The AI-native
talent investor.
Capital is abundant. Agents are commodities. The scarce input is the person who can run a company that runs itself — and the credential that used to identify them is dissolving.
A business leader used to be proved by an org chart. Span of control was the credential: how many people, how large a budget, how many layers beneath. It was a crude measure, but it was legible, comparable across companies, and it took roughly two decades to accumulate. An entire profession — executive search — exists to read it.
That credential is being dismantled, and not slowly. This piece argues that the dismantling creates a new class of leader, that this class is now the scarce input to company creation, and that the correct institutional response is to invest in them rather than merely to hire them. That is the thesis Living Scale Up is built on, and the rest of this piece is the case for it — including the strongest argument against it.
The pyramid is being removed while people are still climbing it
SignalFire's State of Tech Talent 2026, published 22 June 2026 from employment records covering more than 650 million individuals, reports that each engineering manager at a large technology company now oversees about twelve engineers, up from ten, and about fifteen at startups. In the same dataset, entry-level hiring has fallen roughly 65 percent at those large companies and about 76 percent at early-stage startups compared with 2019.
ICONIQ Growth's State of AI (July 2026), surveying roughly 300 executives at software companies building AI products, found that 72 percent of companies earning most of their revenue from AI run with four or fewer management layers between the chief executive and the most junior individual contributor — against 56 percent of their peers. Carta, reading its own cap-table records rather than a survey, reported on 4 May 2026 that average headcount at Series D had fallen 29 percent from its 2023 peak to 131 employees, and that the median seed-stage team is now four people.
Revelio Labs, tracking AI-adopting firms against a control group since October 2022, adds the shape of the change: at adopting firms, senior roles have grown 31 percent while junior roles have grown 6 percent. The organisation is not shrinking uniformly. It is hollowing out at the bottom and thickening at the top.
Put those together and a specific, under-discussed thing is happening to a specific group of people. The ladder by which someone demonstrated they could lead — take a team, then a larger team, then a function, then a P&L — is having its lower rungs sawn off and its middle compressed, while the people currently standing on it are in their late thirties and forties. Their credential is not being devalued in the future. It is being devalued now, mid-career, by forces none of them chose.
The denominator that replaces it
If not headcount, then what? The market is converging on output per person, and the numbers are large enough to be disorienting.
Start with the denominator, because a ratio without one is decoration. SaaS Capital, surveying more than 1,000 private SaaS companies and publishing on 30 July 2026, put median revenue per employee at USD 141,125, up from USD 129,724 the year before. That is the baseline: a well-run private software company generates roughly USD 141,000 of revenue per person.
Against that baseline: Lovable told TechCrunch it crossed USD 400 million in annual recurring revenue in February 2026 with 146 full-time employees — about USD 2.74 million per person, roughly nineteen times the median (TechCrunch, 11 March 2026). Cursor's chief executive Michael Truell stated the company had passed USD 1 billion annualised with more than 300 employees — over USD 3.3 million per person, roughly twenty-three times the median (Fortune, 8 December 2025). Gamma's co-founder said the company reached USD 100 million ARR profitably with 50 people (TechCrunch, 10 November 2025).
What these three numbers are, and are not
All three company figures are self-reported and unaudited. All three were selected by the press precisely because they are extreme, which means they are the tail of a distribution, not its centre. Neither Lovable nor Cursor nor Gamma publishes financials.
We publish them anyway, labelled, because the comparison holds even under heavy discount: halve them and they are still an order of magnitude above a benchmark drawn from more than a thousand companies. The direction is not in dispute. The magnitude at any individual company is.
We considered and rejected six other frequently-cited figures in this family. They are listed at the end of this piece, with the reason each failed.
Nor is the lean company itself new. In February 2014, Sequoia Capital — the investor, with board access — published that WhatsApp supported 450 million active users with 32 engineers, one developer per fourteen million users, at USD 1 per year with no marketing spend. What has changed is not the existence of the outcome but the accessibility of it. WhatsApp's ratio required a rare team and an unusually simple product. The current ratios are being reached by ordinary teams inside two years, and the speed is independently observable: Stripe, reading actual payment flows rather than self-reports, found the top 100 AI companies on its platform reached USD 1 million annualised in a median of 11.5 months (30 June 2025).
