Skip to content

Companies are built, not born

The practice of building companies from within investment firms, systematically—rather than simply backing independent founders—remains one of the most underappreciated drivers of value creation in the venture capital industry.

One of my least favourite narratives in entrepreneurship is that startups are some sort of mythic creation. That founders must be struck with a bolt of lightning, have founder-market fit, or be the person uniquely well positioned to solve whatever problem they end up working on, and so on and so forth.

This has to be one of the most passive, low-agency ways of viewing the world. It is basically to say that you can only act on the world when the world conspires to create the perfect circumstances for you to act on it. It’s just plain false, and a disservice to any ambitious person who wants to leave their mark on the world.

A far more inspiring view is that you can use your own agency to discover and solve important problems and, in doing so, create immense value for yourself and the world.

What follows was written in response to a conversation with a friend who viewed company creation this way. I set out to lay out the empirical evidence for why this is not the case. But empirical evidence aside, what matters is that we inject belief into as many people as possible that entrepreneurship is open to them. It is not a mythic creation, and the more smart people spending their time on important problems in the world, the better.

Let’s begin.

Where I think we really disagree is the claim that systematic company creation has “accomplished fuck all,” and that personal insight is a prerequisite rather than one path among several.

We should definitely have a longer conversation on this, because I have an extremely different view, and I think my view is much closer to the US view. I had the same view as you before I spent a large amount of time in the US, but I’d go as far as saying that it is extremely divorced from reality here. I do think it’s the favourite narrative of a lot of VCs and founders outside the US, because it romanticises the act of company creation and sells stories.

What I mean by incubation

Maybe I need to define incubation more clearly, and that’s causing the problem. When I say incubated, I don’t mean “factory incubator,” like Antler. (Antler is also probably the worst performing example you could use of incubation.)

When I say incubation, I mean a systematic process to identify people or problems worth working with or on, and supporting the full founding and lifecycle of the company. You develop a thesis on a new market and find the team to build it. You give them their first capital. You find someone to be CEO. You find the person who can build “Halter for X,” you find their co-founder, and you earn the right to be their most important funder by doing so.

Incubation does not assume extractive “venture studio” economics. It assumes that you use your differentiated insights into markets, capital, and people to create a competitive advantage in the act of company creation. You fund on terms that allow you to work with the best people for the specific opportunity.

The scale of it

Before I get into the rest of my reasoning, I want to put the empirical scale of this on the table, because I don’t think it’s appreciated in the VC narrative. I built a list of every significant venture-incubated company in US history. Their combined peak value exceeds $2.5 trillion. An enormous amount of value creation has come from systematically created companies. Here are just the top 15:

  1. Intel—Arthur Rock systematically orchestrated the founding, raised $2.5M in 48 hours, and served as first chairman. ~$500B peak.
  2. Palantir—Peter Thiel co-founded it with Joe Lonsdale, Stephen Cohen and Nathan Gettings, then recruited Alex Karp as CEO from outside. ~$500B peak.
  3. Palo Alto Networks—Nir Zuk worked from Greylock’s offices for nine months; Greylock’s Asheem Chandna co-developed the vision. ~$340B peak.
  4. Adobe—A Xerox PARC spinout: John Warnock and Charles Geschke left to commercialise what became PostScript. ~$330B peak.
  5. CrowdStrike—George Kurtz joined Warburg Pincus as an entrepreneur-in-residence in 2011, developed the concept inside Warburg’s offices, and Warburg funded the $25M Series A. ~$285B peak.
  6. Sun Microsystems—Co-founded by Vinod Khosla with Kleiner Perkins backing; Khosla joined KP as a partner soon after. ~$200B peak.
  7. Moderna—Originated as internal project “LS18” inside Flagship Pioneering’s labs; Noubar Afeyan recruited Stéphane Bancel as CEO. ~$200B peak.
  8. Snowflake—Covered in detail below. ~$130B peak.
  9. Waymo—An internal Google X project, spun out as an Alphabet subsidiary. ~$126B valuation (2026).
  10. Genentech—Bob Swanson, a former Kleiner Perkins associate, partnered with Herbert Boyer; Tom Perkins served as chairman. The first biotech company ever, and roughly a 200x return on Kleiner’s money by its 1980 IPO. ~$105B implied value at Roche’s 2009 buyout.
  11. Illumina—ARCH Venture Partners co-founded it around university research on optical bead arrays. ~$75B peak.
  12. Juniper Networks—Vinod Khosla incubated it at Kleiner Perkins and personally suggested the internet router concept. ~$70B peak. Khosla reportedly made KP $10B in profits from incubated companies.
  13. MongoDB—Kevin Ryan co-founded it with Dwight Merriman and Eliot Horowitz out of his AlleyCorp incubator, and still serves as chairman. ~$50B peak.
  14. Twitter—A side project that spun out of Ev Williams’s Obvious Corp. ~$47B peak.
  15. Affirm—Spun out of Max Levchin’s HVF (“Hard, Valuable, Fun”) innovation lab in 2012. ~$47B peak.

