[BidClub_]
20VC · · 76 min

20VC: Sequoia's Leadership Transition | Michael Burry Shorts NVIDIA and Palantir | Gamma Raises $100M at $2BN | Has Defensibility Died in a World of AI | Datadog Surges as Duolingo Plummets: What is Happening

Harry Stebbings

Podcast ↗
TL;DR
  • AI’s revenue evidence now makes outright pessimism harder than arguing about the slope of growth. Harry cited Altman saying OpenAI would hit $20 billion of ARR this year, Anthropic projecting $70 billion by 2028, and Gamma reaching $100 million of revenue with 50 people. Rory’s verdict: “Right here, right now, the revenue has shown up”; the investable uncertainty is whether next year’s CapEx is $80 billion or $40 billion, not whether demand exists.
  • Michael Burry’s $1.1 billion NVIDIA-and-Palantir short can be directionally right and still lose because options force the clock. With NVIDIA at $188, Rory used a 47-day $180 put and a roughly $9 stake: the stock must reach $160 to double the stake, while failing to fall below the strike wipes it out. A two-year put costs $50–55, loses money unless NVIDIA falls below roughly $150, and needs to fall below $100 to produce the same 2X. Moving from an “arm-wavy bullshit podcast statement” about overinvestment to a profitable timed trade is “damn hard.”
  • The next software TAM opens when AI stops being a copilot and becomes part of the team. Jason’s Replit v3 remembers prior work, discusses mistakes, and shipped a new production page in 15 minutes; Gamma already turns company data into bespoke sales collateral in roughly 10 minutes instead of weeks. The threshold is being “sufficiently autonomous, knowledgeable, and powerful” to complete material work with human check-ins — and then “the amount of revenue that’s accessible is so high.”
  • AI has not killed moats; it has moved the point at which they become credible further to the right. Incumbents that once took a year and a half or more to decide whether to clone a product can now produce credible copies in 30–90 days, so an early revenue explosion is “not as defensible” as it used to be. Jason still sees a later “plane of stability,” while Rory argues defensibility emerges only after distribution, engineering depth, and market anointment narrow the field.
  • Later-stage evidence reduces operational risk, but valuations are expanding fast enough to consume that benefit. Rory thinks the probability can narrow from one-in-ten at seed or early A to perhaps one-in-three by the B; Harry argues horizontal AI markets may still be one-in-seven to one-in-ten once Codex, Claude Code, Vercel, Figma Make, Salesforce, Atlassian, and other adjacent competitors are counted. Harry’s cited company moved from roughly $4–5 million of revenue at a $200 million valuation to $80–100 million at $2 billion, showing both genuine de-risking and the price of consensus.
  • More outcome variance logically calls for more diversification, but the ownership math can push a seed fund toward institutional scale. Jason’s example requires 40 initial $5 million checks, another $200 million of reserves, and roughly $100 million for fees and flexibility — a $500 million “little seed fund.” Harry raised the possibility that larger outcomes permit smaller ownership; Fabrice cited a 100–150-position LP fund using $100,000–$150,000 checks and argued that $1 million on a $50 million valuation can work if the company becomes worth $100 billion.
  • Fundraising has become brutally binary, and the best process is cultivated before the company formally raises. Jason sees a “Captain Obvious era” in which YC, Neo, or South Park Commons pedigree—or hot AI-native growth—gets funded while conventional SaaS struggles; Harry’s example grew from $400,000 to $3 million, took 120 meetings, and received one $10 million term sheet at a $40 million post-money valuation. Their shared conclusion: “The best-run processes don’t feel like a process” because several investors are already primed before diligence begins.
  • Public-market AI credit goes to companies capturing new budgets, not merely adding AI features. Datadog’s 23% jump fit the rule “sell shit to the people who are making AI,” while Duolingo’s 25% fall showed that “sprinkling AI dust” earns no premium without new economics. Jason’s hierarchy is explicit: attach to compute spending, replace human labor, or use AI to displace an incumbent and steal its revenue.
Digest · the substance, structured for research

1. Sequoia’s handoff exposes the strain on the old venture playbook

  • Harry framed Roelof Botha’s departure as steward after three years, and Pat Grady and Alfred Lin’s elevation, as unusually consequential. Rory’s outside-in interpretation was blunt: “Whenever you have a CEO change happen, it’s ’cause something is wrong.”

  • Rory carefully stopped short of declaring Sequoia behind in AI. His claim was that an internal perception likely existed that the firm “could do better” after missing rounds, passing on strong companies, and competing against a newly aggressive market.

  • Jason widened the issue beyond Sequoia: many investors, executives, and founders from the last 10–15 years may not suit the next decade. “The old playbook doesn’t work”; for those unwilling to relearn, his prescription was to take their NVIDIA shares, buy a beach house, and check out.

  • Harry’s structural framing was “Walmart versus Chanel”: mega-platforms and walls of capital on one side, focused boutiques such as Benchmark and USV on the other. Splitting early and growth leadership may help specialization, but Rory warned that the person above them becomes a precarious “manager of managers,” increasingly removed from “eat what you kill.”

2. Sequoia’s ruthlessness may be healthier than partnership stability

  • Rory admired Sequoia’s willingness to reject the idea that leadership belonged to someone because “it’s so-and-so’s turn.” If the partnership believed change was needed, it acted rather than protecting an untouchable figure for another five or six years.

  • Jason’s less romantic view was that partnerships are intrinsically dysfunctional because performance rarely maps cleanly to economics. Equal carry feels collegial until one partner produces the winners and another makes the large losing bet: “The daggers are always out.”

  • Rory’s pushback — worth keeping — was not that the daggers disappear, but that economic performance should drive professional change. Sequoia’s willingness to say “It’s not working. Let’s make a change” was, in his view, both healthy and characteristically Sequoia.

3. Michael Burry’s short is a timing trade disguised as a macro thesis

  • Harry introduced Burry’s reported $1.1 billion short exposure to NVIDIA and Palantir. Rory agreed with the broad premise that AI CapEx will eventually overshoot and correct, but tested whether that belief could actually be turned into a profitable trade.

