Intelligence Over Headcount

For most of my career, I have thought about organizations through the lens of people. I am an organizational psychologist by training, and much of my work has involved helping companies decide whom to hire, how to structure teams, where decisions should sit, how managers should manage, and what kind of organization a company needs to become as it grows.

Underneath much of that work was an assumption so basic that we rarely needed to articulate it: if a company wanted to do substantially more, it generally needed more people.

The relationship was never perfectly linear, but it was reliable enough to shape how we built companies. A larger engineering roadmap meant hiring more engineers. More customers eventually required larger sales and support organizations. Geographic expansion meant adding people in new markets. Growth produced complexity, complexity produced coordination costs, and those costs eventually produced additional layers of management. Headcount therefore became more than an expense on the P&L. It became one of our most common proxies for organizational capacity.

AI is beginning to make that proxy considerably less useful. This is not because people cease to matter, or because every company is destined to become a ten-person organization surrounded by agents. I suspect the quality of the people inside an organization actually becomes more consequential as AI improves. What changes is the assumption that adding productive capacity necessarily requires adding another person.

From Headcount to Intelligence

The modern corporation was built around a fundamental constraint: human capability is scarce, specialized, and difficult to scale. If you needed more software written, you hired engineers. If you needed more analysis, you hired analysts. If you needed to serve more customers, you hired people to serve them. Those employees generated coordination requirements of their own, which meant managers, meetings, systems, and increasingly elaborate mechanisms for moving information through the organization.

Ronald Coase was thinking about a version of this problem almost 90 years ago. In The Nature of the Firm, he argued that firms exist in part because using markets has transaction costs, while organizing activity within a company has costs of its own. Crucially, he also argued that those internal organizing costs tend to rise as firms expand. The size of a firm is therefore partly determined by the economics of coordination.

AI changes an important variable in that equation because a person working with capable AI systems can increasingly command resources that previously would have been distributed across several people or functions. An engineer can use agents to write, test, and document code. A marketer can conduct research, develop variations, and analyze performance without assembling the same supporting cast. A generalist can interrogate a dataset without waiting for an analyst to produce the first cut. A manager can synthesize information from across a business without asking several people to prepare it first.

There is now good evidence that at least some of this increase in individual capacity is measurable rather than anecdotal. A set of randomized field experiments involving 4,867 software developers at Microsoft, Accenture, and a Fortune 100 company found that developers given access to an AI coding assistant completed 26.08% more tasks when the results of the three experiments were combined. The effect varied, and less experienced developers tended to benefit more, but the underlying implication is significant: the productive capacity represented by an employee is becoming less fixed.

We see a similar pattern outside engineering. Erik Brynjolfsson, Danielle Li, and Lindsey Raymond studied the introduction of a generative AI assistant across 5,179 customer-support agents and found a 14% average increase in issues resolved per hour, with substantially larger gains among novice and lower-skilled workers. Their evidence suggests that AI was, in part, allowing less experienced workers to access and apply patterns associated with more capable colleagues.

The relevant measure of organizational capacity therefore starts to shift. The question is no longer simply how many people a company employs, but how much intelligence and capability those people can effectively direct.

This Is Not an Argument for Smaller Companies

It would be easy to interpret this as another prediction that AI will produce ten-person billion-dollar companies. I think that misses the more important point. Productivity gains do not necessarily translate into lower employment because companies can reinvest additional capacity into doing more. If an engineering organization becomes significantly more productive, management could reduce the size of the team, but it could also keep every engineer and increase the number of products it builds, enter markets that previously did not justify the investment, or attempt problems that were uneconomic when engineering capacity was more expensive.

We do not yet know where that equilibrium will settle, and the emerging research gives us reason to be careful about assuming that every productivity gain immediately translates into organizational transformation. A 2025 field experiment across 66 firms and 7,137 knowledge workers found that employees who actively used generative AI spent approximately two fewer hours per week on email and reduced work outside normal hours. Yet researchers did not find broader changes in the overall quantity or composition of their work. Individual efficiency can change before the organization around it does.

That distinction matters. AI may increase what an individual can accomplish relatively quickly, while companies take much longer to redesign roles, workflows, management structures, and incentives around that new capacity. The first-order effect is productivity. The more consequential second-order effect, if companies choose to pursue it, is organizational redesign.

The Scarce Resource Moves Toward Judgment

When execution is expensive, leaders spend a significant amount of their time acquiring and coordinating execution capacity. They determine whom to hire, which team should own a problem, how responsibilities should be divided, and how information should move among the people doing the work. As execution becomes cheaper, more of the constraint moves upstream toward deciding what should be done in the first place.

This is where I think the conversation about AI and organizations becomes considerably more interesting than the debate over how many jobs it will eliminate. AI can produce options, analysis, code, content, and recommendations at a volume that would have been prohibitively expensive only a few years ago. But abundance at the production layer increases the importance of discrimination at the decision layer. Someone still has to understand which customer problem matters, which output is actually good, what should be delegated, where a human needs to intervene, and which opportunities deserve the organization's finite attention and capital.