Snowflake is the example that should settle this. Mike Speiser at Sutter Hill identified the thesis (a cloud data warehouse), convinced ex-Oracle engineers Benoit Dageville and Thierry Cruanes to build it, served as founding CEO (typically two days a week), and later recruited Frank Slootman to scale it. Sutter Hill owned roughly 20% at the IPO—the largest software IPO ever at the time—and its stake was worth about $12.6B by the end of the first day of trading, on an investment of less than $200M.

That is arguably the single best venture return of the decade, and it was pure systematic creation. Not “backed a founder with a childhood revelation.” The firm identified the opportunity, recruited the operators, and served as founding CEO until the product was built.

And it wasn’t a one-off. Speiser had already run the same play at Pure Storage, where Sutter Hill owned 27% at the 2015 IPO. Same partner, same process, two outcomes most firms never see once.

Other reasons I don’t agree with your take

1. Company building is skill, not art

Skills compound with reps; art supposedly requires a muse. The process is:

  1. Identify the problem.
  2. Test the hypothesis.
  3. Build the solution.
  4. Scale the solution using well-known playbooks.

2. Australia has a history of systematically built companies too

Eucalyptus was arguably a systematically built company (a copy of Hims & Hers and the Unilever playbook). Fittingly, Hims & Hers was itself incubated inside a venture studio, Atomic.

The canonical non-tech example is Fortescue Metals Group. Andrew Forrest did not invent Pilbara iron ore mining—BHP and Rio Tinto had been there for decades. FMG (2003) was a systematic bet that a third major producer could be built by:

  1. Acquiring tenements adjacent to or between BHP and Rio deposits.
  2. Building independent rail and port infrastructure as a “third force” strategy.
  3. Raising capital at exactly the right moment in the China demand cycle.

Forrest had already crashed Anaconda Nickel. The second swing was systematic execution in someone else’s commodity category. His personal attributes—Pilbara family connections, political risk appetite, willingness to personally guarantee debt—absolutely mattered. But they were operator-fit for a systematically identified opportunity, not a childhood insight that birthed the company idea. The thesis came first; the operator was selected for the thesis. That’s the pattern I’m defending.

Other classic Australian copies:

  • Flight Centre—Graham Turner took what he’d learned running overland tours at Topdeck and applied it to discount airfares, opening the first store in 1981.
  • Westfield—Frank Lowy explicitly built American-style shopping centres, after studying the format in the US.
  • Bunnings Warehouse—The Home Depot warehouse model.
  • Seek, realestate.com.au and Carsales—The classifieds trio all took the newspaper classifieds model online, the same playbook as Monster, Rightmove and AutoTrader.
  • Boost Juice—Janine Allis cloned the US juice-bar model after a trip there.
  • Aesop—A Kiehl’s-style apothecary aesthetic.

3. Companies are just a coordination mechanism to solve problems in the real world

The founder is a coordination mechanism for talent, capital, and market timing—not a source of mystical insight.