  • With NVIDIA at $188, Rory used a 47-day December $180 put as an example. For every roughly $9 staked, NVIDIA reaching $160 would produce about a 2X, reaching $100 would produce roughly an 8X, and failing to fall below the strike would mean “you lose it all.”

  • Burry’s unusually early filing mattered to Rory because a tightly timed position benefits from publicity. Having reportedly filed on the final permissible day previously, releasing this position early looked like an attempt to “pile on the bad news” and shit-talk the stock without saying so directly.

  • Extending the horizon does not remove the difficulty: a two-year $180 put might cost $50–55, lose money unless NVIDIA falls below roughly $150, and require a drop below $100 to double. Options are zero-sum, and being vaguely correct about an eventual correction is insufficient.

4. Present revenue makes fighting the AI trend “dumb as rocks”

  • Harry challenged the show’s repeated worry about whether AI revenue would arrive: Altman was said to expect OpenAI to hit $20 billion of ARR this year, Anthropic projected $70 billion by 2028, and Gamma had reached $100 million of revenue with only 50 people.

  • Rory put more weight on estimates being raised during the current year than on distant forecasts. Anthropic’s upward revisions were therefore meaningful, while CoreWeave’s problem — getting data centers operational, not finding customers — reinforced the evidence of compute demand.

  • His conclusion was unusually categorical: “The cynics sound smart and optimists get rich.” AI is perhaps the largest megatrend since the early internet, so “leaning into it is the only sensible thing to do, and playing against it is dumb as rocks.”

  • The legitimate second-order debate is sizing. A company spending $80 billion next year rather than $40 billion creates a consequential delta, even though both figures describe explosive demand; questioning extrapolation is therefore risk management, not denial of the trend.

5. Gamma and Replit show how tools become revenue-bearing teammates

  • Gamma raised $100 million at a $2.1 billion valuation after reaching $100 million of revenue. Jason’s SaaStr use case goes beyond “AI PowerPoint”: Gamma pulls Salesforce and marketing-automation data, calculates a sponsor’s prior leads and ROI, identifies competitors and similar companies, and builds bespoke collateral in about 10 minutes.

  • SaaStr pays roughly $100 a month, or $1,200 annually, for work that previously used free Google Slides or Microsoft Office and might take a marketing-operations team three weeks to complete badly. Jason called that stealth TAM expansion. He also argued that if Gamma continued toward $1 billion of ARR, a 20× revenue valuation would not sound expensive; Harry added that the business was profitable.

  • Jason had launched 10 applications in 125 days without an engineer. Replit v3 was the first agent he considered literally part of his team: it retained roughly a month of context, remembered earlier implementation choices, discussed mistakes, and put a new idea into production in 15 minutes.

  • His dividing line is autonomy, knowledge, and the ability to complete material high-value work with periodic oversight. The copilot was the 2024 story and failed as a paid-on tool; the current story is that AI finally works; the 2026 opportunity is AI embedded inside the team rather than merely helping individual employees.

  • Jason framed the surge as a recent capability break: Gamma was founded in 2020 but had no revenue before this year, while he said Replit and Vercel likewise exploded only once the models became good enough.

6. Faster clones move defensibility later, not necessarily to zero

  • Harry described the investor’s new burden as continually relearning what became technically possible during the prior 12 months and what will become possible in the next 12. “What I knew six months ago” can become useless quickly, making stale technological judgment equivalent to betting blind.

  • Jason contrasted the old response cycle, in which a company might take a year and a half to decide whether a clone was worth building and roughly two years before staffing it seriously, with an investment that attracted five clones — including one from a cloud leader — in 30 days. Canva’s presentation product had also become “borderline competitive” with Gamma in the short period since an earlier interview.

  • The counterweight is a later, still-fragile “plane of stability.” Jason argued that sophisticated products such as Replit can pull away because competitors cannot reproduce the underlying agent; Bolt, once an early leader, had fallen to third and outsourced its agent layer to Claude. He also warned that companies now need to work at a “996” intensity because a three-year product cycle can compress to 90 days.

  • Harry proposed vertical data accumulation through Solve Intelligence’s patent workflow, while Jason noted patents are public and ingestible by rivals. Rory’s broader rule was that seed defensibility is mostly imaginary: teams must run fast, stay technically superlative, win distribution, and become one of the market’s “anointed” winners.

7. The valuation question is whether investors are paid for wider variance

  • Jason accepted the high-risk team-and-speed model at a $3 million or $5 million post-money valuation, but questioned it at $50 million with $5 million checks. If innovation can be cloned immediately, investors need to ask whether the entry price compensates them.

  • Rory distinguishes stages probabilistically: at seed, investors mostly know the people; around an early A, five or six companies near $1 million of revenue may still imply one-in-ten odds; by the B, rate of change and competitive ordering can sometimes narrow the plausible leaders to roughly one-in-three.

  • Harry disputed that confidence using code generation and vibe coding. Cursor, Cognition, Replit, Lovable, Vercel, Codex, Claude Code, Salesforce, Atlassian, and Figma Make leave the field “one in many”; being a top venture-backed startup means nothing if an adjacent platform ultimately wins.

  • Rory conceded the remaining platform risk but argued the information gain is still real. Harry’s cited company moved from about $4–5 million of revenue at a $200 million valuation to $80–100 million at $2 billion, removing category and execution risk; valuation then expands to fill the space, leaving the final question: is the TAM large enough?

8. More uncertainty pushes fund construction toward diversification

  • Jason’s arithmetic started with 40 seed investments at $5 million each: $200 million deployed initially, another $200 million in reserves, and perhaps $100 million for fees and flexibility. Maintaining both ownership and diversification could therefore require a $500 million seed vehicle.

