Venture Capital Was Built Around Headcount Economics

For most of modern venture capital, there has been an implicit relationship between money and people. When a startup raised a round, the capital was expected to become organizational capacity, and much of that capacity came in the form of hiring. Seed funding brought in the first engineers and operators. A Series A built out product, sales and marketing. Later rounds financed management layers, larger commercial organizations, international expansion and the infrastructure required to coordinate all of it.

This relationship became so familiar that hiring itself became a signal of progress. A rapidly growing startup was usually a rapidly hiring startup. Investors tracked headcount plans alongside revenue forecasts. Founders justified financing needs partly by showing the teams they intended to build. A company that raised $30 million and hired aggressively looked like it was deploying capital; one that raised $30 million and kept most of it in the bank might invite questions about whether it knew how to grow.

There were good economic reasons for this. For most companies, increasing productive capacity required increasing the number of people capable of producing, selling, servicing and coordinating the work. Software changed parts of that equation long before generative AI arrived, but even software companies generally scaled by building increasingly large organizations around their products.

AI is beginning to complicate that relationship. If companies can increase productive capacity without increasing headcount proportionally, then one of the assumptions buried inside the venture model starts to weaken. And if that continues, the implications are larger than smaller rounds or better burn multiples. Venture capital may eventually have to reconsider what financing rounds are for, how progress should be measured, what organizational scale looks like, and ultimately how funds themselves should be constructed.

The firm has always been partly a coordination technology

There is a much older economic question underneath this.

In 1937, economist Ronald Coase asked why firms exist at all. If markets are effective at allocating resources, why don't individuals simply contract with one another for every piece of work that needs to be done? His answer, simplified considerably, was that using markets carries costs. Finding suppliers, negotiating contracts, coordinating activities and enforcing agreements all require time and effort. Under certain conditions, bringing those activities inside a firm is cheaper than continually coordinating them through the market.

Two decades later, Edith Penrose approached the firm from another direction. In The Theory of the Growth of the Firm, she described companies not simply as administrative structures but as collections of productive resources. Growth depended in part on management's ability to organize and deploy those resources. The firm could not expand infinitely simply because capital was available; managerial capacity itself imposed limits on how quickly productive resources could be absorbed and coordinated.

Neither Coase nor Penrose was writing about venture capital, obviously, and neither theory should be reduced to a prediction about startup headcount. But both help illuminate an assumption that became deeply embedded in the way companies have historically scaled: productive capacity depends substantially on assembling and coordinating human capability.

Venture capital developed inside that economic reality. 

A startup raised capital because there was something it wanted to accomplish faster than its existing resources allowed. In many cases, the mechanism for doing so was straightforward: hire more people. Want to build the product faster? Hire engineers. Want to accelerate distribution? Build a sales team. Want to enter Europe? Hire people in Europe. As those teams grew, hire managers to coordinate them and build finance, HR, legal and operations functions to support the larger organization.

Capital purchased labor. Labor created capacity. More capacity, deployed well, produced growth. Of course, that was never the entire story. Capital also paid for servers, factories, inventory, acquisitions, advertising and countless other inputs depending on the business. But particularly in venture-backed software, labor has historically represented one of the dominant uses of capital and one of the primary ways a company increased what it was capable of doing. That is the relationship AI is beginning to disturb.

What happens when capacity stops tracking headcount?

There is already evidence that AI can increase individual output in certain kinds of knowledge work, although the magnitude varies significantly by task, worker and implementation.

In one widely cited study by Erik Brynjolfsson, Danielle Li and Lindsey Raymond, customer-support agents given access to a generative AI assistant increased the number of issues they resolved per hour by roughly 14 percent on average, with much larger gains among less experienced and lower-skilled workers. A controlled experiment involving GitHub Copilot found developers completing a specific programming task approximately 56 percent faster with AI assistance. A later randomized experiment involving Google engineers estimated a roughly 21 percent reduction in time spent on a complex coding task, although the researchers cautioned that the confidence interval was large and that results from one task should not automatically be generalized to software development as a whole.

Microsoft researchers have found something subtler. In a six-month randomized field experiment involving 6,000 knowledge workers, people with access to generative AI spent less time on email and appeared to complete documents somewhat faster, but the technology did not significantly reduce meeting time. The researchers noted an important distinction: AI changed activities workers could alter independently more easily than activities that depended on coordination with other people.

That last finding may ultimately be one of the more important ones for venture.

AI can make an individual faster without automatically making an organization faster. Companies still have coordination costs. Decisions still need to be made. Customers still need to be acquired. Teams still need context, incentives and direction. The existence of AI does not repeal Coase.

But it may change the boundary.