This is the inversion, and it is uncomfortable for institutions because it makes a leader's value illegible. You cannot read judgement off an org chart. There is no equivalent of "ran 400 people, CHF 80 million P&L" for "knew which three things mattered and killed the other forty." The market has not yet built the instrument that measures the thing that now matters most — which is precisely the arbitrage.
Who this produces.
Two words are needed, and both are contested. We will define them rather than assume them, and say honestly where each one fights us.
The AI nomad
Search "AI nomad" today and you will find laptop-on-a-beach content: location-independent freelancers using AI tools while travelling. No research house, consultancy or major publication has defined the phrase. It is unclaimed, and it is unclaimed because nobody with authority wanted it. We mean something different.
An AI nomad is an operator whose productive capacity is no longer bound to an institution. Four properties, all of which must hold:
That fourth property is where the term fights us. "Nomad" in common usage connotes drift, low commitment, a person passing through. That connotation is historically wrong: pastoral nomads were not drifters. They moved because the resource moved, their knowledge of the terrain was deeper than any settler's, and their commitment to the herd was total. What moved was the location, never the obligation. The AI nomad moves because the leverage moves. Someone who cannot commit is not an AI nomad — they are simply unemployed with good tooling.
Equally, what an AI nomad is not. Not a freelancer, who sells hours into someone else's system. Not a consultant, who sells recommendations and carries no consequence. Not an entrepreneur-in-residence — a job title four decades old that carries an understood connotation of a parking space between real roles. And not a "solo unicorn," a phrase originating with Sam Altman in 2023 and now attached to a branded community: that model takes no outside capital and no partners by definition, which makes it a different bet entirely.
The AI business leader
If "AI nomad" describes how these operators move, "AI business leader" describes what they do. Here the crowding problem is the opposite: the phrase is used constantly and means almost nothing. Consultancies and certification bodies have staked "AI-native leader" through 2026, and Microsoft coined "agent boss" in its Work Trend Index of 23 April 2025 — a study of 31,000 knowledge workers across 31 countries — for employees who build, delegate to and manage agents inside existing companies.
Every one of those framings describes adaptation: an incumbent leader learning new tools. We mean origination: someone who designs a company around artificial agents from the first day, where there is no legacy organisation to convert. The job differs in three specific ways.
The honest limit of this claim
No dataset counts these people. We looked. The closest available evidence is adjacent and does not substitute: Carta reports solo founders rose to about 36 percent of companies founded on its platform in 2025, from 31 percent in 2024 (12 March 2026), and Challenger, Gray & Christmas recorded a record 446 public-company chief-executive exits in 2025 (4 February 2026) — but Challenger does not track where departing chief executives go, so no causal claim is available from it.
We are describing a pattern we can argue for, not a population we can measure. Anyone who tells you the size of this cohort is estimating and should say so.
What this class
will demand.
These operators have strong alternatives: raise on their own track record, take a funded seat, or build solo with agents and keep everything. No studio beats "solo with 100 percent" on ownership, and we will not pretend otherwise. The competition is on speed, infrastructure, calibration and company — and the terms of trade are set by the candidate, not the studio. Five conditions follow, and they are conditions on us.
The talent investor,
updated.
The idea that talent — not the idea, not the deck — is the investable unit is not ours. Entrepreneur First named it in February 2019, when Alice Bentinck described a category of "talent investors" who fund individuals before they have a team or a company, paying a stipend that is not repayable even if no company results. It was a genuinely new category, and EF has spent a decade proving it works.
What has changed since 2019 is the machinery. In 2019, funding an individual pre-team meant funding a search: for a co-founder, for an idea, for a market. The company still had to be built from nothing, and most of the risk sat in that construction. In 2026, the construction is the part that has collapsed in cost and time — validated demand, an agentic operating core, compliance substrate, instrumentation — while the person who can run the resulting machine has become the binding constraint.