  • Harry argued that expanding outcomes permit lower initial ownership. Fabrice then cited an LP investment holding 100–150 positions through $100,000–$150,000 checks that had produced a 7× fund; he also argued that a $1 million investment at a $50 million valuation can work if the company becomes worth $100 billion.

  • Rory observed that Harry’s answer implicitly accepted greater diversification rather than denying the higher risk. Longer exit times and wider operating variance logically imply somewhat more positions, achieved through smaller checks, larger funds, or consciously lower ownership.

9. Investor meeting cadence must fit the strategy and the person

  • Harry said the partnership does 20 new-company meetings per partner weekly; with his four investing partners, that is 80 net-new companies met in person per week and more than 3,500 company meetings annually. Jason’s response was that he would “give you all my carry back” rather than adopt that schedule.

  • Rory called himself a “meeting junkie” because even an average opportunity can reveal a market insight unavailable in a deck. His interrupt-driven style skips most slides to extract the operator’s “crucial kernel of knowledge.”

  • Jason prefers a strong email, deck, financials, and five prior investor updates, now augmented by Claude. Only a truly great founder adds enough beyond that work to justify the meeting; founders with a 0% chance of receiving his investment gain nothing from forcing a coffee.

10. The best fundraising process is built before it officially begins

  • Harry found it abrasive when a founder rejected his ready-to-sign term sheet to launch a process on November 19. If he had already met the requested price, he reasoned, the founder was implicitly optimizing for a different partner and should simply say so.

  • Rory separated a committed offer from vague interest. A real term sheet deserves serious consideration, but sharing data serially with one or two uncommitted firms creates an “accidental process”; if they decline, the company has effectively failed a financing before formally starting one.

  • Jason’s preferred founder cultivates several investors through updates and relationships, then announces a round only when those investors are primed. The optimal version needs no traditional data room: once investors already want to invest, the remaining material can be a diligence file, such as the example Jason labeled “Box-diligence investment 12/21/25.”

  • Rory’s synthesis sharpened the point: “The best-run processes don’t feel like a process, but they are.” Timing pre-existing interest creates a competitive outcome without overt games, although only a sufficiently attractive and high-performing company can execute it.

11. Fundraising is binary, while AI rewards new budget rather than features

  • Jason called the market the “Captain Obvious era”: affiliation with YC, Neo, or South Park Commons can unlock pre-seed attention, while later companies need hot AI-native positioning and top-quartile venture growth. The formerly fundable middle has largely disappeared.

  • Harry’s concrete specimen was a conventional enterprise SaaS company growing from $400,000 to $3 million. It held 120 meetings and received one term sheet: a $10 million round on a $40 million post-money valuation, about 12× revenue for a business that had grown 10×.

  • Datadog represented the opposite outcome: its shares rose 23%, and Harry cited “$15 million plus AI-native customers,” without clarifying the metric. Rory summarized the playbook as “Sell shit to the people who are making AI”; observability, switches, routers, and interconnects all scale with unprecedented compute consumption.

  • Duolingo fell 25% after slightly softer guidance despite a good quarter. Jason called it the “wrong kind of AI”: improving an existing product earns no special credit today. A company must attach to compute, replace human labor, or use AI to seize an incumbent’s revenue.

12. Education and Hummingbird show two versions of capital efficiency

  • Rory resisted reducing every application to layoffs. In education, AI can give students individualized instruction closer to one-on-one tutoring than a class of 20; for language learning, the credible budget may come from adults already paying human coaches, not cash-constrained public schools.

  • Jason accepted the product benefit but kept asking, “Where does the budget come from?” If AI neither replaces labor nor captures compute spending, it must displace a legacy vendor — difficult for incumbents such as Duolingo that may need to cannibalize their own existing revenue.

  • Harry closed with Hummingbird’s first biotech investment, BillionToOne, producing an approximately $800 million position at IPO. He also cited BillionToOne at $5 billion and Nirvana at $4.5 billion, both as examples of the power of capital-efficient companies and concentrated venture ownership.

  • Jason called maintaining ownership from an eight- or nine-figure fund “God tier,” while Rory defended accepting dilution: going from 20% to 12% after investing $4 million can still create a spectacular fund. The small, high-MOIC vehicle is often the more compelling home for the marginal LP dollar.

Jason Lemkin

Tools are great. When the AI is part of your team, for real—not VC talk—the amount of revenue that's accessible is so high.

Rory O'Driscoll

Sell shit to the people who are making AI, and if they grow, you'll sell more shit too.

Jason Lemkin

You just can't take that early, first-month explosion as seriously as you used to. It's not as defensible.

Rory O'Driscoll

The pace of evolution is so fast. If you decide, “Well, what I knew 6 months ago is still useful,” you're probably going to be wrong very quickly, right? That's what I find the most stressful about right now.

Harry Stebbings

Guys, it is so good to be back with you. I've just come back from my AGM, and it's a humbling thing for me because for 10 years I did this show solo, and then I go to my AGM, and do you know what everyone says? “Oh, we love Rory and Jason. We love Rory and Jason.” And I suddenly realized that I wasn't the star of the show.

Jason Lemkin

Just kick us off. End this now.

1. Sequoia Changes Leadership

Harry Stebbings

Fine, I'll kick us off with big, big venture news. We've said before how Sequoia are the kings of venture. There's been a leadership transition at the top. We saw Roelof Botha moving out as steward after a 3-year tenure and being replaced by Pat Grady and Alfred Lin, which was, I think, quite surprising news to the venture ecosystem. I'd love to hear how you thought about it.

Rory O'Driscoll

Sure. It just brings home how tough venture is right now. This is the best firm in the world, and they're feeling what's going on in AI: that they're behind. I think everyone in venture, especially if you have bought a large existing portfolio and you're trying to compete for these new deals in the last 3 years, is feeling stretched. Everyone is feeling tired. Everyone is feeling it's brutally competitive.

This is going to sound kind of—not schadenfreude—but it's more reassuring in a way. I get up and I go, “God, we need to do better. We need to sort this out.” Then you realize the best firm in the world is having exactly the same feelings. That was the first comment.