If an engineer can produce more without adding another engineer, a customer-support team can resolve more cases without adding agents, or a small operating team can perform work that previously required several specialized functions, then the relationship between organizational capacity and organizational size becomes less linear.

We do not yet know how far that relationship will move. We should be skeptical of confident predictions that ten-person companies will routinely produce billion-dollar outcomes simply because the tools now exist. Productivity improvements observed in controlled tasks do not translate automatically into equivalent company-level productivity improvements.

But venture capital does not require the relationship to disappear for the economics to change. It only requires the slope to move.

Venture rounds were designed to buy capacity

This is where the implications become more interesting. A financing round has traditionally answered two questions at once: how much money does the company need, and what will that money allow the company to become? The second answer frequently involved organizational scale. The company would use the capital to hire the people necessary to reach the next milestone, at which point it would raise another round to purchase the next tranche of capacity.

That logic helped produce many of the conventions of venture financing. Companies raised enough money for roughly a certain amount of runway because burn was expected to increase as teams expanded. Milestones were partly calibrated around what an organization of a particular size could reasonably accomplish. Later stages involved larger rounds because scaling distribution, management and international operations required increasingly large organizations.

If productive capacity can increase faster than headcount, those relationships become less predictable. Imagine a company that raises $10 million and discovers that its existing team can accomplish what its original operating plan assumed would require twice as many people. The conventional description is that the company has become more capital efficient. That is true, but incomplete. The more important question is: what should happen to the $10 million?

Should the company simply extend its runway? Should it raise less next time? Should it spend dramatically more on distribution? Should it pursue another product simultaneously? Should it acquire a competitor? Should it enter a market that would previously have been too expensive to justify? Should it return capital? Should it have raised $10 million in the first place?

Those are not questions about AI tooling; they are questions about capital allocation. And they become increasingly important if the connection between capital and hiring weakens.

We may need different signals of scale

This creates a related problem for investors. Many of the signals venture has historically used to understand a company's development were formed in the same headcount-intensive environment.

Headcount itself is an obvious example. A company growing from 20 employees to 100 suggested that something meaningful was happening. It indicated access to capital, confidence in future demand and increasing organizational capability. It also created an enormous new cost base, which made subsequent growth necessary.

Hiring plans provided investors with a rough map of how capital would translate into execution. Burn revealed how quickly that organizational machine consumed resources. Revenue per employee offered one way of measuring the efficiency of the machine. Even the language of “scaling” often conflated two things that usually happened together: the business was becoming larger and the organization was becoming larger.

Those two things may no longer move together as reliably. That does not make headcount irrelevant. It makes headcount a less complete proxy for capacity. A 30-person company and a 100-person company may increasingly possess capabilities that are difficult to infer from their organizational size alone. Conversely, a company can adopt dozens of AI tools without meaningfully increasing its productive capacity if the underlying workflows, decisions and coordination remain poor.

Investors therefore need to become better at measuring what the organization can actually do.

How quickly can it turn an idea into a shipped product? How many customers can the existing organization support before service deteriorates? How quickly can it run and learn from experiments? Where does another dollar of capital actually increase throughput? Which constraints disappear with software and which remain stubbornly human? At what point does adding another employee create more value than another unit of compute, distribution or data?

These questions are harder than counting employees; they are also closer to what investors were trying to understand all along.

The meaning of a financing round may change

We have written previously about [what capital-efficient growth looks like when companies can accomplish more before adding permanent cost]. The natural extension of that argument is that venture rounds themselves may need to become less standardized around organizational expansion.

If a Series A is no longer primarily financing the construction of a much larger organization, what is it financing?

There will not be one answer. For some companies, particularly those operating in the physical world, capital will remain heavily tied to people and infrastructure. For others, distribution may become the dominant constraint. Proprietary data may matter more. Regulatory approvals, acquisitions, hardware, compute or access to scarce networks may absorb capital that previously would have gone toward payroll.

The point is not that startups will stop hiring. It is that capital allocation may become more idiosyncratic as labor becomes less universally synonymous with capacity.

That has implications for round size and timing. A company that can reach a meaningful milestone with $4 million rather than $12 million may rationally raise less. Another company might still raise $12 million but deploy the difference toward a strategy that would previously have been impossible at that stage. Some companies may go much longer between rounds because their burn does not rise as quickly. Others may choose to raise aggressively precisely because their organizational leverage allows them to deploy capital into distribution or acquisitions faster than competitors.

The familiar progression from Seed to Series A to Series B does not disappear overnight. But the economic logic underneath each stage becomes less uniform.

And eventually, the fund has to change too

This is where the argument reaches beyond the company.

