Satya Nadella’s Insights on the Future of Organic Intelligence (OI)

Microsoft CEO Satya Nadella has published a deep and subtle post on the future of intelligence—both artificial and organic together: “A frontier without an ecosystem is not stable.” Nadella essentially argues that companies exist to learn. He envisions a “cognitive loop” of “human and token capital”.

The fusion of economics and cognitive science language is both fascinating and useful. His reframe puts meat on the bones of the vision of humanity’s future in which AI augments human potential rather than replaces it. At the same time, it’s in tension with the “humans in the lead” framing I wrote about recently and needs augmentation to achieve the human-positive vision that Nadella is reaching for. The piece is worth a careful examination.

The frame: humans working together

Nadella’s very first sentence gives us our first clue that his post is not going to be a typical CEO-writes-PR piece:

I’ve been thinking a lot about the future of the firm in an AI-driven economy.

The word that jumps out as a little unexpected is “firm.” Not “company.” Not “business.” “Firm.” While one could write it off as an idiosyncratic word choice, the piece is too carefully writtenfor that word to be accidental. The canonical reference to that word is Nobel Prize-winning economist Ronald Coase’s 1937 paper, “The Nature of the Firm“. Nadella, as a widely read person, may have encountered it in its original essay or through the open-source treatise “Coase’s Penguin, or Linux and the Nature of the Firm“. The question both essays address is simple: “When does it make sense for people to give up some autonomy and work within a firm (as opposed to working as free agents)?” Coase’s answer is that people join together in firms when the cost of coordinating as individual agents is too high. Benkler, writing in 2002, accepts Coase’s frame but argues that internet technologies lower coordination costs past a threshold for collaboration on open-source projects like Linux. Nadella seems to argue that AI, when deployed as he envisions, has the potential to change coordination costs again, tipping the balance back toward the firm.

He frames the problem this way:

This is the first time we can create a real cognitive loop between people and digital systems. That is a mind-bender, because it changes how we even conceptualize work inside an enterprise.

What is at stake is not some digital tool or system and its use, but how organizations continue to learn, build IP, differentiate, and thrive in a world where AI models can continuously absorb the expertise of humans and organizations and commoditize it.

Nadella is doing several jobs at once with these few sentences. First, he’s posing the ability of AI to “absorb” and “commoditize” expertise as a challenge rather than an unquestioned goal with assumed positive outcomes. While AI executives mouth the “augment rather than replace” sentiment from time to time, Microsoft’s CEO describes a specific trajectory for AI in which he believes augmentation rather than replacement will be the outcome. At the same time, I don’t think the term “cognitive loop” is accidental either. Up until about 10 years ago, AI and cognitive science researchers were joined at the hip. We should read Nadella as arguing that AI does cognitive work and, in a real sense, can concentrate and reorganize what had previously been diffuse distributed cognition.

Every company is going to have to build what I think of as human capital and token capital. Human capital comprises the knowledge, judgment, relationships, ingenuity, and pattern recognition of its people, while token capital is the firm’s AI capability it builds and owns…. This loop becomes the new IP of the firm. I think of it as a hill climbing machine. And unlike most assets, it compounds. Every improved workflow generates better training signal, which accelerates the accumulation of tacit knowledge unique to the firm. The companies that build this early will have an advantage that is hard to replicate, regardless of any new individual model capability.

In Nadella’s vision—which, it should be said, requires some innovations in AI design that do not yet exist—humans generate new ideas; AI identifies, absorbs, and operationalizes them at scale; and the firm accumulates them as differentiators. Companies won’t use generic ChatGPT or Claude; they will use their own models that dynamically accumulate and distribute the learning happening within the organization. In fact, we might consider Nadella to be arguing that a “learning orgnization” empowered by its own AI is something closer to a “learning organism.” His reference to “tacit knowledge” is another callout to a research tradition: knowledge management. Former Hewlett-Packard CEO Lew Platt famously stated: “If only HP knew what HP knows, we would be three times more productive.” Platt likely meant that knowledge is not evenly distributed within an organization, causing inefficiency. The knowledge management research argues the barrier is harder than that: You and I don’t even know what we know. Tacit knowledge is knowledge that we aren’t consciously aware of and can’t articulate. Nadella seems to argue that AI can identify know-how that we don’t know how to express or aren’t even aware we have. I think this is loosely plausible. My own experience using AI is that it is very good at both identifying my personal ways of thinking and organizational patterns of thinking that are buried in a previously unusable mass of documents on Google Drive, Slack, email, and so on. AI can be very effective at interpreting the lossy exhaust of thinking that resides in our systems, along with observations about the thinking contained in prompts themselves. If all that can be durably captured and redistilled into an evolving AI model—which is what Nadella is arguing for, and which is not a technological given—then yes, AI can increasingly reduce coordination cost across the company by learning from us as we work.