Harry Stebbings

Why do you think a leadership transition is showing that they're behind in AI?

Rory O'Driscoll

Look, whenever you have a CEO change, it's because something is wrong, right? And again, if you want to go down into the weeds of the interpersonal stuff and people's perceptions of other people, we can talk about that later. But my outside-in reading is that, in part, this is dissatisfaction about how the firm is doing relative to the competition.

Look, I'm not saying they are behind; I'm saying there's a perception internally that they could do better. They missed some rounds and some deals. They passed on some great companies. That can be frustrating.

Jason Lemkin

Well, maybe 2 thoughts. One is that, in general, more people should be stepping aside today. I think more people—VCs, executives, founders. And I'm not saying this is exactly what happened at Sequoia. I mean, that's the articles, right? It's about AI and missing Cursor and missing these deals. Maybe that's true.

Most folks from the last decade or 15 years are not the right people for the next decade. I could only imagine that we don't even need half the VCs we have today for the AI world. Maybe they'll spend the money. Maybe they'll throw a few nickels into Cursor at $30 billion and get a few logos on the website. The old playbook doesn't work, and the pace is so fast. Take your coins, take your NVIDIA shares, and buy a beach house. Seriously, check out. It's a good time to check out, guys.

Rory O'Driscoll

One of the things I admire most about Sequoia is their toughness and their willingness and ability to evolve. I'm not commenting on the merits of the case, but if the internal group feels they need to make a change to continue to execute, what they did not do was make that fatal error—which we can talk about in politics in a second—of saying, “It's someone's turn,” and then saying, “It's so-and-so's turn, so we'll leave him in.”

They did the exact opposite. They ruthlessly said, “If we're going to compete, we need these people, not those people,” and they made a change. And I admire that. All they needed to do was call a vote and have a discussion, sit down with the partner in question, and have a discussion. I think that's healthy.

I think with some of these organizations where so-and-so can't be touched, you always say to yourself, “Does that create a false sense of security when you just can't afford to have that in this market?”

Harry Stebbings

To me, it's like a specialization in leadership that's correlated to the winners and losers in venture in the next 10 years. And I think the winners are Walmart, which is your megaplatform. It's your Thrive, it's your Lightspeed, it's your General Catalyst, it's your walls of money, and it's your boutiques, which are your Chanels. Walmart versus Chanel.

The boutiques are your Benchmarks, your USVs, your specific products. And I think when you see this splintering of leadership between Alfred Lin on Early and Pat Grady on Growth, I wonder if it's this kind of attempt to play into one of those. Because right now I do see them, and I love Sequoia and respect them intensely, but actually in the middle, which I think is quite a hard place to be.

Rory O'Driscoll

It's an interesting comment. They are more like a manager of managers. You have an early-stage team and a late-stage team. What that means, by the way—and I want to talk about this for a second and then come back to the megafund comment—is that the person on top is no longer even a manager of a venture firm; he's a manager of managers of venture firms.

One of the bigger hardships is being on top of those organizations where you have these structures underneath. It's just a very hard and precarious place. If you're not actually running one of the groups that's putting out the money, if you've allowed your job to become helping other people do things who, in turn, are managing money, you're more removed from being able to eat what you kill, and it just becomes a more precarious position.

Jason Lemkin

Having watched venture firms and law firms and others, I really can't think of something more dysfunctional in many ways than partnerships. One of the reasons partnerships are dysfunctional, and maybe it does tie this in, but maybe we could move on, is that it's almost impossible for performance to tie to economics.

If Harry and I are equal partners, if the 3 of us are equal partners, and I have all the winners, it's all fun the first year. But the second year I've got the winners, and then Harry made that big bet and it blew up on us.

And we're friends, but our carry is equal, and then we're raising another fund and we have to argue over carry in the next fund. My limited experience is in partnerships: the daggers are always out. I love your kumbaya view, Rory, but I haven't seen it in the real world yet.

Rory O'Driscoll

I didn't say the daggers weren't out. Because seriously, let's talk about this. In the end, economic performance drives change. One of the things about Sequoia you have to admire is that they didn't say, “We've made this decision. Let's stick with it for 5 or 6 years.” They said, “It's not working. Let's make a change.” It's very Sequoia. It's very on-brand.

2. Burry Shorts AI Leaders

Harry Stebbings

Listen, I want to discuss Michael Burry, famed for The Big Short, pulling another big short: a $1.1 billion short on NVIDIA and Palantir. It had some pretty significant ramifications on the market. How did we think about Michael Burry and the subsequent downfall that it caused?

Rory O'Driscoll

Zooming out, you look at the AI CapEx spend and you go, “At some point this is going to overshoot, then there's going to be a downturn.” Uncontroversial statement. Even Sam Altman would say it, right? I did the numbers. What I actually did was, I decided in this case, since I've been opining on AI CapEx, to say to myself, “What would it look like to make Michael Burry's bet?”

Let's make it real here and actually look at the numbers. NVIDIA stock on Monday was $188. If you wanted to buy puts at $180, which means these things have no value unless your stock is below $180, and let's just say you buy December of this year, so you've got 47 days for that stock to go down, for every $9 you bet, you make 2x your money if the stock goes to $160. You make 8x your money if the stock goes all the way down to $100. In other words, it almost halves.

If you're buying puts, you're betting that in the next 47 days, you have to have that stock go from $188 to $160 just to make 2x on your money. Remember, if it doesn't go down, you lose it all.

Jason Lemkin

You don't even get a quarter.

Rory O'Driscoll

You lose it all. It's not like a stock. I look into that and I'm going, “Hmm, I've got to get it right to make 2x,” and that's pre-tax, right? I'm saying, “Would I have the guts to do this? I believe the CapEx is overinvested. I believe at some point it'll correct. Would I take that bet?”

Would I put $1 million in where I could get $2 million if NVIDIA goes from $188 to $160 by December 20th, or whatever it is, and I lose it all if it stays above $180? No.