In other words, as certain forms of intelligence become cheaper, judgment becomes relatively more valuable.

We are seeing versions of this in the companies and founders we work with at Misfit Labs. The strongest operators are not necessarily the people automating the greatest number of tasks. They are becoming more deliberate about where machine intelligence creates leverage and where human judgment remains disproportionately valuable. That distinction matters because it is entirely possible to automate a badly designed organization. A company can create agents around every existing workflow and generate extraordinary quantities of code, analysis, documentation, and communication without becoming meaningfully better at deciding what deserves to exist.

Organizations already make versions of this mistake with humans. We have spent decades consuming highly capable people's time moving information between systems, preparing status updates, reconciling documents, sitting in meetings whose primary function is transferring context, and producing analyses because producing them became part of a process. AI gives us an opportunity to remove a meaningful amount of that work, but only if we resist the temptation to automate every existing process exactly as we found it. Otherwise, we will have made the machinery more efficient without asking whether we still need all of the machinery.

What Happens When Headcount Stops Signaling Importance?

There is another implication that I find particularly interesting as an organizational psychologist. Corporate status has historically been tied to the accumulation of resources. Senior executives control larger budgets, promotions often come with larger teams, and we routinely use the number of people someone manages as a proxy for the significance of the role.

There were sensible economic reasons for this. Managing a thousand people meant responsibility for an enormous amount of organizational capacity because, for most of modern corporate history, those thousand people were the capacity.

That logic becomes less obvious if a leader with 30 exceptional people and well-designed AI systems can eventually command more productive capability than another leader with 300. In that environment, a manager who responds to every capacity constraint by requesting additional headcount should be evaluated differently from one who first asks whether the work needs to exist, whether the workflow can be redesigned, and where another person's expertise and judgment create enough incremental value to warrant the investment.

This is not an argument against hiring. It is an argument for becoming considerably more thoughtful about what we ask humans to spend their time doing. There is nothing particularly humanistic about preserving low-value work simply because humans historically performed it.

Intelligence, Not Efficiency

There is understandable fascination right now with the prospect of extremely small teams building very valuable businesses. The market itself is beginning to provide hints that value creation is migrating beyond the companies building the largest foundation models. Crunchbase found that 29 companies joined its Unicorn Board in May 2026, and described the standout trend not as new model builders but as businesses helping enterprises put AI to work, alongside AI infrastructure, autonomous software, and robotics companies. Ten of the 29 companies were less than three years old.

I think the important lesson is broader than the possibility of smaller teams or faster unicorns. Minimizing headcount should not become the objective any more than maximizing headcount should have been. A 40-person company should become an 80-person company if the next 40 people create more value than alternative uses of the capital. A 4,000-person organization should question whether it needs to become a 5,000-person organization if much of that incremental capability can be created another way.

The objective is to build the most intelligent organization, not the smallest one.

That requires changing where organizational design begins. Growth planning has traditionally moved fairly quickly from strategic objectives to an organizational chart: this is what we want to accomplish, therefore these are the functions we need, these are the roles inside them, and these are the people we need to hire. Increasingly, I think there is an important step before the organizational chart. Leaders should first map the intelligence required to operate the business. They should understand which decisions need to be made, which capabilities are required to execute them, what information those decisions depend on, where human judgment genuinely differentiates the company, and where machine intelligence can provide leverage.

Only after answering those questions does headcount become particularly meaningful.

People have historically been the default unit of organizational capacity because, until very recently, there was no serious alternative for most forms of cognitive work. As that changes, headcount can become an output of organizational design rather than one of its principal inputs. Some companies may end up remarkably small. Others may become enormous because AI expands the opportunity available to them faster than it reduces their need for labor. In either case, the size of the organization tells us less about its capability than it used to.

For those of us who have spent our careers thinking about organizations, that is a more consequential shift than whether AI eliminates one job category or creates another. We have spent more than a century developing increasingly sophisticated ways to organize people around work. The next generation of companies will have to become equally sophisticated about organizing human and machine intelligence around one another, and about reserving the scarcest resource in that system, excellent human judgment, for the places where it actually changes the outcome.

That, much more than having the smallest team, is what I think an AI-native organization should mean.

‍---

Sources

• Ronald H. Coase, “The Nature of the Firm,” Economica (1937). 
• Zheyuan Cui, Mert Demirer, Sonia Jaffe, Leon Musolff, Sida Peng, and Tobias Salz, “The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers,” Management Science (2025). 
• Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, “Generative AI at Work,” National Bureau of Economic Research (2023). 
• Eleanor W. Dillon, Sonia Jaffe, Nicole Immorlica, and Christopher T. Stanton, “Shifting Work Patterns with Generative AI,” National Bureau of Economic Research (2025).
• Gené Teare, “AI Services And Robotics Lead Diverse Crop Of 29 New May Unicorns,” Crunchbase News (June 9, 2026).