Summary

Stage-specific firms appear to pick better, while megafunds win by owning more of the board. Across roughly 15 years, Benchmark made about 63 Series A investments and produced six $5 billion hits—a 10% hit rate—while Andreessen made 454 and produced ten, or 2%. Rory O’Driscoll called that “exactly the dilemma of the megafund versus the focus fund”: superior selection versus greater absolute capture.
Megafunds’ decisive advantage is that early-stage venture can become a loss leader for later private-market ownership. With 50–60% of industry capital, large firms can offer aggressive seed pricing, treat Series A like “milk at the grocery store,” and monetize at C, D, and beyond. Their long-term math still requires several companies to compound from roughly $100 billion to $400 billion while private; Rory saw only SpaceX and OpenAI fitting that pattern today.
Windsurf’s potential $3 billion acquisition underscores both rapid product pivots and the strategic value of AI distribution. Rory framed the price as roughly 1% of OpenAI’s market capitalization for entry into one of AI’s largest use cases; Jason Lemkin, already “anesthetized” by $10 billion startup valuations, found $3 billion almost modest. Cursor’s decision to remain independent, assuming it rejected an offer, now means competing against the acquirer-backed competitor: “It takes real courage.”
Every investor’s M&A advice is filtered through their own cap table, fund size, and entry price. A recent investor may welcome a quick 1X, an earlier investor may see the same sale as the destruction of a fund returner, and seed holders may already have sold secondaries. The corrective is to “reinforce success, starve failure”: the final double—from $6 billion to $12 billion, perhaps in 12 months—can matter more than years spent reaching the first $10 million of ARR.
Jason Lemkin and Jason Calacanis expect AI to eliminate white-collar roles much faster than Rory does. Jason Calacanis said running his own AI on 130,000 conversations changed him from skeptical to “100% convinced” that half of many knowledge-work categories could disappear within 24 months; Jason Lemkin separately said SaaStr had gotten rid of five people in 90 days due to AI. Harry argued that deep research requires no hardware-installation cycle. Rory rejected mass unemployment within 12–24 months, but conceded that tech’s larger economic share could make adoption “boomier quicker.”
Decagon’s 100X ARR valuation is a bet on support automation’s measurable ROI, not merely AI branding. References suggested generative AI can raise automated resolution from 30–35% to 60–70%, supporting a huge labor-replacement market; the bullish case turns $15 million of ARR into $50 million, $150 million, then $300 million. Harry’s counter was that Intercom, Sierra, incumbents, and vertical YC startups make the implied probability absurdly generous: “You’re paying for it.”
AI may maim incumbents without killing them—and that is enough to erase venture returns. Jason’s feared outcome is a leader falling from 50% growth at $500 million to 30% at $300 million through churn, downgrades, and pricing pressure: alive, but no longer IPO-ready. Rory still expects enterprise integration and default-vendor status to produce three- or four-player oligopolies, while conceding that models absorbing more of the software stack make today’s stability unusually uncertain.
Endowment pressure and venture’s “arrogance score” both favor established capital providers over emerging managers. If universities build precautionary cash reserves, the $150 million first-time fund loses a core LP while the $8 billion platform turns to sovereign wealth; weakening research universities would also damage venture’s talent and IP pipeline. Josh Kopelman’s underlying test remains unforgiving: calculate the market share and duration needed for the fund model, remembering that “time value of money is a bitch” and a 2X in 10 years can resemble a 4X in 17.

