Organizational Capital: How to Make 1 + 1 > 2

This discussion will explore how to make a group of excellent individuals form an outstanding company. You might think that merely gathering top talent will lead to great success, but that’s not always the case. Let’s look at an ongoing story.

In the summer of 2025, with its own large models performing below expectations, Meta faced the prospect of falling behind the leading AI group. Mark Zuckerberg could no longer sit idly by, directly orchestrating an expensive, talent-driven “Avengers” assembly. He first invested $14.3 billion in Scale AI, inviting its founder, Alexandr Wang, to lead the newly established “Superintelligence Labs.” He then heavily recruited dozens of key R&D personnel from OpenAI, Google DeepMind, and Anthropic, with some reported salaries as high as $300 million over four years, exceeding those of top NBA stars [1].

You might ask, with such a star-studded team, possessing capital, computing power, and data, shouldn’t they, if not immediately reach the top, at least return to the leading tier?

However, that wasn’t the case. Just two months after its formation, eight people left this all-star team, some even returning to OpenAI in less than a month [2]… Now, a year later, they’ve finally managed to release a model, yet it still can’t rank among the top tier alongside GPT and Claude.

This is not an isolated incident. As of this writing (June 2026), Google’s models are also half to one step behind OpenAI and Anthropic. Other major Silicon Valley companies, like Apple, Amazon, and even Microsoft, have long since fallen out of the leading group.

Founding a company is not a simple chemical reaction; merely piling up resources like talent, capital, and GPUs does not automatically generate top-tier products. 1 + 1 doesn’t necessarily equal 2; it can be greater than 2, but it can also easily be less than 2.

The thinking tool for this discussion is “organizational capital”: if human capital is attached to individuals, then organizational capital is rooted in human collaboration.

Definition and Value of Organizational Capital #

The concept of “organizational capital” was introduced by American economists Edward C. Prescott and Michael Visscher in 1980 [3]. Their insight was that an efficiently operating company naturally accumulates an exclusive asset that cannot be purchased externally: valuable information about who is good at what, who works best with whom, and who is most reliable for which tasks. This information constitutes the company’s organizational capital—it is not brought by any individual, but gradually cultivated by the team through long-term collaboration, belonging exclusively to the company, not to any individual.

The secret to team competitiveness lies here.

California law prohibits non-compete agreements. Even if you are a core technical backbone, you can freely move to a competitor. You might take theoretical knowledge, research acumen, personal reputation, and even networks with you, but you cannot take the original company’s collaboration practices, evaluation systems, failure archives, user feedback, internal tools, nor the tacit understanding developed by the team through long-term cooperation.

It is the existence of organizational capital that often leads to excellent individuals performing below expectations once they leave their original teams.

For example, a 2006 paper specifically studied cardiac surgeons [4]. Logically, performing surgery is a highly individualized skill. However, this study found that doctors who were very successful in their original hospitals, with high patient survival rates, experienced a drop in their surgical survival rates once they moved to another hospital. We can only conclude that doctors take their own hands with them, but cannot take the cooperation of the operating room.

Another paper from 2008 found [5] that star securities analysts on Wall Street, once they change jobs, often see their performance immediately decline and may not recover for several years; however, if they bring their original team with them, or join a company with stronger organizational capabilities, the decline is much less severe.

The red flower needs green leaves, a star needs a team; organizational capability is not merely the sum of individual abilities.

Organizational capital is the exclusive ability a company accumulates through long-term collaboration to transform dispersed knowledge among many members into shared judgment, shared products, and repeatable actions.

Knowledge Flow: The Key to Building Organizational Capital #

So, where does this capability come from? Is it the boss’s brilliant command? The star’s exemplary role? The processes written in a manual?

In 1998, a paper by British management scholar Janine Nahapiet and Indian-born management scholar Sumantra Ghoshal proposed that the quality of organizational capital hinges on how knowledge flows within your company [6].