The ITN framework that effective altruists use to prioritise global problems (Importance × Tractability × Neglectedness, formalised by Open Philanthropy and 80,000 Hours) maps cleanly onto venture-style company building. Pick a market that’s important (big), neglected (incumbents won’t defend it), and tractable (the enabling conditions—regulation, technology, demographics—have just become favourable).

Every one of the Australian examples above fits this framework in retrospect. Which means it can be applied prospectively.

4. The Elon formula

Find a civilisational-scale problem, identify the core monetisable milestone or bottleneck to solving it, then throw unimaginable force at solving it.

Every part of this is systematic and teachable. “Find a civilisational problem” is importance. “Identify the bottleneck” is tractability. “Throw force at it” is execution. No step requires a childhood revelation or a personal muse—it requires reading, research, capital, and discipline.

5. China is one of the greatest case studies in the systematic creation of technology

The 863 Program, “indigenous innovation” policy (2006), Made in China 2025, and the explicit state strategy of fast-following Western technology in semiconductors, EVs, batteries, solar, and AI have produced BYD (now the world’s largest EV maker), CATL (the largest battery manufacturer), DJI (drone market dominance), and SHEIN (the Zara and H&M killer).

None required “deeply personal founder insight.” They required identifying a category, mobilising capital and talent, and executing faster than incumbents could respond.

6. Books that support this thesis

  • Where Good Ideas Come From, Steven Johnson
  • Borrowing Brilliance, David Kord Murray
  • American Independent Inventors in an Era of Corporate R&D, Eric S. Hintz (MIT Press, 2021)
  • The Myths of Innovation, Scott Berkun
  • A Technique for Producing Ideas, James Webb Young
  • Disciplined Entrepreneurship, Bill Aulet
  • The Idea Factory: Bell Labs and the Great Age of American Innovation, Jon Gertner

7. The literature on building companies from nothing

The literature on building companies from nothing—searching for problems—starts with Paul Graham’s essay on startup ideas. The key passage is Graham’s argument that “the way to get startup ideas is not to try to think of startup ideas. It’s to look for problems, preferably problems you have yourself.”

Which is the opposite of the “personal insight from childhood” myth. Graham is saying: put yourself in a position to notice what’s missing, and then go build it. That’s a systematic posture.

Virtually every major figure in the startup world has published a systematic process. This was settled decades ago in the operator community:

  • Steve Blank—The Four Steps to the Epiphany (2005) and The Startup Owner’s Manual. Blank coined “customer development” as a systematic parallel to product development, and teaches it at Stanford, Berkeley, and Columbia. His core claim—“no business plan survives first contact with customers”—is why you build a systematic discovery process instead. This is the intellectual foundation of the entire Lean Startup movement.
  • Eric Ries—The Lean Startup (2011). Build-measure-learn loops, validated learning, minimum viable products. Ries explicitly frames entrepreneurship as management: a learnable discipline, not an art. The book has sold millions of copies and is the standard text at most accelerators.
  • Bill Aulet—Disciplined Entrepreneurship: 24 Steps to a Successful Startup (2013). Aulet runs the Martin Trust Center for MIT Entrepreneurship and teaches a literal 24-step framework. The title alone is the argument. The companion workbook is used in more than 100 universities.
  • Marc Andreessen—“The Pmarca Guide to Startups” (2007), plus his a16z writing. His “only thing that matters” essay is a systematic framework for product-market fit. His entire thesis at a16z is that venture is a craft with learnable patterns.
  • Reid Hoffman—Blitzscaling (2018, with Chris Yeh), taught as a Stanford course. A systematic framework for when and how to prioritise speed over efficiency through defined company stages (Family → Tribe → Village → City → Nation).
  • Ben Horowitz—The Hard Thing About Hard Things (2014). Explicit operational playbooks for the situations nobody else writes about. Horowitz’s whole thesis is that the “wartime CEO” role has systematic patterns, not mystical instincts.
  • Sam Altman—“Startup Playbook” (2015). Literally titled a playbook. Altman ran YC for five years and distilled the pattern: great idea, great product, great team, great execution.
  • Geoffrey Moore—Crossing the Chasm (1991). The systematic framework for taking technology from early adopters to the mainstream, and required reading at every B2B startup. Moore treats market development as engineering.
  • Clayton Christensen—The Innovator’s Dilemma (1997) and Competing Against Luck (2016). His jobs-to-be-done framework is a systematic method for identifying what customers actually want, stripped of demographic or psychographic intuition.
  • Tony Ulwick—What Customers Want (2005) and Jobs to be Done (2016). Ulwick’s Outcome-Driven Innovation is arguably the most rigorous systematic method in the field; he built a consulting practice (Strategyn) around applying it to Fortune 500 innovation.
  • Rob Fitzpatrick—The Mom Test (2013). A systematic framework for customer interviews that don’t produce biased answers.
  • Alexander Osterwalder—Business Model Generation (2010) and Value Proposition Design (2014). The Business Model Canvas is now the default whiteboard tool in every accelerator on earth. It is the definition of a systematic framework.
  • Hamilton Helmer—7 Powers (2016). Helmer, a working strategy advisor to Netflix and others, argues durable advantage comes from exactly seven structural powers. Not genius. Not vision. Seven identifiable, learnable positions.
  • Peter Thiel—Zero to One (2014). Despite the “invention” framing, Thiel’s actual content is systematic: seven questions every startup must answer, definite vs. indefinite optimism, last-mover advantage.
  • Keith Rabois—Hasn’t written a book, but his Stanford lectures on execution are widely treated as canonical playbooks. Rabois is explicit that execution beats ideas in virtually every interview he gives.
  • Andy Rachleff (Benchmark, Wealthfront)—Coined “product/market fit” as a measurable, systematic concept, and teaches it at Stanford GSB.

Every one of these authors is among the most respected operators or investors in tech. Not one of them frames company building as art. They all frame it as a learnable, systematic discipline with identifiable patterns, frameworks, and processes. The “company building is personal art” view is contradicted by the entire published corpus of the people who have actually built the most valuable companies in history.

8. Most unicorn founders are on their second or third company, and around 40 when they start

The same pattern holds for many of the greatest artists. Again, company building is more skill than art.

  • Van Gogh painted roughly 860 oil paintings, but only worked as an artist for about ten years. The Starry Night (1889) came near the end, after he’d already made 700+ works. The Potato Eaters (1885), his first major ambitious work, came after roughly four years and hundreds of drawings and studies.
  • Picasso produced an estimated 50,000 works across all media. Les Demoiselles d’Avignon (1907), the painting that arguably birthed Cubism, came when he was 25, after thousands of drawings and paintings through his Blue and Rose periods.
  • Monet painted Impression, Sunrise (1872) in his early thirties, after more than a decade of serious work. The water lilies came decades later, after thousands of canvases.
  • Leonardo da Vinci is the interesting outlier—only 15 to 20 paintings survive, but he produced thousands of drawings, studies, and notebook sketches. The Last Supper came in his forties, the Mona Lisa even later.
  • Michelangelo completed the Sistine Chapel ceiling at 37, but had been sculpting and painting seriously since his early teens—roughly 25 years of intense work before that particular peak.

9. We allocate capital so poorly

One of the saddest parts of entrepreneurship is seeing people work on bad problems. We allocate capital so poorly. This can be solved with better coordination mechanisms, created by less asymmetric information. VCs, for example, have differentiated insights from research into new markets, and those can be shared with founders.