So the category splits. One branch keeps investing in individuals before the company exists. The other invests in operators who already have a track record, and hands them the company already built and already proven. That second branch is what we mean by the AI-native talent investor, and it is where Living Scale Up sits.
Three things follow from taking the word "investor" seriously rather than decoratively. First, the leader must hold a stake large enough to behave like an owner — and must hold it in a company that actually owns something, which is the part usually skipped. Second, that stake must vest against what the venture achieves rather than how long the leader stayed, because an investor is buying outcomes. Third, the studio must publish the structure of the investment before asking anyone to consider it — no serious investment category operates on terms discovered late.
We are one month old, so we will say exactly where we are on the third. The complete terms go to every candidate in writing, early — before they have spent real time on us. They publish here in September 2026, when the Venture 02 mandate opens, as a whole capitalisation table rather than a chief-executive figure quoted alone, because a figure quoted alone is misread in both directions. Publishing late and complete beats publishing early and unusable.
The argument against us,
at full strength.
It would be convenient if the analysis above led cleanly to "and therefore join a venture studio." It does not, and the strongest counter-argument is not the one usually raised.
The usual objection is about equity percentages, and the published ranges are wide: Steve Blank puts studio stakes at 30 to 80 percent (Harvard Business Review, 13 December 2022); Ben Yoskovitz — himself a studio founding partner — puts them at 15 to 80 percent and writes plainly that he struggles with the math above 40 to 50 percent (17 October 2023); Vault Fund data cited in Venture Studio Perspective gives 21 to 43 percent (19 May 2025). Those numbers matter, but a number can be answered with a number.
The deeper objection cannot. In 2026 the Journal of Business Venturing published "Founders for hire? The role of venture studios in breaking the individual-opportunity nexus" (Coelsch-Foisner, Vandeweghe, Clarysse and Murray), built from 16 venture studios and 50 interviews. Its central finding is the sharpest thing written about our model:
The mechanism the authors identify is an inversion of agency. A conventional founder begins as a principal and gradually cedes control to investors as the company grows. A studio-recruited leader enters as an agent from inception, because the studio performed the entrepreneurial work — opportunity identification, validation, initial structuring — before any leader was involved. The paper also observes, drily, that studios use the word "co-founder" as a motivational device rather than a description.
We think this is correct as a description of the model's default. We also think it is a structural problem rather than a messaging problem, which means it cannot be answered with adjectives — not with "founder-first," not with "real ownership," not with any phrase a studio can write about itself.
It is worth being precise about what the answer has to address, because the obvious response is the wrong one. The instinct is to reach for a bigger equity number, and equity is the thing studios argue about publicly. But the distinction between a principal and an agent is not a quantum — it is a locus of control. A chief executive holding thirty percent of a company whose product, brand and customer data belong to someone else, who cannot set price without approval, is an agent with a large stake. A chief executive holding less, in a company that owns its own business and whose operating decisions are genuinely theirs, is closer to a principal. Ownership of the asset and authority over the decision do more work here than the percentage does.
Our structural answer — in order of how much it actually matters
- The venture owns its own business. A venture is wholly studio-owned while it is being incubated — before it exists as a separate company there is nothing else it could be. At incorporation, when you are seated, the product and codebase, brand, domains, customer relationships, customer data and any model trained on venture data are assigned to the venture. The studio retains and licenses only the reusable operating core, on terms that survive our exit from the cap table. You are not running a company that rents its own existence.
- Operating authority from day one — pricing, roadmap, technical direction, hiring, go-to-market, capital allocation within budget. The studio holds only shareholder-reserved matters, and its board weighting reduces as external capital enters. By Series A the direction is set by you and your investors, not by us.
- Consequence that runs both ways. If we stop the venture you keep everything vested with no clawback, receive a cash bridge, take first look at the next mandate — and we state publicly that the studio stopped it and why, so you do not spend two years explaining a decision our gates made.