They broke down an organization’s knowledge flow into three dimensions: cognitive, structural, and relational—

“Cognitive capital,” refers to whether dispersed knowledge in individual minds can be integrated through a common language, shared standards, and a collective problem awareness, thereby fostering mutual understanding;

“Structural capital,” refers to whether the right people are effectively connected to ensure that relevant personnel can be quickly found when specific knowledge is needed;

“Relational capital,” refers to whether the relationships among members are sufficiently safe and trusting to encourage everyone to express truthful information.

If company information is like a football, then the leader is like the team’s head coach: you’re managing not just a starting lineup, but the multiplicative effect generated between talents.

Cognitive Capital: Coordinating Team Diversity #

The key to cognitive capital is coordinating diversity.

We’ve discussed the “Good Regulator Theorem” in cybernetics, which states that a good regulator of a system must be a model of that system. In short, to effectively control something, you must possess at least a complexity matching that of the controlled system. One of the proponents of the Good Regulator Theorem, W. Ross Ashby, also had a similar theory called the “Law of Requisite Variety” [7]: to cope with a complex and changing external world, you must have adequately matched diversity internally.

Therefore, a company needs diverse talent, preferably from different fields, who can offer unique perspectives and each hold independent views, in order to brainstorm and leverage collective intelligence. You can’t say there are 100 scenarios in the market, but your management team only knows two tricks: “add more people” and “work overtime.”

However, diversity is not simply about gathering a group of smart people from different backgrounds; the key lies in effective coordination.

Researchers care about model capability, engineers about system stability, product managers about user experience, sales about actual orders, and security teams about avoiding incidents that make headlines… Everyone is correct, but if they don’t understand what each other is saying, diversity becomes noise, and the organization turns into a marketplace.

Those who haven’t managed large projects may find it difficult to deeply understand the severity of coordination problems. For example, you have two departments under you. You say, “We need to prioritize product quality,” and everyone agrees—but one department understands quality to mean reliability, while the other thinks you mean user experience. You say, “Launch as soon as possible,” one understands it as “release this week,” and the other as “release immediately when it’s mature”…

This is essentially why NASA’s Mars Climate Orbiter mission failed in 1999—Lockheed’s team used Imperial units for calculating the impulse of the thruster firings, while NASA assumed Metric units… The result was the probe disintegrating and burning up in the Martian atmosphere [8].

Effective cognitive capital does not require all members to agree, but rather for the team to maintain uniformity in language, standards, and problem awareness, ensuring their “coordinate systems” are perfectly aligned. The leader’s most important power lies in setting the agenda, clarifying standards, boundaries, and priorities. You must require the team to reach consensus on the following three core questions—

1. What problem are we truly solving?
2. What constitutes a good product?
3. When quality, speed, cost, and risk conflict, which takes priority?

This is like directing a group of AIs to work together, where the leader’s main responsibility is to articulate instructions clearly. An effective way to manage cognitive capital is to concretize abstract concepts—if you say “user first,” you must clarify during a rushed release: for the user, is it better to delay the launch, or launch first and then fix it; if you say “pursue excellence,” you must explain: which defects are absolutely intolerable, and which can be left for the next version…

As for the specific implementation methods, experts can have their own insights, and we can delve into them—but “what to do,” “to what extent,” and “what we are willing to sacrifice to achieve the goal” must reach a consensus.

Structural Capital: Building an Efficient “Expertise Map” #

The key to structural capital is that anyone with a question knows who to approach.

If a technical service person discovers a major product issue on the client front, can they directly find the product owner? If an engineer detects anomalies in training data, do they first have to write a report, inform their supervisor, wait for a weekly meeting, and then have the supervisor inform another supervisor? To increase productivity, knowledge must be able to flow quickly.

An excellent organization cannot demand that everyone be proficient in everything, but every member should clearly know who possesses what knowledge, whose judgment is most reliable in which area, and whom to ask for help when encountering certain types of anomalies. For this, you need an “expertise map,” academically called a “transactive memory system” [9], used to store the “addresses” of knowledge. It might exist in Feishu, DingTalk, knowledge bases, and project documents, but more often it resides in people’s minds and in interaction habits.