One of the interesting things Michael Burry did was release his SEC filing early. In other words, he didn't have to disclose it until the last day—I think it was 45 days after the end of the quarter. He disclosed it early, and to me that was probably him trying to pile on the bad news. If you're taking a very tightly coupled time bet, where you've only got 47 days to be proven right, you have every incentive to shit-talk the stock to try and move it down.

You're not just a passive investor; you're actively trying to say, “Hey, everybody, look, this is a pile of shit. It's going to go down.” That's what he was doing in a very polite way by giving people the information earlier than he legally had to. As someone pointed out, he filed on the last day every other time, and this time he filed early. So this was someone saying, “Hey, everyone, look, this is my bet.”

The other interesting math is, if instead you decided, “No, I don't think it's going to crash in 47 days, but I think it's going to crash over the next 2 years, so I'm going to buy these long-dated options, LEAPS.” I'm going to buy puts. The stock's at $188, and I want to buy puts at $180. They're going to cost a lot more. They're going to cost, like, $50 to $55 per trade.

Jason Lemkin

Hmm.

Rory O'Driscoll

Now you lose money unless the stock in 2 years gets below $150. To get that same 2x, it has to go below $100. People talk about shorting, and obviously shorting is more risky than buying puts, but it just brings home how hard a business it is to bet against AI CapEx. My big takeaway was that—

Harry Stebbings

I don't understand, as an asset allocator, how you can rationalize that as an economic decision to make. Given what you have to believe for that 2x to be real and risk-adjusted, it's not a good decision to make.

Rory O'Driscoll

Yes, it's a very hard decision to make. I mean, remember, we did see an 80% Nasdaq decline in 2000. If you time it right, you look like a genius, but it's so hard. I agree with you. On average, the return, especially for amateurs like me trying to do options, is just net negative, because it's a zero-sum business.

Unlike equity investing, where there's an intrinsic overall return, for every winner in options, there's a loser. For every idiot like Rory, there's a smart guy on the other side of the table who prices it better. I agree. I struggle to think how most people can make money shorting, which is why, to your point, Harry, in some weird way, it's good that there are some guys like that out there.

Was it me, or was it Jason? You made the point about just keeping the whole system honest. That's an expensive way to be a policeman. You have to bet your own money to police the system.

Harry Stebbings

It's courage.

Jason Lemkin

If you're great at it—well, Rory's point is that it's almost impossible to be great at this. If you are, you get leverage on your investment. It's a great way to get leverage, but you better be really good at it.

Rory O'Driscoll

You have to be right. Jason, the thing that really impressed me when I did this analysis is that you not only have to be right, but you have to be right on timing. I think it's easy to be roughly right.

I'm going to say it here: I think Michael Burry is right. It's very hard to imagine a company trading north of 110 or 120 times revenues, like Palantir, growing at 50% or 60% last quarter, having a great, amazing quarter, and kicking off cash. It's very hard to imagine that, in the next 2 years, it doesn't have a significant correction. It's very hard to imagine that the AI CapEx boom doesn't have a significant correction.

But going from that arm-wavy, bullshit podcast statement to actually being able to make money on it, that's damn hard.

3. AI Revenue Is Here

Harry Stebbings

You talked about the excitement waning around AI and air being let out of that bubble and excitement. That's all predicated around, “Will the revenue show up? Will the revenue show up?” Well, the revenue is showing up. Altman says OpenAI is going to hit $20 billion ARR this year. Anthropic projects $70 billion in ARR by 2028. The revenue's showing up in the billions. Are we not answering our own question? Do you think we're almost being overly negative in asking, “Where's it showing up? Where's it showing up?” when it's already showing?

Rory O'Driscoll

Agreed. You are, and they are increasing their estimates. I don't care if you're changing a 2027 estimate up, but the real fact is, if you're changing a 2025 estimate up as the year goes on, that's an enormously positive signal, and I think Anthropic in particular has been doing that.

So you're right: the revenue is showing up. The growth rates are showing up. Even when you look at—funny, CoreWeave's had a little bump today, but their problem is not lack of demand. Their problem was, “Oh my God, we couldn't get the data center up and running.”

It's very hard to make an intellectual case right now for anything other than there being massive demand for compute and massive revenue traction. What you're left with is saying some version of, “I don't think it'll grow quite as quickly as other people think.” It's all hypothetical. Right here, right now, the revenue has shown up.

Harry Stebbings

Gamma announced last night—I was reading about it at midnight after my AGM—raising $100 million at a $2.1 billion valuation, having hit $100 million in revenue with 50 people. That's $2 million a head.

Jason Lemkin

Yeah, we run SaaStr on Gamma. It's great. It rocks.

Harry Stebbings

So I guess I'm just asking: are we being overly British? Are we looking for a problem that's not there?

Rory O'Driscoll

First of all, I'm definitely not being over-British, and if that's what's happening, it's time to end this show now. Sorry, Harry. No, but I think you're right. There's wisdom in what you're saying.

Cynics sound smart and optimists get rich. This is a great, enormous megatrend. It's the biggest megatrend we've seen, maybe since the early days of the internet. Maybe it's even bigger. It's an enormous, ginormous megatrend, and leaning into it is the only sensible thing to do. Playing against it is dumb as rocks.

The only reason you even have these discussions is that you have to make the second-order comment to make your question useful, because on an overall trend basis, you're of course right: the demand is huge. It's some version of the question, “Are we going to see $80 billion of CapEx next year or $40 billion?” Both of them are still huge. In both cases, the trend is exploding, and in both cases, everything is still amazing.

But one of them is $40 billion more than the other, and that's where you do have to start saying, “Things are enormously great, but are we overextrapolating?” So I don't think it's being negative to simply say, in a hyper-growth company, exactly how much should you lean in, and how much risk should you take.

4. AI Joins The Team

Jason Lemkin

That's all that's going on here. Let's talk about Gamma. I actually did an all-company meeting for Replit last night. They want to end the year at 250. I can tell you what I learned from both.