- Vesting against outcomes, not time served — a time-vested core with a twelve-month cliff, plus tranches released against gates published before you start and not moved afterwards, accelerating on Series A.
- A meaningful minority stake at incorporation, fully diluted before external financing. The band is decided, and every candidate receives it in writing early rather than late. It publishes in September 2026 as a complete capitalisation table rather than a chief-executive percentage in isolation, because a figure quoted alone is misread in both directions and tells you nothing you can act on.
Whether that is sufficient is not for us to declare. The authors describe a mechanism by which psychological ownership fails to develop; we have tried to reinstate each condition it removes — asset ownership, authority, consequence, and a stake that tracks outcomes — and to publish the structure so the claim is checkable rather than asserted. If a candidate reads it and still concludes that founding alone is better for them, they are probably right about themselves, and we would rather they reached that conclusion in week one than in month nine.
The Swiss paradox
we operate inside.
We are based in Lavaux, in the canton of Vaud, between EPFL and the Lake Geneva ecosystem. The case for building here is strong and independently documented — and the case against recruiting globally into here is equally strong. Both belong in the same paragraph.
On supply: the 2026 European Deep Tech Report (Lakestar, Walden Catalyst and Dealroom, March 2026) ranks ETH Zurich first globally for alumni-founded, VC-backed deep-tech startups since 2020 with 192, and EPFL second in Europe with 94 — together 289 spinouts, eight unicorns, and USD 24.2 billion of enterprise value, exceeding Oxford and Cambridge combined. Startup Genome's 2026 profile puts the Greater Lausanne region at USD 18.7 billion ecosystem value and tenth among emerging ecosystems globally.
Against that, three facts that cut against our own recruiting story. Swiss third-country work-permit quotas have been frozen for a third consecutive year at 4,500 B permits and 4,000 L permits (BAL, 4 December 2024; Clark Hill, 25 November 2025) — 8,500 permits for the entire Swiss economy, across every sector and all 26 cantons, subject to a labour-market test. The Swiss Startup Association's Swiss Startup Agenda (February 2026) ranks Switzerland last among 25 European countries for startup equity friendliness, citing the possibility of taxation before any gain is realised, and notes that 78 percent of Switzerland's unicorns were founded by international teams while no streamlined founder visa exists. And in Startup Genome's own ranking of Greater Lausanne, talent strength is the region's weakest of four dimensions.
The conclusion is unflattering and worth stating plainly: a thesis that assumes frictionless global mobility into Switzerland is wrong on the facts for non-EU/EFTA operators. The realistic recruiting ground is EU/EFTA, and anyone who tells a candidate otherwise is setting up a disappointment in month four.
There is a second conclusion, and it happens to favour us, so we hold it more loosely: if Swiss equity is genuinely hard to structure and hard to hold, then the structuring itself is a real service rather than a claimed one. An individual operator arriving alone confronts a regime the country's own industry body calls the worst in Europe. That is an argument for a studio — but only for one that has actually solved it, and only once it is willing to show how. We show ours in writing to every candidate, and publish it in September.
What would change
our mind.
A thesis that cannot be wrong is not a thesis. Three observations would materially weaken this one, and we commit to revisiting the piece against them.
This piece will be reviewed on 6 February 2027. Changes will be logged on our facts page, alongside the corrections record.
Every figure, with its
provenance attached.
Three verification classes are used below. Observed means the source has direct visibility into the underlying activity. Survey means a stated methodology and disclosed denominator. Self-reported means the company said so and nobody independently checked.
