The Other Door

Every investor in technology has heard the cliché that a company's most valuable asset walks out the door every evening. It is among the oldest observations in the business — true, repeated to the point of furniture, and generally filed under culture and retention, the soft matters a good operator manages. What the cliché has never been followed through to is its consequence on the cap table. If the most valuable asset walks out every evening, then the thing an investor's equity is a claim on is not that asset. It is the building the asset walks out of. The familiar fact is about people. The consequence is about property — about what equity actually attaches to, and what it does not.

Start with what an investor believes he is buying. He believes he is buying the company: its product, its customers, its technology, and the team that built them, bound inside a single entity whose shares he now partly holds. The premise is that these travel together, because they are the same company, and his shares are a claim on all of it. The exit is supposed to be the moment that claim pays — someone buys the company, the entity changes hands, and the proceeds flow up through the shares.

A transaction now common across artificial intelligence breaks that premise cleanly in two, and it does so by design. An acquirer does not buy the company. It hires the company's people and licenses the company's technology — and leaves the company behind. The founders and key engineers depart, employed and paid, often handsomely. The technology goes too, under a license. What stays is the legal entity: its customer contracts, its infrastructure, its remaining staff, its shares. The thing the investor owns continues to exist. What made it worth owning has left, under contract, for a building the shares have no claim on.

This is not hypothetical, and it is not rare. In March 2024 Microsoft hired the chief executive and substantially the entire team of Inflection AI — a company that had raised on the order of $1.3 billion at a $4 billion valuation — and paid roughly $650 million, structured as a non-exclusive license to Inflection's models plus a sum to settle any claim over the hiring. Microsoft took no equity and acquired no assets; Inflection remained, by the deal's own architecture, a "separate, independent company," now without its founders, pivoting to a new business under a new chief executive. Months later Amazon did the materially similar thing with Adept: it hired the founding chief executive and most of the team, licensed the technology and models, and left roughly a fifth of the staff behind in a standalone entity under a new leader. In 2025 the pattern reached its most elaborate form with Windsurf, an AI coding company. An acquisition by OpenAI, reportedly near $3 billion, collapsed when its exclusivity window lapsed; within hours Google paid about $2.4 billion to license Windsurf's technology and to hire its chief executive, a co-founder, and a small group of senior researchers into its own labs — no acquisition, no equity — while the roughly 250 people left behind, along with the product, the brand, and some $82 million of recurring revenue, were sold separately to a different company entirely. The value left by one door; the entity by another.

The structure is engineered, and the engineering is the point. The acquirer wants the team and the technology and does not want the rest — the customer commitments, the infrastructure, the obligations a going concern accumulates. So it takes the first and leaves the second. The people are hired, which requires no one's permission but their own. The technology is licensed, which compensates the entity in form while removing the asset in substance. The entity is left standing, technically solvent and technically still owned by its investors. The cap table is undisturbed; every percentage is exactly what it was. The percentages now describe a shell.

What this exposes is a fact about venture equity that the ordinary case keeps hidden: the claim attaches to the entity, and in a great many of the businesses venture finances, the thing of value is not owned by the entity. The entity owns its contracts, its code, its cash, its furniture. It does not own its people — their skill, their judgment, their working relationships, the tacit capability that sits on no asset register because it cannot. People can be employed, incentivized, retained; they cannot be owned, and so neither the entity nor, through it, the investor holds any title to them. The equity is a claim on everything the company owns, and the thing the company most needed it never owned. It rented it, the way every employer rents its people, terminable at will. Where the value lives in the team, the investor's claim is a claim on the one asset that can legally choose to leave.

This is why the transaction is a revelation rather than merely a raid. In a company whose value sits in a factory, a patent estate, a book of signed contracts — assets the entity genuinely owns — value cannot simply walk out, because it is bolted to the thing the shares have a claim on. The acquirer who wants it must buy the entity, and the proceeds reach the shares. But in a company whose value sits in forty people, nothing is bolted down. The acquirer who wants it hires them and licenses what they have made, and the entity — the thing the shares attach to — becomes beside the point. The equity's claim was on the wrong object all along. It was on the container, when the value was in the contents, and the contents had legs.

That the buyers doing this are the most sophisticated in the market is the strongest evidence that it is a structural feature and not a run of misfortune. When the most capable acquirers repeatedly choose to take the people and license the work rather than buy the company, they are not improvising. They are acting on a judgment about where the value sits and where it does not — that for this kind of company, the entity is the part not worth buying. The investor's shares are a claim on precisely the part the most discerning capital in the market has decided to leave behind. He is, in effect, holding the position the smartest buyer declined to take.

It is fair to object that investors are not defenseless — that vesting, non-competes, and well-drafted assignment of intellectual property exist to bolt the people and their work to the entity, and that an investor stripped this way simply failed to paper the deal. There is something to it, and the correction draws on it. But it overstates what paper can hold. Vesting can be renegotiated or bought out by a buyer who wants the team enough; non-competes are unevenly enforceable and, in some states, barely at all; assignment secures yesterday's work, not the work that exists only in a person's head and walks out with them. These instruments raise the price of leaving. They do not convert a person into an owned asset, because nothing does. They buy friction, and friction is worth buying — but friction is not title, and the claim here is about title.

