The Future Company Is Smaller

David Holz raised $94 million from Andreessen Horowitz, Founders Fund, and Highland Capital to build Leap Motion. The technology was remarkable — hand tracking precise enough to follow all ten fingers to 1/100th of a millimeter. The product worked. The investors were among the most celebrated in Silicon Valley. The outcome was a sale to a British haptics company for $30 million. Ten cents on the dollar.

Much of the capital went where venture capital often goes when it exceeds what a company needs: office space in San Francisco's SoMa neighborhood, perks designed to signal momentum rather than produce it, and an engineering team that expanded while the company searched — unsuccessfully — for product-market fit across three consecutive pivots. Apple attempted to acquire Leap Motion twice. Both efforts collapsed.

Holz spent twelve years there. When he left, he started Midjourney with ten people, no outside capital, and what appears to have been a deeply personal conviction: the capital itself had been the problem.

Within two years, Midjourney was generating $200 million in annual revenue. By 2025, it had reached $500 million — profitable from its first month of operation, with no marketing expenditure and no external investors. The same venture firms that had backed Leap Motion, and many others, have since sought to invest. Holz has declined every approach.

What interests me about Midjourney is not the artificial intelligence. It is the organizational structure that Holz chose — and what it reveals about a broader shift that is already underway.

Midjourney compensates its employees through profit sharing rather than stock options. There are no equity grants. No option pool. No four-year vesting cliffs, no liquidation preferences, no 409A valuations. The company generates revenue and distributes a share of it to the people who do the work.

This is not the economics of a startup. This is partnership-shaped economics — and a growing number of the most consequential AI-native companies are converging on the same model, whether they use that language or not.

To contextualize these figures: the leading traditional SaaS companies — Salesforce, Adobe, ServiceNow, Workday — average roughly $610,000 in revenue per employee, and employ tens of thousands of people. The emerging class of AI-native companies is producing five to six times that figure with teams of two dozen. This is not a marginal efficiency gain. It is a different category of company.

Cursor illustrates the fork in the road. Four MIT graduates. Twelve employees. $100 million in annual recurring revenue — the fastest any SaaS company has reached that milestone. Then $3.4 billion in venture capital. Then three hundred employees. The small, high-output firm became a conventional corporation. Whether this was the correct decision is beside the point. What matters is that the decision was binary: accept venture capital and scale the organization, or remain small and independent. No instrument existed to support a middle path.

Gamma reveals the other side of this constraint. $100 million in annual recurring revenue. Profitable for over two years on $23 million in total funding. When Gamma raised its Series B, the round included a secondary component — not because the business needed primary capital, but because early employees needed a mechanism to convert their paper wealth into cash. Without a fundraising event, no such mechanism existed. The company was producing partnership-shaped economics but had no infrastructure to distribute them.

A Structural Successor to Something Very Old

Before the era of venture-backed technology companies, the highest-performing professional organizations operated on what I would describe as partnership-shaped economics. Goldman Sachs maintained this structure for 130 years before its 1999 initial public offering. McKinsey & Company still does. So do the most enduring law firms, surgical practices, and architecture studios.

The defining principles were consistent across all of these institutions. Teams were small relative to the value they produced. Each individual was significant to the firm's output. Compensation was drawn directly from the firm's economic performance — not deferred into abstract equity instruments or contingent on a future liquidity event. There was no "exit" in the venture capital sense. The firm endured, and its people were compensated continuously as they contributed.

Goldman's pre-IPO structure offers the most instructive historical case. SEC filings from the 1999 corporate conversion reveal that partner compensation had been classified as "distributions of partners' capital" — distinct from salary, equity grants, or discretionary bonuses. Partners received their share of the firm's profits directly. This mechanism operated successfully for over a century.

Whitehead's warning proved accurate. Following the public offering, Goldman's culture of shared economic fate and long-term orientation gave way to the incentive structures of a public corporation: quarterly earnings pressure, stock-based compensation, and the externalization of risk. Partners who had been with the firm for twenty-five years departed. The institutional knowledge they carried left with them.

I am not drawing an equivalence between Goldman Sachs and a thirty-person AI company. The industries, risk profiles, and regulatory environments are entirely different. What I am observing is that the principles which sustained Goldman's model for 130 years — small teams of high-leverage individuals, compensated continuously from the firm's cash generation, retained through economic alignment rather than contractual illiquidity — are now reasserting themselves in a completely different context, driven by completely different forces.

