If large models were to be commercialized solely through providing APIs, such a model might not even be sustainable for a company like OpenAI. Integrating with various industries and domains is essential to expand application boundaries. As intelligence merges with these sectors, applying large models will likely take the form of multiple Agents. This shift, though seemingly small, has the potential to change the rules of the game fundamentally. After all, making bows doesn't necessarily equate to being skilled in hunting!
1. The Scale of Intelligence
Whether Agents are feasible and can be widely applied hinges critically on the intelligence scale; simply put, it's about what intelligence can achieve. The level of intelligence leads to different sources of power. The potential applications of using animals versus steam engines as power sources are vastly different.
When the internet was first introduced, it was a purely new concept, unfamiliar
to most people, leading to relatively simple perspectives focused solely on the internet. However, intelligence is different. Every individual is an intelligent being with their own understanding of intelligence, leading to myriad interpretations that can overwhelm and intermingle.
Thus, in creating intelligent products, it's crucial first to clarify the intelligence scale.
All constructs of human civilization are based on intelligence, so theoretically, if the progress of intelligence is comprehensive enough, it could lead to a complete reconstruction of society. However, creating products based on reconstructing civilization is like drifting into the unknown.
With a solid anchor, the rhythms of capital, product development, and business operations can be balanced, leading to a situation akin to a Ponzi scheme: inflating expectations versus a reality of hesitation, fostering a breeding ground for a game of hot potato. The real competition is about who can run faster. Ultimately, it's hard to be sure who the real winner is. The scale of intelligence provides this anchor for products. However, it's challenging to establish a parametric scale due to the inherent generality of intelligence (hence the term' general intelligence'). Any metrics or test sets are inevitably partial and insufficient as a reasonable scale (like using various rates expressed in percentages — they are helpful but not a scale of intelligence). Using them is like using weight to measure someone's intelligence.
Returning to a Turing Test 2.0, which incorporates randomness, becomes necessary in this context.
The evaluation of smart speakers reflects this mismatch. Everyone perceives them as intelligent beings with general intelligence, but the manufacturers can only conduct limited testing and provide a finite set of responses. The ultimate conclusion is that smart speakers could be more intelligent.
Defining the scale of intelligence for Agents is just the first step.
Agent-type products are more complex than large models, akin to a pure mathematical genius entering the market industry. The inevitable multi-dimensionality of thought processes requires a blend of logical clarity and understanding of human emotions and relationships.
The shift begins with the realization that an Agent is not just a single point but must handle entire business scenarios along extensive chains. Focusing on a single moment is ineffective; more comprehensive optimization is needed.
Building competitive strength at a single point may not be viable for agents, but overall balancing might be the key.
2. From Focusing on Single Points to Balancing Optimization
When the market is sufficiently pure, a focus on single-point excellence is required.
For instance, with large models themselves, conveying a sense of accurate intelligence is essential. Aspects like sales and branding are secondary.
This is even more pronounced in the early days of the internet as a tool. If you don't position your product at a sufficiently strong demand point and make it highly usable, your operational costs will overwhelm you. OpenAI essentially relived this story. The first wave of AI tools, like image and video generators, followed a similar path. However, Agents are different.
When intelligence merges with specific domains, the chain of operations significantly lengthens, and more than standalone technology is needed to leverage the market. Can your best technology impact the chemical industry, for example?
This necessitates a coordinated effort involving pre-sales, sales, product, technology, delivery, and after-sales. In other words, no single aspect can fully resolve issues; it's about overall harmony. Achieving this harmony requires balanced optimization.
It's akin to prescribing medicine, where principal and complementary ingredients exist, and the exact amount of each varies with different situations. But underlying this balancing act is a dominant rule.
Behind this balance lies an age-old concept: organizational strength.
This is crucial because, to date, there's no shortcut to significantly enhance organizational strength.
The challenges are far more significant for a startup to transform into a coordinated organization than focusing on a high-profile product.
Understanding this balancing act for entrepreneurial teams means actively participating in the field rather than cheering from the sidelines. Often, a team's expectations can be at odds with reality, whether in terms of economic returns or the pace of success, presenting substantial challenges. However, the real issue isn't just the divergence of expectations but that new teams inherently lack an advantage in balancing critical elements.
Industry veterans are naturally better at managing this balance. New teams tackling balance-critical tasks are akin to a less efficient entity competing against more efficient teams.