Deep dive

1. Windsurf’s potential $3 billion acquisition followed a two-year pivot

Jason had previously considered $3 billion enormous, but startup pricing has moved so far that he now feels “anesthetized to these numbers” unless they reach tens of billions. He compared the mood with 2021, when a project-management company might reject the same offer.
Rory’s framing was celebratory and concrete: start a company three or four years ago, pivot roughly two years ago into a VS Code-style fork, execute relentlessly, and potentially sell for $3 billion. “All you have to do is just get it right. Bob’s your uncle.”
If completed, Rory saw straightforward strategic arithmetic for OpenAI: spend roughly 1% of market capitalization to enter one of AI’s largest use cases, loved by the company’s developer constituency. The acquisition therefore “just makes total sense for both sides.”
Cursor’s risk was foreseeable when it allegedly declined its own offer, if such an offer existed. Rory’s distinction: a number two often folds before an adjacent buyer acquires number one; the leader can bet on remaining “the independent winner,” but must now compete against the acquirer-backed competitor.

2. M&A advice is inseparable from each investor’s position

Jason observed that investors “talk their book,” even when their behavior does not look economically logical. Before weighing boardroom advice, a CEO should understand every participant’s entry valuation, ownership, liquidity, fund exposure, and internal calculation of “What does it mean for me?”
At Cursor, an investor who just entered at $10 billion might calmly accept a risk disclosed during diligence; someone who invested around $1 billion might consider refusing a sale disastrous. Seed investors, perhaps already half-liquid through secondaries, can afford to say, “I love you guys, and you be you.”
Fund scale has softened the stigma of quick, low-multiple exits. Andreessen appeared content recovering roughly 1X from Loom, while Thrive’s near-immediate 2X on Instagram generated reputational value far beyond its raw return: appearing spectacularly prescient helped establish the franchise.

3. The final double is venture’s scarce advantage over private equity

Jason suggested that an early investor becomes risk-averse once a 10X markup could materially return a smaller fund. Rory recognized the fear but called overprotecting winners a mistake: venture’s first rule is “let the winners run,” or, in military language, “reinforce success, starve failure.”
Rory recalled taking money off the table in a winner and later realizing another 2X could have transformed the outcome. The emotional instinct—“Oh my God, let nothing go wrong”—protects a paper gain while potentially sacrificing the portfolio’s defining return.
Harry preserved Brian Singerman’s lesson about “the final double”: moving from $6 billion to $12 billion may require only another 12 months, yet doubles the portfolio return. Rory contrasted that with grinding from $1 million to $5 million of ARR and celebrating comparatively little value creation.
That asymmetry is venture’s edge over PE. Occasionally an investor owns 10% of a business already worth billions and compounding toward tens or hundreds of billions; the correct response may be to “lie back” and discover how far the outlier can travel.

4. Megafunds have already won access, but not yet proved returns

Harry said he had reversed his view: five to seven large private-market providers may dominate because sovereign wealth funds and other major LPs have largely selected their partners, while companies such as Anthropic, Glean, and Rippling have few places capable of financing enormous private rounds.
Rory separated possession of capital from profitable deployment. A firm holding $7–8 billion “has won” the first contest, and firms controlling 50–60% of capital should capture a similar share of wins through sheer participation; complaints that they own everything often reduce to “just math.”
The unresolved test arrives when LPs assess whether each multibillion-dollar pool cleared its hurdle. For the next three to five years, Rory expects “great big walls of capital” to trample smaller firms’ economics, even if some mega-strategies ultimately disappoint over five to ten years.
Harry’s defense depends on outcome size, not better early-stage selection. Rory agreed: large strategies work if several businesses compound from roughly $100 billion to $400 billion before listing; beyond SpaceX and OpenAI, another four or five such assets could make the arithmetic work.

5. Seed and Series A are becoming loss leaders in a bundled product

Rory compared the megafund model with supermarket merchandising: seed and Series A are cheap milk that brings founders inside, while Series C and D are “all the strawberries you can buy.” The most useful feature of an $8 billion platform remains “$8 billion.”
Harry asked whether a multistage firm can still win Series A without doing pre-seed. Rory said Scale was actively wrestling with that question after losing a deal largely because another investor had built the seed relationship; if Scale enters seed, it will not “bullshit” founders with a token program.
Founders generally want abundant money, minimal friction, and perhaps maximum help—not fidelity to an investor’s nuanced stage strategy. Harry countered that founders ask whether he makes two or three investments rather than 12 because they want attention; Rory replied that higher portfolio activity also creates news flow, relevance, and constant visible success.