We use Gamma at SaaStr, and some folks may not have heard of Gamma. People talk about it as an AI PowerPoint, but they miss the point. I'll tell you how we use it. We use it now instead of sending the same dated prospectus when we have to close $8 million in sponsorships at SaaStr a year to keep the lights on. Before Gamma, it was the same crappy thing.

Now Gamma automatically pulls all of our data from Salesforce and our marketing automation system. If they've been before, it knows the exact number of leads and ROI from the calculation. It knows who their competitors and similar companies are. It makes a fully dynamic piece of collateral for them in about 10 minutes.

It's actually a great deal. We spend $100 a month for Gamma. A couple of ways to think about that: it ain't much for what I just described, is it? But it's $1,200 a year. How much do we spend for Google Slides? Zero. It's built in. How much do we spend on PowerPoint? I don't even know where my key is for Microsoft Office.

It's a stealth TAM expansion. We're spending $1,200 a year on PowerPoint, but we are because of Gamma. I'm the biggest superfan. It's all over SaaStr because you can do epic things that we would have to wait three weeks for a marketing ops team to do, and they would do a crappy job of it. Now we do it in 10 minutes.

If they keep going, it's a billion-dollar ARR business. 20× revenue doesn't sound expensive compared to some of the deals we've done, does it? It doesn't sound expensive. If it goes from 1 to 100 in 11 months at 20× revenue, it sounds cheap.

Harry Stebbings

And profitable.

Jason Lemkin

Yeah. Look, there are some meta issues. I think we're going to see a lot of these folks adding revenue teams, adding sales teams, adding marketers. Never at the ratio of 2021. We'll never see those levels of staffing of humans, but more power to Gamma if it gets to a billion with no sales team.

Knowing a few folks on the team, the B2B use case I'm describing is pretty small today. As that blows up, they're going to add a whole GTM team. They're going to need 100 people to service it. But I think we are underestimating it.

For what it's worth, the other thing I did was this presentation. I've been vibe coding for 126 days. It feels like a lot of change. So I did this presentation at an all-hands at Replit: “125 days and 10 apps.” I launched 10 apps in 125 days without an engineer. I shared the data, most of the stuff that went well, and a few areas for improvement. I learned a lot of things, including that the engineering team there is really, really, really good.

It occurred to me in the middle of this, while I was talking, that the Replit agent—in terms of all the agents we use, we use about 20 agents, 12 real ones—the Replit v3 agent is the first one that is literally part of our team. It's not making us more efficient like our SDRs and BDRs. Replit is part of our team.

It now has an essentially infinite context window. It remembers everything I've done for the last month with it. We talk about it, and we talk about our mistakes. So I'm doing the presentation, and I'm like, “I have this new idea. I want to build a page that spotlights all the AI apps we've spotlighted on SaaStr. I want to rank them, and I want to do links and everything so you can go discover them,” all the ones we talk about.

I fired up Replit. I'm like, “Here's my idea.” And it just remembers: “Yeah, that's what we did on the other one. That's how we're going to do it this one.” In 15 minutes, we're in production. Like Copilot, the lame thing about Copilots is that they were just tools. Tools are great, but when the AI is part of your team—for real, not VC talk—the amount of revenue that's accessible is so high.

If you go to our office, we have little signs. We make fun of them. We have Repli for Replit, Arty for Artisan, which is an SDR, and Quali for Qualified. You can see how clever we are in our nicknames, right? Arty, Repli, Quali. They all have these little desks where there's no human at them anymore. It's kind of weird.

But Replit v3, this latest agent, is the first one that jumped the line from being one of the tools for our team to literally being part of our team, just like a human being. That is going to unlock so much revenue. So much revenue. It's just starting to be capable now.

When Gamma is part of your marketing team rather than a marketing tool, there's a lot of revenue expansion if they can pull it off. We have literally just gotten going because these agents are so much better than they were 90 days ago. If you're not doing it, it's hard to see how much better it is. I think people miss it because they're not doing it, and they're missing when we cross the line to where AI is part of your team.

That's what's coming in 2026. We're missing this. The agent was the story of 2025. The Copilot was the 2024 story. It didn't work. It was a rip-off—spend $30 more a month on Office. No one wanted the rip-off.

This year, OpenAI and Claude finally actually got good. That's why Lovable, Replit, and Gamma exploded. Gamma was founded in 2020. It had no revenue before this year. Replit was founded, like, in the 1800s. It had no revenue until this year. Vercel had no revenue until this year, and they exploded.

That was this year: AI works, right? Next year is AI as part of your team. It's not replacing folks or layoffs. It is literally embedded in your team. I talked about it, but now I see it. That's where we should be investing, as VCs, as humans, as leaders: what happens when AI is good enough to be part of my team?

Harry Stebbings

And you distinguished that. I want to go down this because I'm actually here to learn, too. You distinguished that from just having an individual agent. What's the difference between Replit as an agent, or Arty as an agent, versus being part of your team?

Jason Lemkin

It is sufficiently autonomous, knowledgeable, and powerful to complete material, high-value tasks on its own with some daily discussions, just like on our team, where you have to check in. With some folks, we can check in just once a month—that's enough, right? But it does need some oversight and some discussion, like a human does.

The level of autonomy and capability is: “Gamma, go out. We've got 20 sales calls this week. Gamma, go into my Google Calendar, create prospectuses and sales collateral for all of them. Pull all the data on them from last year from Salesforce, HubSpot, and Marketo. Put them all together, review them once, and then distribute them to the team. And, if you can, join the meeting.”

When Gamma can do that, Gamma's part of our team. It's not just, “Make me a PowerPoint.” It's not just, “Oh, Gamma's the AI PowerPoint.” That's not so interesting, right? Microsoft will figure that out. Canva actually has a Gamma clone now that's not bad.