The AI-native company did not arrive at this structure by studying financial history. It arrived there because the underlying conditions demand it. When a firm's entire workforce consists of twenty-five people, each of whom is genuinely difficult to replace, the economic relationship between the firm and the individual becomes personal, consequential, and mutual. The economics become partnership-shaped whether anyone designs them that way or not.

The previous generation of technology companies adopted the language of partnership — "we're all owners," "your equity aligns you with the company" — and applied it to organizations of thousands. But an employee holding 0.002% in stock options at a company with three thousand colleagues is not an economic partner in any meaningful sense. That was the language of partnership applied to the structure of industrial employment. The AI-native company dissolves this contradiction by eliminating the need for scale. It does not need three thousand people. It needs thirty.

The $61 Billion Symptom

When an economic reality outgrows its institutional infrastructure, the strain becomes visible in secondary mechanisms — workarounds that emerge to solve problems the primary system was not designed to address. In venture-backed technology, that strain is now quantifiable.

Secondary transactions have now surpassed initial public offerings as a liquidity mechanism for venture-backed companies. This is not a cyclical fluctuation. It is a structural inversion — and it reflects the fact that the primary system no longer serves the needs of the companies and people operating within it.

The individual cases are instructive. OpenAI conducted a $6.6 billion tender offer — a single transaction representing 6.2% of all annual US secondary volume. Stripe facilitated a tender at a $91.5 billion valuation. Databricks structured its $10 billion capital raise to include a secondary component specifically so that employees could access cash. Even Gamma — profitable, small, capital-efficient — included a secondary in its Series B for the same reason.

The participation data confirms the underlying pressure. Median tender offer participation rose from 36.6% in 2021 to 56% in the first half of 2025. Subscription rates — the proportion of buyer demand that sellers were willing to meet — reached 99.9%. The people who built these companies are willing to sell nearly every share that buyers will take. This is not speculation or profit-taking. It is the rational behavior of individuals who hold significant paper wealth with no native mechanism to realize it.

Every tender offer is, in effect, a patch applied to a system that lacks a distribution mechanism. The $61 billion secondary market is what results when partnership-shaped economics are forced through corporate-era financial infrastructure. It is an enormous, expensive workaround — and its growth is evidence that the underlying architecture requires redesign, not repair.

Consider what Holz built at Midjourney. He decided that the people who generate the firm's revenue should be compensated from that revenue, on a continuous basis, without waiting for a liquidity event that may never occur. This is not a novel principle. It is the principle on which professional firms operated for centuries before the venture capital model introduced a different set of assumptions. Holz did not innovate on compensation theory. He simply declined to adopt the deviation.

The Open Question

The structural conditions are now clear. A new class of company is emerging — small, high-output, profitable early, generating economics that historically only appeared inside professional partnerships. The financial infrastructure surrounding these companies — how they are capitalized, how their people are compensated, how liquidity is created — was designed for a fundamentally different type of organization: one that requires hundreds of employees, burns capital for years, and achieves returns through a terminal exit event.

The mismatch is producing visible consequences. A $61 billion secondary market. Tender offers as a standard operating procedure rather than an exceptional event. Founders who reject outside capital entirely — not because they are philosophically opposed to investment, but because no instrument exists that aligns with what they have built.

The question, then, is not whether the economics of the AI-native company are real. The data has settled that. The question is what it would look like to build financial infrastructure that matches these economics — capital structures that assume a company may never go public and are designed accordingly, compensation mechanisms that distribute value continuously rather than contingently, and investor relationships that resemble the alignment between a firm and its limited partners rather than the adversarial dynamics of a board seat.

Holz built one answer: reject all outside capital. This works in rare circumstances. It does not scale as a solution for the broader category of founders who need capital but do not need — and should not accept — the structural concessions that currently accompany it.

I believe the most valuable AI-native companies of the coming decade will look less like startups and more like the firms that came before them. The economics will be partnership-shaped. The teams will be small. The individuals will be irreplaceable. The only open question is whether the financial infrastructure evolves to serve them — or whether they continue to be forced into instruments designed for a company type that no longer represents the highest and best use of capital.

What if the financial infrastructure is the last thing standing between these founders and the model they actually want?

More soon.