However, just as Microsoft disrupted IBM's dominance and Google weakened Microsoft's hold, the success of Agents demands a new approach. This approach must provide a tenfold efficiency in balancing.
This tenfold efficiency returns to the essence of intelligence. For Agents to succeed, they must reconstruct the components of organizational strength based on intelligence.
3. Organizational Strength
The key to influencing balanced optimization is organizational strength, which essentially encompasses various elements discussed in traditional MBA programs, such as culture, organizational structure, and tools. These elements collectively form the comprehensive power of a company, which is organizational strength.
Different starting points (like the values or even the personality of a startup's founder) can give birth to different styles of organizational strength. Some may resemble the zombies in "Plants vs. Zombies," while others might be akin to the T1000 in "Terminator," capable of achieving a blend of flexibility and rigidity.
Previously, there was no shortcut to this path. It was time-consuming, costly, and depended mainly on luck. These factors are necessary for organizations to avoid ending up in a mediocre state, engaging in mutual undercutting.
However, a significant change is now occurring because the critical element of "people" within companies is evolving.
Imagine a company with only two people but a thousand intelligent agents. Its organizational strength would be equivalent to the capability of those algorithms — intelligence becomes organizational strength. This represents a fundamental difference from an organization of a thousand people.
So, can companies transform in this way? And when?
This brings us back to the scale of intelligence and the Turing Test 2.0. The progression of these scales is gradual, but such changes are already occurring at the fringes. For example, in coding, tasks that used to require a team of five to ten people can now be handled by one advanced programmer. Under these circumstances, we'll likely hear more about small teams, like MJ's, generating significant revenue with limited personnel. Even if they don't reach the scale of MJ, many small teams around us are using similar methods to develop their products.
As this becomes more widespread, knowledgeable native organizations will emerge. In such organizations, intelligent agents will be integrated wherever possible, avoiding needing a large workforce for detailed division of labor and coordination. Every step forward in these organizations will push the boundaries of intelligence, expanding the human core only when necessary.
When organizational strength undergoes such a transformation, and we return to the perspective of balanced optimization, it becomes apparent that fundamental opportunities are emerging.
Traditional organizational balancing results from management and operational refinement, whereas algorithms and intelligence levels govern the new style.
These new organizations can produce tenfold efficiency and have a universal appeal. Meanwhile, the competitive advantages previously embedded in organizational balancing could become liabilities.
The old balancing relied on a specific production relationship. Breaking and upgrading this relationship means undergoing a transformation, which could be incredibly challenging and have a low chance of success.
4. Intelligence-Native Organizations vs. Traditional Organizations
The organizations emerging from this evolution are intelligence-native, as intelligence is inevitably at their core. When discussing intelligence-native entities, some people say, "A definition could be that an application which ceases to function without intelligence is intelligence-native." However, this definition seems too absolute. For instance, if a weather app only offers facial recognition login, it's intelligence-native, but if it provides both facial recognition and password options, then it's not.
So, while Ericsson's definition is mostly accurate, this article expands upon the potential gradual progression of such a concept in organizational contexts.
There isn't a precise definition of AI-native. Among the definitions I've encountered, Ericsson's comes closest to what was discussed earlier:
This article adds a perspective: intelligence-native organizations are more likely to emerge in peripheral areas initially and then gradually move toward the center stage. They are starting from something other than large, well-established companies like Ericsson.
Most enterprises will still view intelligence as a singular tool in the early stages of developing Agents. The most crucial competitive factor — balanced optimization — remains unchanged from the past, hence the likelihood of high failure rates.
5. The Great Fold
When intelligence-native organizations truly emerge, an accelerated folding of the existing economic system is inevitable.
This process resembles how agriculture once constituted maybe 80% of the economy but now accounts for only about 10–20%.
During this transition, the total economic volume, including both individual and collective aspects, will be many times larger than it is now.
This marks the end of the old model and the beginning of a new one.
What should be the initial setting for this new model, and what kind of initial setting would lead to a positive trend, is indeed an interesting topic.
6. Summary
Intelligence must transcend various boundaries, after which Agents in different fields can gradually establish themselves. These boundaries can be measured using the Turing Test 2.0. Once these limits are crossed, organizational methods will likely be restructured. Only after the essence of organizational strength has changed can a tenfold efficiency be universally achieved, impacting industry after industry. At this point, the various accumulations embedded within corporate relationships across different sectors might become liabilities, ironically paving the way for new opportunities from large models to Agents.