This “map” is gradually formed through team collaboration.

A 2005 paper specifically studied joint replacement surgery teams in hospitals [10]. The researchers broke “experience” into three parts for separate calculation: how many surgeries the lead surgeon personally performed, how many the entire hospital performed, and how many this specific team performed “together.” The results showed that even if the surgeon was experienced and the hospital was large, operations were still clumsy and difficult if the team had not collaborated enough.

Only through “working together” can this expertise map be cultivated. Initially, team members are exploring who is good at what, whose judgment is most reliable at which stage, and who to seek help from immediately if deviations occur. In the long run, it becomes a tacit understanding that needs no words: I know you’re good at paving the way but tend to overlook the finishing touches, so I’ll keep an eye on them for you; you know that when I say “no big deal,” I’m actually weighing the risks; and before I even extend my hand, you’ve already placed the next instrument in my palm.

Relational Capital: Fostering Team Psychological Safety #

The key to relational capital is team psychological safety.

We previously discussed this topic specifically when talking about a “sense of security.” Psychological safety is the primary common gene of excellent teams because it encourages team members to speak up boldly.

An engineer discovers a major defect in a project, but speaking up would offend the person in charge; a young researcher has a strange hypothesis, but fears being seen as ignorant; a middle manager determines the strategic direction is wrong, but publicly opposing the boss means they are “not firm enough.” These are typical manifestations of a lack of psychological safety.

The concept of “team psychological safety” was coined by Amy C. Edmondson [11], whom we mentioned earlier. Back in the 1990s, Edmondson was a PhD student at Harvard, assigned to study errors in hospitals, such as nurses giving wrong medication or incorrect dosages. She originally hypothesized that a well-managed group with a good atmosphere and a strong nursing supervisor should make fewer mistakes… but that was not the finding.

She compared two independent sets of data—one was error records from various nursing teams, and the other was ratings given by people to the quality and leadership of these teams—and made a counter-intuitive discovery: groups with better relationships and stronger leadership did not have fewer error records, but more.

Was it because these teams were lax in management, leading to member negligence? Not at all.

Edmondson didn’t stop at the surface statistics; she specifically hired someone to observe in the wards and conducted one-on-one interviews to uncover the truth: nurses in good teams were not clumsier or made more mistakes; rather, they were more willing to report errors—which is why their error records were higher! In groups with a stifling atmosphere, mistakes led to reprimands and accountability, so nurses’ instinct was to quickly cover up errors, often leaving no record at all.

Therefore, if you don’t hear bad news within your team, you should be wary. Your team is concealing the truth from you. In reality, bad news is part of the system’s daily routine; how can a healthy team not have bad news?

Silicon Valley Evidence: Organizational Differences Between Emerging Giants and Traditional Tech Companies #

Over the past year or two, I’ve had private conversations with several frontline R&D personnel from OpenAI and Anthropic. Some of their insights happen to corroborate the theories discussed in this article.

On several occasions, I asked them, “What is your company’s moat?” They all responded, without exception, that it was the research atmosphere.

The most telling point is that even a young newcomer, if they discover valuable insights in their research, their findings can be quickly integrated into the company’s large model products. At major Silicon Valley tech companies, such as Google, this is almost impossible—you have to report it through multiple layers, and it’s very likely it won’t even reach the front lines.

Both of these companies—and many emerging tech companies in Silicon Valley—conduct almost all of their daily operations through office software called Slack, which functions roughly like Feishu in China. Leaders issue tasks, employees claim tasks, deliver tasks, and all internal communications are done via Slack. You can have your AI read your Slack, post on your Slack, and submit work reports to Slack for you. From Slack, everyone knows what others are doing and who to contact if there’s a problem.

In contrast, at Google, if you want to assign a task to an engineer outside your team, you have to schedule a formal meeting with them and their direct supervisor, going through various procedures, which is quite cumbersome.

One R&D staff member even confessed to me that he would rather directly assign tasks to an AI than interact with a Google L5 (Senior Engineer level) or lower engineer—whereas at Anthropic, he has never encountered such communication difficulties.