There is a stronger objection: that the investor is not left with nothing, since the license fee and the residual entity have value, and the acquirer pays something that reaches the shares. Sometimes a great deal — in the largest of these arrangements the proceeds to investors were substantial, and the Windsurf remainder was sold rather than abandoned. But the payment is set by what the acquirer must give to take the team and the technology, not by what the company was worth as a going concern, and those are seldom the same figure. The license compensates for the asset's use, not its loss; the residual entity is worth what an emptied entity is worth. The investor is paid the price of the container and the rent on the contents, having believed he owned both. That he is paid something is exactly what lets the outcome read as an exit rather than register as what it structurally is — the sale of access to an asset the investor never held, conducted over the head of the equity that thought it did.

So venture equity is not, reliably, a claim on the value of a company. It is a claim on an entity — and where the value lives in people, the entity and the value are separable things, parted by any acquirer willing to use the other door. In the ordinary case they are never parted, because the value is bolted down and the only way to take it is to buy the whole, and so the investor never discovers that his claim was on the container. He discovers it in the case where the value gets up and leaves under contract while the shares stay behind. Venture capital prices a claim on a company and learns, at the door, that what it held was the company minus the part that could decide to go — and that the part that could decide to go was the part it was paying for.

The correction is partly mechanical and partly a matter of attention. Mechanically: bolt down what can be bolted — vesting that survives a change of control, assignment that is total and current, the structural rights that make a team-and-license deal expensive enough to deter, or expensive enough to share. But the deeper correction is to underwrite the human-capital company for what it is: a business whose principal asset is not owned but retained, and whose value can therefore leave by a door the cap table does not watch. That means pricing the departure risk at entry rather than meeting it at exit, and treating any company whose worth is its people as a company whose worth is, in the last instance, free to walk. The investors who are paid in these situations are not the ones who held the largest share of the entity. They are the ones who understood that the entity was not where the value lived, and built their claim — through structure, through alignment, through the price they paid going in — around an asset that could always, lawfully, choose to leave.

An investor can own every share of a company and still not own the thing he was buying.

Sources & Methods. The argument is structural; the transactions are verified public reporting, used to illustrate a mechanism. No claim of wrongdoing by any company is made or intended — each transaction described was, by all available accounts, lawful and disclosed. The empirical record the analysis is set against:

■ Microsoft–Inflection (March 2024) — Microsoft hired CEO Mustafa Suleyman, co-founder Karén Simonyan, and substantially all of Inflection's staff; paid ~$620M for a non-exclusive license to Inflection's models plus $30M related to the hiring ($650M total, per Bloomberg/The Information); took no equity stake and acquired no assets; Inflection continued as an independent company under a new CEO, pivoting from its consumer "Pi" product to enterprise model licensing. Reporting noted the non-acquisition structure may have been designed to avoid triggering a merger review. Inflection had raised ~$1.3B at a ~$4B valuation.

■ Amazon–Adept (June 2024) — Amazon hired co-founder/CEO David Luan and most of Adept's ~100-person team and licensed Adept's technology, models, and some datasets; Adept continued as a standalone company under a new CEO with ~20 employees remaining. A coda underscoring the thesis: by early 2026, four of the five Adept co-founders who joined Amazon had themselves left — the retained asset departing the acquirer in turn. (A structurally similar Amazon–Covariant arrangement followed in August 2024.)

■ Meta–Scale AI (June 2025) — a variant worth distinguishing: Meta took a 49% non-voting stake for ~$14.3B (valuing Scale ~$29B); founder/CEO Alexandr Wang and a small number of staff moved to Meta's "superintelligence" effort; Scale continued operating under an interim CEO, and investors and employees shared in the proceeds. Unlike Inflection/Adept, the equity holders were paid substantially — included here as the minority-stake form of the same talent-extraction structure, not as an emptied-shell case.

■ Windsurf (July 2025) — the most elaborate instance, and the one carrying the most moving parts. OpenAI's reported ~$3B acquisition lapsed (entangled with Microsoft's IP rights under its OpenAI agreement); Google then paid ~$2.4B to license Windsurf's technology and to hire CEO Varun Mohan, co-founder Douglas Chen, and a small group of senior researchers into DeepMind — no acquisition, no equity. The remaining ~250 employees, the product, the brand, the IP, and ~$82M ARR were subsequently acquired by Cognition. Reporting also noted that some employees who had joined Windsurf within the prior year were excluded from the Google deal's financial participation — the container/contents split made literal.

■ The pattern and the regulatory backdrop — CB Insights, AI Agent Bible (2025), characterizes reverse acqui-hires (hire-the-team, license-the-tech, leave the entity) as a leading exit structure in coding AI. The FTC has scrutinized several of these arrangements over concern that the structure absorbs startups without triggering formal merger review; that regulatory pressure is itself part of why the structure is chosen, and is presented as context, not as a finding against any party.

■ Human capital and the limits of retention instruments — general venture-financing practice on vesting, IP assignment, and non-compete enforceability (with notable jurisdictional variation; non-competes are sharply limited in California and, more recently, the subject of federal action), as the basis for the claim that retention instruments raise the cost of departure without conferring title. The observation that human capital is unowned is itself long-standing (the "assets walk out the door" maxim); the letter's contribution is the consequence for what equity is a claim on, not the maxim.