A lot has changed. AI is so fast. When Cliff was on the show, I accidentally was a little triggering on Gamma. I didn't even really think of them as competitors. But that was the only time Cliff got a little thoughtful. I mean, he was always thoughtful. But now their version of Gamma is pretty good. That wasn't even that long ago, was it?

Harry Stebbings

Yes. Which is why Gamma's got to keep swimming and add all that other functionality.

Jason Lemkin

That also goes back to the beginning of Sequoia. That's why investing is so stressful today.

Harry Stebbings

I think that's the real answer. I'm glad you said that, Jason, because I'm sitting here thinking about it. We were talking last weekend in a partnership, and technically what you really need to understand is what the improvements in the last 12 months—and I think this is what you're saying—and the changes in the model in the next 12 months mean in terms of what can be done that wasn't doable even 12 months ago.

That requires a quantum of time to just get your head around it, right? It is so stressful because you have to make finding that time a priority to know where it's going. The pace of evolution is so fast. If you decide, “Well, what I knew six months ago is still useful,” you're probably going to be wrong very quickly, right?

That's what I find the most stressful about right now: making sure you actually know where the technology is right now versus your opinion 12 months ago. You might look back on what people said 12 or 24 months ago, and they're laughably wrong, right? The whole Copilot thing—I mean, that was like, “Thank you for sharing, but no.” That's just so done.

Trying to have clarity on the next 12 months is necessary just to be able to play. Without it, you're just betting blind, and you're going to get it wrong.

5. Clones Arrive Overnight

Jason Lemkin

There's that. What I think is even more stressful for seed—maybe by B it's a positive—

Harry Stebbings

Yeah.

Jason Lemkin

Maybe Andreessen investing in Gamma at $100 million was the smartest play of all. Or at Replit, right? At later stages—well, they did early, too. I think the quality of clones is only going up.

When we all started in this industry, you'd laugh. You'd be like, “Well, Salesforce...” Or forget it, HubSpot. Pick whoever you want, whatever leader. It would take them like a year and a half to decide if it's worth cloning. Then they'd launch something, and it would be okay because they have smart engineers, but it wouldn't actually do anything for the first 6 months. It would be so feature-poor.

Then after 2 years, they'd decide, “Well, should I put 100 people on this, or should I put 100 people on the—

I'm not kidding. I can think of 1 investment I've made that has had 5 clones in the first 30 days, including 1 from a cloud leader. The same thing will be true: will they sustain it, or will it just be a feature? But the ability of AI to enable us to clone better stuff faster, and the fact that Canva is borderline competitive with Gamma and wasn't when Cliff was on the show, just disrupts what the hell seed investing means. When anything, even with progress, you might see 10 better versions in 30 days.

Harry Stebbings

What does that mean you do, then? I'm a student of this business.

Rory O'Driscoll

Yeah. What do you do?

Jason Lemkin

For seed? I don't know. We should ask the new Benchmark guy who's smarter than me. But for me, I think the answer is the old one. I'm worried it won't hold. Innovation plus the best founders get there. You've got to still bet on the best founders. You just can't take that early first-month explosion as seriously as you used to. It's not as defensible, that innovation. But if you have the best founders—

Rory O'Driscoll

But—

Jason Lemkin

What else are you going to bet on?

Rory O'Driscoll

The interesting question is, to your point on doing Gamma at $2 billion, do you think that same statement is true? Fast-forward: you're now Gamma, you're doing $100 million in ARR. Do you think the next clone who starts at zero can catch up? Or do you think, over time, the distribution moat, the market-brand-leader moat—do you think some kind of moats accrue over time with scale? Or do you think everything's up for grabs all the time?

Jason Lemkin

No, what I think is that there is a plane of stability that is later than it used to be, but is still fragile. I'll give you an example. At Replit, let's say they go from 1 to 250 this year. It's going to be less than Lovable. We'll stipulate it's not as good as Lovable, okay?

Here's what has happened in Replit: most of the competition can't build the AI agent they can. Even Bolt, which was the early leader, is now just at number 3. They don't even have an AI agent anymore; they've outsourced it to Claude. These products that are really, really good are building a deep layer of sophistication.

But if you don't get there fast enough—and we can argue whether Gamma is there or not—if it's sophisticated enough... If anybody on this hasn't used Gamma, I tell everyone to use it. Go into a Google Doc, write 10 points about this show, and then just give it to Gamma and say, “Make me an amazing deck.” Your jaw will drop at what it does when it's coding this deck in real time. Just give it 10 bullet points. It does so much.

But you sure better be working that 996, right? Because Cliff's got a few good folks. It used to take Cliff 3 years. Now it takes Cliff 90 days. That's why I think $100 million, $250 million—these Gammas start to build a moat. I do believe Replit has a moat. Maybe not versus Lovable, but versus all the rest. But, man, you used to get a moat earlier, didn't you?

Harry Stebbings

But that's where I think vertical specialization does accrue benefits with scale. An example is Solve Intelligence, which is AI for patent law. The more patents that go through their algorithms, the better they are at writing, editing, and predicting. It's a very specific use case that gets better with more and more data ingested. You don't get that with horizontal products.

Jason Lemkin

No, and data is going to be defensible in the age of AI. Data is going to be defensible. But those patents are public, right? They can still be ingested by other people, can't they?

Rory O'Driscoll

Yes, they can. I don't know if it's true that you don't get it with horizontal products. I'm thinking aloud. I think you probably do. Take Cursor as a horizontal product. Would you fund another company now from scratch to do exactly the same thing? I think not. I think there does come a point when you do pull away.

I don't think it will remain unstable forever. I don't think you'll have these $200 million–$300 million outcomes and then someone else doing roughly the same thing come and displace them. I think there's this uncertainty period at the start, but I could be wrong. I'm processing in real time here.

6. Defensibility Moves Later

Jason Lemkin

Well, I think that Harry's point about the patent one—let's step away for a minute—

Rory O'Driscoll

Yeah, yeah.

