $1 billion in investment, with only one success among thousands of candidates. Meanwhile, a major pharmaceutical firm may employ thousands more marketers than elite researchers, simply because world-class scientists are extremely rare.
Today, most business leaders remain in the phase of “trying to accept AI” rather than “truly trusting AI.” They’re conditioned to avoid problems perceived as too difficult or expensive. But with AI, the constraint is no longer “can we come up with a solution?” but rather “how quickly can we test and validate good ideas?”
All of this will have profound implications. When every company can instantly access several “PhD-level AI experts,” the pace of innovation will naturally accelerate dramatically. Just as Henry Ford’s assembly line enabled rapid iteration and refinement in manufacturing, AI allows ideas and solutions to be continuously improved. Companies can experiment faster, learn quicker, and pivot with unprecedented speed.
Of course, if an organization lacks the ability to execute and integrate ideas proposed by its AI “brain trust,” even the most brilliant suggestions will go to waste. The real competitive edge will lie in execution and integration.
My Daily Life with AI
Over the past 18 months, I’ve gradually built an “AI ecosystem” to support my work. For example, in June 2024, I invoked these AI systems 38 times in a single day, exchanging a total of 79,000 words for research purposes.
By January 2025, I stopped counting word volume altogether. Unless objected to by others, I now bring an AI assistant to nearly every meeting for real-time note-taking. In daily research, I routinely use multiple different AI tools. Just during the week I wrote this article, I issued at least 144 queries to various large language models—not including 26 voice-to-text transcriptions or usage of code-assistant tools. I now use next-generation AI tools more frequently than Google Search.
Surprisingly, despite handling a heavier workload at greater speed, I spend less time in front of my computer screen than I did years ago—a welcome personal benefit.
When the cost of intelligence approaches zero, the real bottleneck is no longer “how to get smart minds,” but “how well we use them.” Individuals and organizations that excel at asking the right questions, objectively evaluating answers, and acting decisively will emerge as winners. They must also ask themselves: now that I have more time, what should I do with it?
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