If you’re worried about the concentration of AI power in a few companies, then Nadella’s vision may be attractive to you. (It’s certainly attractive to Microsoft, since they can still make lots of money if AI becomes a commodity.) If you’re worried about AI replacing humans, this vision may also have some surface appeal. Nadella asserts that, in the firm he describes, human capital “compounds” (with an exclamation point!). Your value, as a human, compounds. The AI takes over much of “that which has been learned,” freeing you to keep focused on learning.

When that happens, companies will create value for themselves and for the economy around them. Employees will see their expertise amplified and their judgment become part of systems that make it replicable and scalable and the benefits accrue to the companies and communities around them.

That is how companies drive value for themselves and the broader economy. And it is the stable equilibrium we should build together.

So Nadella asserts. And he doesn’t cleanly separate “should” from “will”. Is this inevitable, or something we should build together? And if we do build this, will it be good for humans?

I have mixed feelings.

The problem

While I understand why the CEO of Microsoft would focus on the firm as a unit of analysis, I think he misses a critical point in the literature he’s alluding to. When Coase wrote about the nature of the firm in the 1930s, large companies with centralized management were fairly new, historically speaking. He was trying to explain why they came to exist. His answer was that management and employment lower the cost relative to paying for independent workers on an open market. Benkler’s analysis points us toward the social contract dimension that Coase took for granted. If coordination costs were lower, would workers automatically continue to do their work through a firm? Not necessarily.

Like Coase, Benkler was trying to understand a new development. Why do people devote their labor to open-source software without getting paid either as an employee or a freelancer? It turns out that workers are humans, and humans do work for a variety of reasons. Now let’s consider Benkler’s insight in light of the “gig economy,” which is a kind of a middle case. For a company like Uber, the software platform lowers coordination costs, creating new economics of coordination. That’s what the Uber apps do. I log in as a driver when I want to. You call for a car. Uber routes that request to me. I am not an Uber employee. The platform takes a firm’s coordination role without a binding employment contract between the firm and the worker. In fact, some of the legal debates over this kind of arrangement are on whether the workers deserve some of the benefits and protection that employees get in return for effectively working for the company enough that the firm owes them something more in return for their labor. The legal contract is governed by a social contract. Employers of gig workers are accused of using a different contractual structure to avoid commitments to the workers that are required by law for employees.

Lowering coordination costs can benefit the workers and the firm alike. But it doesn’t have to work that way. Nadella’s vision as articulated in his essay accrues all benefits to the firm. When Nadella says that human capital will “compound”, he means the value of the workers will go up over time. That would be a pro-worker outcome, if it held. But nothing in the architecture he describes ensures it does. The firm’s loop is engineered to compound; workers are only valuable as long as they are generating new ideas that the AI absorbs and very efficiently makes the company’s property. Rather than being “good at their job”, workers in Nadella’s firm are good at generating ideas for the job to be done better by AI, or by cheaper labor that the AI guides. Ideas would essentially become instant company property that could be efficiently passed from the people capable of thinking of them to people merely capable of executing them under guidance. That’s not compounding; it’s extractive. This is a serious flaw in Nadella’s claim of the compounding value of human capital, but it doesn’t need to be fatal (and I believe his pro-worker sentiments are genuine expressions of his desired outcome). It is possible to rescue his theory so that value accrues to the workers as well as the firm.

The solution: Recognition

Nadella’s system can become a true human capital compounding machine by adding one simple question to its token compounding machine: Who? Who thought of this? Who invented this technique? Who created a new skill? Who learned and applied it?