| Claim | Source | Date | Class |
|---|---|---|---|
| Median revenue per employee, private SaaS: USD 141,125 (n > 1,000) | SaaS Capital | 30 Jul 2026 | Survey |
| Lovable: USD 400M ARR, 146 employees | TechCrunch | 11 Mar 2026 | Self-reported |
| Cursor: > USD 1B annualised, 300+ employees | Fortune | 8 Dec 2025 | Self-reported |
| Gamma: USD 100M ARR, 50 employees, profitable | TechCrunch | 10 Nov 2025 | Self-reported |
| Top 100 AI companies on Stripe: median 11.5 months to USD 1M annualised | Stripe | 30 Jun 2025 | Observed |
| WhatsApp: 450M active users, 32 engineers | Sequoia Capital | 19 Feb 2014 | Observed |
| Engineering-manager span ~12 (majors) / ~15 (startups); entry-level hiring −65% / −76% vs 2019 | SignalFire | 22 Jun 2026 | Observed |
| 72% of AI-revenue companies run ≤4 management layers, vs 56% of peers (n≈300) | ICONIQ Growth | Jul 2026 | Survey |
| Average Series D headcount −29% from 2023 peak to 131; median seed team of 4 | Carta | 4 May 2026 | Observed |
| Solo founders 36% of 2025 Carta incorporations, vs 31% in 2024 | Carta | 12 Mar 2026 | Observed |
| AI-adopting firms: senior roles +31%, junior roles +6% since Oct 2022 | Revelio Labs | 28 Jul 2026 | Observed |
| Record 446 public-company CEO exits in 2025 (destination not tracked) | Challenger, Gray & Christmas | 4 Feb 2026 | Observed |
| "Agent boss" definition; 31,000 workers across 31 countries | Microsoft Work Trend Index | 23 Apr 2025 | Survey |
| "Talent investor" category coined; funding individuals pre-team, pre-idea | Alice Bentinck, Entrepreneur First | 20 Feb 2019 | Primary |
| Studios remove the conditions under which founder ownership develops (16 studios, 50 interviews) | Journal of Business Venturing | 2026 | Peer-reviewed |
| Studio equity 30–80% | Steve Blank, HBR | 13 Dec 2022 | Commentary |
| Studio equity 15–80%, "many taking 40–50%+" | Ben Yoskovitz, Focused Chaos | 17 Oct 2023 | Commentary |
| Studio equity 21–43% (Vault Fund data) | Venture Studio Perspective | 19 May 2025 | Commentary |
| ETH Zurich 192 / EPFL 94 alumni-founded deep-tech startups; 289 spinouts, USD 24.2B EV | 2026 European Deep Tech Report | Mar 2026 | Report |
| Greater Lausanne: USD 18.7B ecosystem value; talent strength weakest dimension | Startup Genome GSER | 2026 | Report |
| Swiss third-country quotas frozen: 4,500 B + 4,000 L | Clark Hill · BAL | 25 Nov 2025 · 4 Dec 2024 | Official |
| Switzerland last of 25 European countries on startup equity friendliness; 78% of unicorns internationally founded | Swiss Startup Agenda | Feb 2026 | Industry body |
Claims we researched and did not publish
- Midjourney revenue per employee. Headcount is irreconcilable across sources — approximately 11 in one Forbes calculation, 40 in the widely circulated version, 10 elsewhere. Midjourney has never disclosed headcount. No denominator, no usable figure.
- The AI revenue-per-employee league table attributed to Dealroom. Published only as a social post in April 2025 with no methodology, no headcount sourcing and no as-of date per company, then republished by a data brand in a way that gives it the appearance of independent verification. It has none.
- "Telegram runs a billion users with about 30 engineers." Traces only to aggregator posts. No filing, interview or founder statement substantiates any engineer count. Telegram's revenue is sourceable; its headcount is not.
- Bessemer's "Supernova" ARR-per-employee archetypes. Attributed only to unspecified portfolio data, with no n, no statement of whether the figures are medians or means, and no publication date on the deck.
- "Fastest company ever to USD 100M ARR." Claimed by more than one company, originating from each company's own communications, and unfalsifiable — no source maintains a complete time-to-milestone dataset across all software companies. We used the underlying dated facts instead.
- "Executives are leaving corporates for AI ventures." We wanted this and could not source it. No public dataset tracks the destination of departing senior operators. We made it an argument rather than a statistic, and labelled it as such.
If this describes you,
the structure is already public.
Venture 01, BuddyLeader, has its chief executive — Jonas Cosendai. The founding-CEO mandate for Venture 02 opens in September 2026.