Isn’t this cognitive capital, structural capital, and relational capital at work?

Effective Strategies for Enhancing Organizational Capital #

Perhaps, large enterprises are indeed not suitable for high-density R&D work. For projects like large models that require frequent iteration, having a lean and efficient team is crucial.

Effective methods for enhancing organizational capital do not rely on ideological education or team-building activities, but rather, as we discussed earlier, through joint operations, especially conducting high-quality debriefs. Research has found that properly designed debriefs can improve team performance by about a quarter on average [12]. But debriefing is not about a leader critiquing their subordinates; it’s about everyone asking four questions together: What was supposed to happen? What actually happened? Why was there a difference? What should be changed next time? – Debriefing is not about finding a scapegoat for failure, but about transforming an experience into the organization’s collective memory.

Thus, an excellent organization requires a stable core team that has undergone arduous collaboration and formed cooperative capabilities, avoiding frequent personnel changes.

However, a static organizational structure is not always optimal. Particularly for R&D teams, research has found [13] that if a team works together for too long, its members become increasingly isolated from key internal and external information sources, and technical performance consequently declines.

Therefore, you need both a stable “skeleton” and a fluid “frontier” to maintain vitality and openness.

Conclusion #

In summary, leadership is not about personally being the most powerful brain in the company; leadership is about creating an environment where many brains can combine to form a larger intelligence.

Organizations possess their unique wisdom. 1 + 1 > 2 is the foundation of human civilization.

Notes

[1] Hugh Langley and Pranav Dixit, “Read Mark Zuckerberg’s Memo Explaining What Alexandr Wang Will Be Running at Meta,” Business Insider, June 30, 2025.

[2] Hayden Field, “What’s Really Happening with the Hires at Meta Superintelligence Labs,” The Verge, August 29, 2025.

[3] Edward C. Prescott and Michael Visscher, “Organization Capital,” Journal of Political Economy 88, no. 3 (1980): 446–461.

[4] Robert S. Huckman and Gary P. Pisano, “The Firm Specificity of Individual Performance: Evidence from Cardiac Surgery,” Management Science 52, no. 4 (2006): 473–488.

[5] Boris Groysberg, Linda-Eling Lee, and Ashish Nanda, “Can They Take It with Them? The Portability of Star Knowledge Workers’ Performance,” Management Science 54, no. 7 (2008): 1213–1230.

[6] Janine Nahapiet and Sumantra Ghoshal, “Social Capital, Intellectual Capital, and the Organizational Advantage,” Academy of Management Review 23, no. 2 (1998): 242–266. Note that this paper used the term “social capital,” but in my view, it refers to “organizational capital,” simply named differently by various scholars.

[7] W. Ross Ashby, An Introduction to Cybernetics. London: Chapman & Hall, 1956.

[8] NASA, Mars Climate Orbiter Mishap Investigation Board, Phase I Report, November 10, 1999.

[9] Yuqing Ren and Linda Argote, “Transactive Memory Systems 1985–2010: An Integrative Framework,” Academy of Management Annals 5, no. 1 (2011): 189–230.

[10] Ray Reagans, Linda Argote, and Daria Brooks, “Individual Experience and Experience Working Together: Predicting Learning Rates from Knowing Who Knows What and Knowing How to Work Together”, Management Science 51, no. 6 (2005): 869–881.

[11] Amy C. Edmondson, “Psychological Safety and Learning Behavior in Work Teams,” Administrative Science Quarterly 44, no. 2 (1999): 350–383; see also Edmondson, “Learning from Mistakes Is Easier Said Than Done,” Journal of Applied Behavioral Science 32, no. 1 (1996): 5–28.

[12] Scott I. Tannenbaum and Christopher P. Cerasoli, “Do Team and Individual Debriefs Enhance Performance? A Meta-Analysis,” Human Factors 55, no. 1 (2013): 231–245.

[13] Ralph Katz, “The Effects of Group Longevity on Project Communication and Performance,” Administrative Science Quarterly 27, no. 1 (1982): 81–104.