Nadella wants a continuously trained AI to recognize useful business patterns, extract them, and incorporate them into the model weights. Fine. Employers capturing and reusing employee innovation is not new. Nadella proposes a method that’s theoretically more efficient at doing so. Employees should get credit, for a variety of reasons. First, the employers benefit by knowing what their employees are good at, reducing the tendency to treat their ideas as resources to extract and increasing the recognition that the ideas come from particular smart, capable humans. That gives workers paths inside the companies. They can establish their contributions efficiently. If their way of doing work is important enough to the firm that it gets trained into the model, then it is important enough to recognize who contributed it. Second, if I have that recognition as a worker in a form that I can take with me, it makes me more legible as a valuable employee to future employers. To accomplish that goal, the recognition should live outside the corporate IP boundary in a Learning and Employment Record (LER), a portable, standards-based skills record. (My employer, 1EdTech, stewards the core standards in an LER, which are built on broader standards such as the W3C’s Verifiable Credentials.) This enables the human contributors to efficiently establish a verified portfolio of their skills, contributions, and growth. I wrote about this idea last September:

AI may help to capture emerging skills that have not yet been codified. For example, recently I’ve been vibe coding as a non-programmer. I’ve figured out how to vibe code Model Context Protocol (MCP) servers in TypeScript and python, use progressive disclosure patterns to reduce AI token usage while increasing accuracy and security, and build a compositor that enables me to orchestrate these workflows using microservices. Some of these skills didn’t exist six months ago…. An AI that understands digital credentials standards could identify, express, and capture evidence for emerging competencies as part of the exhaust stream of my work. And another AI could read that evidence. To be clear, nobody would have any reason to believe that I have any of these skills based on my formal work experience. 

Think about that moment in time. Vibe coding was just emerging. Effective practices were just emerging. An employee working in that situation invents effective practices. That happens all the time now across any field of endeavor where AI is changing how work is getting done. The ways I accomplish some tasks that are core to my job are vastly different than they were six months ago. I have both learned and invented new skills. I expect that to continue. If my growth were visible and portable, that would be valuable for me. And that visibility need not subtract from the corporate IP. My skills at vibe coding, which I teach to my fellow employees but which also travel with me when I leave, is separable from specific work I contributed to my employer’s IP. My skill is owned by me.

And if an AI can say, “Here’s something novel and valuable about the way you’re doing your work that may not be clear to you,” expressing it that way makes it valuable to me as well as to the firm—especially if the verification that I have the knowledge or skill is wrapped up in a secure and portable record that expresses not company IP but my demonstrated human potential as a contributor. If Nadella’s compounding machine gives me that as part of the social contract I make with my employer, I may be willing to make that deal.

The third benefit of designing a system that way is it allows the compounder to ask a different question: “Who would benefit from learning this?” We shouldn’t uncritically accept the assumption that all human work will eventually be performed by AI agents. As I’ve been exploring in my recent posts, economics as well as technological limitations will have their say in how AI ultimately gets used. A system that can recognize and codify skills might eventually become good enough to teach and assess them. At the very least, such a system should be able to identify and articulate the skills that need to be taught and assessed. Compounding requires investment. If you want to increase the value of human capital, you need to invest in the humans. Any system good enough to capture human-generated value within a specific context—like a firm—should provide infrastructure for investing in the humans who can generate value.

Nadella’s final paragraphs are worth reading in this light:

In my view, our priority has to be building a frontier ecosystem, not just a frontier model, so value flows broadly across every company, every industry, and every country. One where every organization can own the learning loop that encodes its institutional knowledge, compounding its human and token capital….

[C]ompanies will create value for themselves and for the economy around them. Employees will see their expertise amplified and their judgment become part of systems that make it replicable and scalable and the benefits accrue to the companies and communities around them.

That is how companies drive value for themselves and the broader economy. And it is the stable equilibrium we should build together.

This is, I think, a worthy aspiration. Notice, though, that the stable equilibrium depends on employees seeing their expertise and judgment valued. Not extracted. Valued. Nadella’s story is somewhere between “humans in the loop” and “humans in the lead.” It’s “humans driving the loop.” I contribute. I accomplish. I learn. I teach. I participate. I grow. That dynamic stabilizes a frontier ecosystem.

Nadella wants to position AI as a tool for improving the capture and dissemination of human knowledge, which puts it in the same technological tradition as inventions like written language and the printing press, which both democratized access to knowledge. That’s an interesting and productive frame. It’s certainly healthier than “AGI”, which centers the question, “When can we replace the humans?” Nadella wants the answer to be “never.” He wants AI to remain a human productivity tool. That’s possible, but it’s not inevitable. A stable ecosystem is one that supports life. If Nadella wants to see his vision come to fruition, he needs to consider the conditions that support human welfare more carefully. His framework allows for that, but it’s currently incomplete.