6. Focused firms’ superior hit rate collides with scaled firms’ absolute wins

The cited Series A analysis showed Benchmark making roughly 63 investments across about 15 years and producing six $5 billion hits—a 10% hit rate. Andreessen made 454 Series As and produced ten such outcomes, yielding about 2%.
Rory’s conclusion was deliberately narrow: assuming both firms employ exceptional investors, the comparison isolates strategy. “The focus fund is better at hit rate” and proportionate success; the high-volume platform sacrifices quality while increasing its absolute number of hits.
Jason cautioned that 2013–2018 bears little resemblance to today. Rory agreed but found the older, easier period even more sobering: the most aggressive firm captured only about 10% of great outcomes, Benchmark roughly 6%, and others clustered below—so six or eight giant funds cannot each own 10% now.

7. AI labor displacement split the table on speed, not direction

Jason Calacanis said his view changed completely after his team ran its own AI on 130,000 conversations. He became “100% convinced” that half of many tech knowledge-work roles could disappear within 24 months. Jason Lemkin separately said SaaStr had gotten rid of five people in 90 days due to AI.
The vulnerable list included SMB sales, marketing management, customer success, mediocre QA, and mediocre product management. Once AI reaches “80% as good as a human,” Jason argued, employers will prefer an agent that “doesn’t complain about the job.”
Harry supported the fast-adoption case: deep research can replace research capacity without buying PCs, laying fiber, or running an implementation cycle. His team could press a button inside a model it already used and eliminate the need for three researchers.
Rory accepted automation and OpEx savings but rejected mass unemployment within 12–24 months. His base case was directionally similar but slower: firms will make people more productive, examine more opportunities, and perhaps remove one or two roles rather than experience an immediate step-function collapse.

8. Tech-heavy economies may absorb AI faster than prior technologies

Rory anchored his skepticism in roughly 2% long-run productivity growth since the Industrial Revolution. Electricity, steam power, PCs, and the internet were also transformative, yet diffusion remained slow; enterprises still running DOS-era software illustrate how uneven adoption can be.
Jason’s strongest rebuttal was sectoral: technology was not the economy’s largest segment when they began investing, whereas today the early-adopter population is enormous. Rory conceded this could create accelerated returns to scale and make AI “boomier quicker” than PCs or the internet.
The labor-market consequence may be severe even without a macro step change. Harry cited shortages in plumbing, roofing, trucking, emergency services, and other vocational work; Rory invoked Peter Thiel’s argument that marginal college entrants can acquire weakly marketable skills plus $150,000 of debt, while “there’s always gonna be room at the top.”

9. Weakening university endowments would reinforce venture concentration

Harry asked whether the threatened removal of Harvard’s tax-exempt status could spread across endowments and reduce venture commitments. Rory avoided the wider political judgment but expected pressure to trigger “precautionary cash planning” throughout the university LP base.
That would fall disproportionately on emerging managers. An $8 billion fund can rely on sovereign wealth; a first-time $150 million fund still needs the endowments that historically back small, early, innovative franchises—another mechanism through which “the big will get bigger.”
Rory also defended universities as venture infrastructure: software, biotech, robotics, and other US advantages depend on research, intellectual property, and technically trained graduates. Whatever Harvard’s institutional failures, “don’t kill the golden goose” and “don’t blow” the talent pipeline.