In Portfolio Construction Is Changing. But So Is the Company Inside the Portfolio, we argued that discussions about changing venture portfolios often begin too far downstream. Before deciding how many companies a fund should own, how much it should reserve or what ownership percentage it should target, investors need to understand whether the economic object inside the portfolio is itself changing.

Headcount economics provides another way of seeing the same problem.

The traditional venture fund is partly constructed around assumptions about how much capital successful companies will consume. Those assumptions influence initial check sizes, follow-on reserves, ownership targets, fund size and return mathematics. If a successful software company historically needed $100 million of equity capital to reach a certain scale, owning 15 percent of it required one investment strategy. If an equally valuable company can reach that scale with $30 million, the strategy begins to look different.

The arithmetic alone does not tell us what the new venture model should be. Lower capital requirements can create their own complications. Smaller rounds can make it harder for large funds to deploy enough capital. Competition can push valuations upward and offset some of the ownership benefits of capital efficiency. Founders who need less money may be less willing to sell large percentages of their companies. Meanwhile, businesses in AI infrastructure, hardware, robotics, energy and other capital-intensive sectors may move in precisely the opposite direction and require extraordinary amounts of capital.

There may therefore be no single “AI venture model.” The more likely outcome is greater divergence.

Some categories of companies may become extraordinarily capital-light. Others may become more capital-intensive. Some founders may use AI primarily to maintain small organizations for longer. Others may use the same leverage to pursue much larger ambitions with the same amount of capital.

A venture industry built around relatively predictable relationships among capital, hiring and growth will have to learn how to distinguish among them.

Capital is still buying something

It would be easy to take this argument too far and conclude that AI makes venture capital less important. We don't think that follows. Capital still buys time. It buys risk. It buys the ability to make investments whose returns will not be immediate. It buys access to markets, infrastructure, talent, compute, distribution and acquisitions. And in many businesses, it will continue to buy a great many people.

What may be changing is the assumption that more capital should naturally produce a larger organization because a larger organization is the primary mechanism through which a company becomes capable of more. For decades, that relationship helped shape the architecture of venture capital almost invisibly. It influenced how much startups raised, what they spent it on, how investors evaluated progress, when companies returned for another round and how funds reserved capital for the winners.

AI does not need to eliminate labor from the equation to make those assumptions worth revisiting. It simply needs to allow productive capacity to grow faster than headcount in enough companies for the old proxies to become unreliable. We are beginning to see evidence that this is possible. We do not yet know how large the effect will become. But if venture capital was built in an era when capital became people and people became capacity, then a world in which capital, people and capacity can move more independently from one another is not simply a world of more efficient startups. It may require a different venture model.

Sources

• Ronald H. Coase, “The Nature of the Firm” (1937), Economica. The foundational paper on transaction costs and why economic activity is sometimes organized within firms rather than through markets. Read via Wiley Online Library
• Edith Penrose, The Theory of the Growth of the Firm (1959). Penrose's foundational work on firms as collections of productive resources and the role of managerial capacity in constraining growth. Oxford University Press edition
• Erik Brynjolfsson, Danielle Li & Lindsey R. Raymond, “Generative AI at Work” (NBER Working Paper 31161; subsequently published in Quarterly Journal of Economics). The study of 5,179 customer-support agents found AI assistance increased productivity by roughly 14% on average, with substantially larger gains for less experienced workers. NBER paper
• Shakked Noy & Whitney Zhang, “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence,” Science (2023). A randomized experiment examining how generative AI affected productivity on professional writing tasks. Science article
• Sida Peng et al., “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot” (2023). In the controlled experiment referenced in the draft, developers with Copilot completed the specified programming task 55.8% faster than the control group. Research paper
• Microsoft Research, “Shifting Work Patterns with Generative AI”. The six-month randomized fi
• eld experiment involving more than 6,000 workers referenced in the draft found changes in email and document-related work but no statistically significant reduction in meeting time. Microsoft Research
• McKinsey Global Institute, “The Economic Potential of Generative AI: The Next Productivity Frontier” (2023). Estimates that generative AI use cases could eventually contribute $2.6–$4.4 trillion annually across the use cases analyzed. This is modeled economic potential, not an observed productivity result, so I would preserve that distinction if you add the figure to the article. McKinsey Global Institute report
• Misfit Labs, “What Capital Efficient Growth Actually Looks Like Right Now.” The earlier Misfit argument about AI changing the relationship among headcount, capital requirements, and growth. Misfit Labs article
• Misfit Labs, “Portfolio Construction Is Changing. But So Is the Company Inside the Portfolio.” The prior argument referenced in the venture-fund section about changing company economics eventually affecting portfolio construction, reserves, ownership, and fund strategy. Misfit Labs article

‍