The gap is that nobody has stitched the existing pieces together into a single instrument for venture capital:

  • Open Philanthropy runs cause prioritisation at scale using the ITN framework (Holden Karnofsky, ~2013), and publicly documents its methodology and grant decisions.
  • 80,000 Hours (Benjamin Todd, Will MacAskill) maintains problem profiles ranking world problems by expected impact, with career pathways—essentially this for careers, not ventures.
  • GiveWell operates the most rigorous cost-effectiveness evaluation in philanthropy (dollars per DALY).
  • Emergent Ventures (Tyler Cowen, Mercatus) is explicit talent-scouting grantmaking—picking people, not projects.
  • Schmidt Futures, and Nat Friedman and Daniel Gross’s AI Grant and Vesuvius Challenge, are the technical-talent versions.
  • Fifty Years is the closest existing “progress studies” venture fund.
  • Convergent Research and Focused Research Organizations are Adam Marblestone’s answer for problem-first R&D institutions that sit between academia, startups, and national labs.
  • DARPA and the UK’s ARIA are the civilisational-scale versions—program-manager-driven R&D allocation.
  • YC’s Requests for Startups is the closest analogue to a “what should people work on” map, but at much lower resolution.
  • Our World in Data (Max Roser, and the Hans Rosling legacy) is the quantitative spine for tracking human progress.
  • The Progress Studies movement (Patrick Collison, Tyler Cowen, Jason Crawford at Roots of Progress, Works in Progress magazine) is the intellectual movement explicitly trying to make this a discipline.

10. The academic data is unambiguous

Tellis and Golder (1993, 1996) studied roughly 500 brands across 50 categories. Market pioneers fail 47% of the time and hold a mean market share of only 10%. Early leaders, entering on average about 13 years later, dominate. Chux invented disposable diapers and died; Pampers and Huggies split the category. Reynolds pioneered ballpoints; Bic won. Netscape invented the commercial browser; Microsoft erased it.

Azoulay, Jones, Kim and Miranda (2020, American Economic Review: Insights) used US Census data on all growth-oriented startups from 2007 to 2014. The average age of founders of the top 0.1% fastest-growing ventures is 45. A 50-year-old founder is about 1.8 times more likely to build a runaway success than a 30-year-old, and prior experience in the same industry sharply raises the odds. That’s not “deeply personal insight from childhood”—that’s 15 to 20 years of pattern recognition earned through time in market.

Even Peter Thiel concedes this in Zero to One: “It’s much better to be the last mover—that is, to make the last great development in a specific market and enjoy years or even decades of monopoly profits.”

Worth noting: Thiel’s last mover isn’t just the last copycat. He means the last durable innovator, the one who finally gets the product right. Google beat AltaVista because PageRank genuinely worked better, not because it copied. But this supports the systematic-creation argument rather than undermining it. “Finally getting it right” is a coordination and execution problem, not a mystical-insight problem. You can systematically identify what previous attempts got wrong and systematically build something better. That’s exactly what studios like Atomic do with every new brand.

11. The “art” framing is low agency

To me, “art” rather than “skill” is a far less inspiring view of company building, because it creates a huge obstacle to becoming an entrepreneur: that you need to start with a unique insight and can’t systematically earn your way to one. It assumes you can’t find a problem worth working on and go solve it. It tells aspiring founders they either have “it” or they don’t.

The skill framing says: study the market, find the best existing approach, and execute with more intensity than anyone else is willing to muster. There are countless examples of this creating some of the biggest companies on the planet:

  • Larry Ellison read Edgar Codd’s 1970 IBM paper on relational databases and beat IBM to market with IBM’s own research. Oracle shipped in 1979, two years before IBM’s SQL/DS. Ellison even tried to make Oracle error-code-compatible with IBM’s System R.
  • Bill Gates bought QDOS from Seattle Computer Products for about $75,000 in 1981 and rebranded it MS-DOS to fulfil the IBM contract. He didn’t write DOS.
  • Steve Jobs lifted the GUI, mouse, and bitmapped display from Xerox PARC in 1979. On camera in 1996 he said: “Picasso had a saying—good artists copy, great artists steal. And we have always been shameless about stealing great ideas.”
  • Mark Zuckerberg launched Facebook after Friendster (2002) and MySpace (2003) had already proven social networking. Facebook was roughly the tenth social network.
  • Google was the 21st search engine to market—after AltaVista, Yahoo, Lycos, Excite, Ask Jeeves, Infoseek, HotBot, and WebCrawler.
  • Sam Walton copied Kmart’s discount-retail playbook and deployed it in towns under 25,000 people that Kmart had decided were too small. He toured every Kmart he could reach, taking notes on pricing systems.
  • Howard Schultz had zero coffee background. He was a housewares salesman on a 1983 business trip to Milan when he wandered into an espresso bar, and said later: “It was so immediate and physical that I was shaking.” He visited 500+ Italian espresso bars, photographed menus, measured counter heights, and cloned the concept.
  • Henry Ford’s foreman William “Pa” Klann brought the assembly line back from a Chicago slaughterhouse (Swift & Company)—meatpackers had been running disassembly lines for decades. Ford acknowledged this in his autobiography. Ransom Olds had been running a stationary assembly line since 1901, twelve years before Ford’s moving line.
  • Ray Kroc didn’t invent McDonald’s—the McDonald brothers did, in 1940. Kroc was a 52-year-old milkshake-machine salesman who saw their operation in 1954 and franchised it.
  • Phil Knight’s Nike started as Blue Ribbon Sports, importing Onitsuka Tiger running shoes from Japan. The waffle sole was his coach’s riff on a literal waffle iron.
  • John D. Rockefeller didn’t invent oil refining. He copied best-in-class techniques from the competitors he acquired, then scaled them ruthlessly.

12. The AI argument

Why this model gets more credible from here, not less—though it cuts both ways.

Everything above is the historical case. I think the forward-looking case is stronger, but it requires honesty about what AI does and doesn’t change.

What AI genuinely makes easier for systematic creators

  • Thesis generation and validation. A founder in 1995 needed months of customer interviews, market research, and prototyping to stress-test a thesis. A founder in 2026 can generate 50 variants of a market thesis in an afternoon, pressure-test each against comparable companies, simulate unit economics, and run real customer discovery in parallel.
  • Prototype building. A competent technical founder with Claude Code, Cursor, v0, or Lovable can now build in weeks what used to require a $2M seed round and six engineers.
  • Operator sourcing and diligence. AI-augmented candidate identification and screening dramatically compresses the recruiting cycle.

What AI does not make easier—and what arguably gets harder

  • The scarce inputs become more valuable. Capital, distribution, networks, regulatory access, compute, and operator density don’t get easier to accumulate. AI doesn’t give you LP relationships, enterprise sales pipelines, political connections, or a distribution flywheel. Those still have to be built the old-fashioned way, over years.
  • Attracting elite operators gets harder. The best founders now have more options—they can start alone with AI far more easily than they could in 2015. Winning them to a thesis-driven opportunity means offering genuinely scarce inputs they can’t replicate solo.
  • Thesis advantages compress faster. If AI makes thesis generation cheap, every other systematic creator is generating the same theses. Your thesis edge decays unless your distribution, capital, or operator-recruiting edge is structural and compounding.

The synthesis

AI commoditises the front end of systematic creation: thesis generation, validation, prototyping, diligence. That is bad for any systematic creator whose only edge was being smarter about thesis work, and good for any systematic creator who has accumulated the scarce non-AI inputs.

The durable edge shifts almost entirely to capital, distribution, networks, political and regulatory access, compute allocation, and operator density. A thesis-driven investor without those edges will do worse in 2026 than in 2015. A thesis-driven investor with them will do dramatically better, because the part they can’t replicate is exactly what becomes scarce.

So the founders with the biggest moats in a post-AI world will increasingly be the ones backed by operators who systematically accumulate the scarce inputs—not the ones who showed up with a childhood revelation. Substitutes for personal insight are getting cheaper. The scarce inputs around personal insight—coordination, capital, distribution—are getting more valuable.

Conclusion

There’s no law of physics saying companies cannot be systematically created. The evidence is overwhelmingly in favour of it.

The most productive synthesis is probably that you use systematic methods to identify the opportunity, and then find the operator uniquely suited to building it—which is exactly the definition of incubation I opened with. Not “factory,” not venture-studio economics, not “replace the founder.” Systematic thesis development plus founder-market fit, underwritten by the scarce resources that matter in a world where information and execution are commoditised.