10. The unicorn cleanup will produce radically different outcomes by cap-table layer

Jason called Census’s quiet acquisition by Fivetran an “acqui-ouch”: a formerly coveted company had raised about $80 million, including from Sequoia, yet did not produce the expected breakout. For a seed investor counting it as a fund returner, shares in the acquirer can be poor consolation.
Rory barely noticed because he expects hundreds more. Of roughly 800–1,000 unicorns, perhaps a couple hundred can become public; most of the remainder will be “scrunched into other companies,” often without a dramatic failure announcement.
Lacework illustrated a potentially rational surrender. Rory, explicitly lacking specifics, had heard that investors may have preferred recovering perhaps 50–70 cents per dollar while substantial cash remained—conceptually, spend $100 million, preserve $800 million, and stop—rather than finance a deteriorating thesis.
Jason’s advice to eclipsed companies was “take the effing offer.” He contrasted Olo’s need to find a buyer with SevenRooms selling to DoorDash for $1.2 billion; DoorDash’s $2.9 billion Deliveroo purchase, versus a UK valuation near $1.4–1.5 billion, showed what a scaled US platform can pay to color in Europe.

11. Decagon’s 100X ARR price sells option value, not intrinsic value

At roughly $15 million of ARR and a $1.5 billion valuation, Decagon priced near 100X. Jason Lemkin said that was “not wholly crazy”: customer-support references suggested generative AI could raise automated resolution from about 30–35% to 60–70%, removing humans from most calls in an enormous labor pool.
Harry’s pushback was competition. Intercom’s excellent product and 17-year journey produced an enterprise value around $2 billion; Decagon must also face Sierra, Brett Taylor, incumbents, and dozens of verticalized YC companies. “Getting to 150 million in ARR from 15 is a journey.”
Rory explained why investors still lean in: a mature company at $400 million growing 20%, bought and sold near 6X, offers a bounded double after roughly four years. A new company jumping from a few million to $15–25 million can be imagined reaching $50 million, $150 million, then $300 million at 20X—a $6 billion story supporting a $1.5 billion entry.
The error may be probability, not possibility. The future could occur one time in 100 while investors price it like one in two: “Venture guys love new shit with option value over old shit with intrinsic value. We are upside junkies.”

12. AI can destroy a venture outcome without displacing the incumbent

Jason’s “moats versus momentum” concern was that Decagon may deploy well today, yet a better competitor could appear next year. Rory countered that market windows close: momentum becomes a moat when a vendor reaches critical mass, becomes the safe enterprise choice, and secures a 10- or 15-year slot.
Rory expects customer support eventually to resemble other enterprise categories: incumbents such as Intercom or Gorgias adapt, two or three AI-native companies explode, and the market settles into a modest oligopoly. Deep integrations, tuning, references, and default status make large installations difficult to remove.
Jason no longer trusts that stable state. Disruption now arrives “every five weeks,” models may absorb more application functionality, and realistic digital support people could emerge outside today’s vendors. Rory conceded the countervailing force: “You just have to internalize what the models are gonna do” because parts of the stack may simply be done for software companies.
The crucial risk is being maimed rather than killed: pricing pressure, churn, and downgrades can turn 50% growth at $500 million into 30% at $300 million, eliminating an IPO. HubSpot’s reported 50% productivity gain—and more features than it can operationally release—was Jason’s warning against assuming any SaaS position is settled.

13. Venture’s arrogance score reduces every strategy to market share and time

Rory endorsed Josh Kopelman’s underlying question: are there enough suitable deals, and what share must a fund capture, for its model to work? Historical data suggest nobody achieved the early-stage market share now implied by several simultaneous $5–10 billion strategies.
The escape route is later-stage private compounding, probably at lower returns but perhaps still above the hurdle. Duration matters brutally: Harry highlighted that a 2X over 10 years can resemble a 4X over 17, and Rory’s summary was blunt—“Time value of money is a bitch.”
Every firm therefore runs a coverage strategy, whether targeting ten deals or monitoring 100. Content lets Harry preserve relevance without financing unwanted companies; Rory’s closing instruction was to accept that abundant capital defines the next five years, even if it later proves unstable: “Quit bitching and play it.”

No Priors Ep. 114 | With Duolingo CEO Luis von Ahn Doug Burgum, Secretary of the Interior | All-In DC