Jason Lemkin

What I think is interesting is the classic question in B2B, maybe all venture, but certainly B2B venture since we all started: how important is it for something to be defensible in the early days? We've debated this for years, since the inception. We've all known deep down that no product that can be built in 60 days can be all that defensible. But we told ourselves the team had domain expertise or this or that.

The question is, today, has the bar gone up? Should we either give up on defensibility for seed investing—just give up that it is a criterion—or should we radically raise the bar, forcing us to go into verticals, corners of the market, areas where there aren't 11 or 100 agents already?

Harry tweeted about support the other day. I can tell you, when I invested in support in the early days, Talkdesk, Gorgias, and Front—no one wanted to do support in the early days. Everyone thought this was the dumbest category. So go find something. Now it's trendy. Don't do that, right? Maybe go find something the cool kids aren't in.

Rory O'Driscoll

I think it's simple. I don't think you can have a major defensibility moat in any of these horizontal or vertical markets at the seed stage, or even, frankly, at the stage we're investing at. The defensibility theorem emerges at scale.

In other words, I do believe what's true in most enterprise businesses is that once you become the anointed winner, once a market coalesces and there are 2 or 3 people, at that point in time, it's yours to lose. You can still screw it up. But provided you have great engineering and stay on top of the trends, on top of the technology, you should be okay.

I think the idea that at the seed stage you're going to find a defensible way to do code generation or code testing and so forth is absurd. You just have to internalize the game you're in, which is that, for most of these deals, unlike deep tech, you're going to have to have an awesome team, run fast, be superlative on technology, and get your distribution early. Then, as you scale up, you become the winner. You can't be anointed the winner up front. Get over it, everybody. It's a high-risk game.

Jason Lemkin

But is that okay at 50 post for a seed round or a pre-seed round? Do the outcomes justify it? That's the thing. Sure, Rory, if the deal's at 3 post or 5 post and I can spread my $500K checks around, I get it. If I've got to spread $5 million checks around at 50 post, it's tougher.

Rory O'Driscoll

That's a much better question, because now we've gone from the abstract of whether it's defensible to the actual nuts and bolts of money. We've recognized that the game we're playing has more variance than we thought the last time, has to run fast, and probably also has more competitors.

Are you getting paid for the risk? I wrestle with the same thing around later-stage companies at $100 million or $200 million pre-money. We have this constant dialogue. If I look at it, we do As and Bs. In the early product-market-fit A, you probably still don't know who the winner is. And in the B, when you know who the winner is, it's going to cost you a fortune. Which of those is the better bet?

Harry Stebbings

But do you think at the B you even know the winner? We look at the B—you mentioned my customer support tweet. Most of those companies had raised Bs. I have no idea who the winner is in that category, and I don't think anyone does.

Rory O'Driscoll

No, I'm going to push on that. I think somewhere between the A and B, you can know a lot more than you know at the seed or the early A.

Harry Stebbings

Do you know when the Bs are preemptive at $3 million to $4 million of ARR, which they are for hot companies today?

Rory O'Driscoll

It's interesting, because this is absolutely what we have to know to do our business, right? At C, Jason can't know. He can just believe, right? But somewhere between the A and the C, you have to know, otherwise we're all ludicrously overpaying.

If you look at all these markets—for example, you mentioned your patent company—I think, without naming it, we have a rough sense of where all the companies are. I think your company is doing very well, Harry. You should be glad to know. That doesn't mean I know where the market is going to end up 10 years from now.

Once the horses are running, and once they round the first furlong, you can actually see the rank order of where they're running in a way you can't at the early A. At $2 million or $3 million in revenue, you're drawing on small pieces of information, but you can see rate of change, and new differences emerge pretty quickly. So I disagree. I think you can have a pretty good idea. Look, it's hard, but—

Harry Stebbings

I genuinely don't want to be rude, but can you give me an example of where you think at the B you have had a clear understanding of a winner?

Rory O'Driscoll

Take code gen. There were a bunch of people doing it. There were 10 to 15 companies at the A or earlier. I think Cursor emerged, I think Windsurf emerged, and then, to a different extent, Cognition. By the B, it was obvious that those were the names.

Harry Stebbings

But I would push back on you there and say that I don't think we know a winner.

I think Codex is making incredible ground. I think Claude Code is making incredible ground. Cursor is very good. But then, as you said, you've got Cognition, you've got Replit and Lovable kind of coming from the prosumer, less developer-centric side. I think that's still an entirely up-for-grabs market.

Jason Lemkin

I think, looking back—and nothing but kudos to the team, right?—when Windsurf sold, it wasn't clear it had a sustainable moat of any sort. It wasn't clear it truly had the staying power. It was a darling of a slightly more enterprise version of Cursor. People did love it, but ultimately it wasn't clear it was a winner then. It wasn't clear that brand was enduring. Looking back on it, it wasn't. We don't know if it would even have survived as a standalone company.

Rory O'Driscoll

I think that's true. But at the same time, what was clear over the past 6 months was that, if anyone was going to be the perceived company worth acquiring, there were only 2 or 3 names in that space, and the other 7 or 8 names that had been around weren't. So I understand what you're saying. The reason they took the deal is that it wasn't clear they could go from where they were to a billion in revenue and an IPO.

But my point is, going back, we're dealing with probabilities here. At the seed, you know nothing but the people. At the early A, when there's 5 or 6 of these companies doing $1 million, plus or minus, in revenue, all you know is you've got a 1-in-10 shot. Somewhere at the B, I think you get down to being able to say it's a 1-in-3 shot. It's still a 1-in-3 shot with a huge amount of variability, but the odds have narrowed.

Let's take— I mean, you guys keep talking about Lovable and Replit, right? You would say, implicit in what you're saying, that those are the 2 names. You implicitly said, "I don't know the space as well, but Bolt has shot its bolt," as it were—pun intended. That's a piece of information you have. When you're making a bet at this stage, you can probably say something like, "There's 2 clear winners here. Probably less likely to be another raw startup. There is adjacent competition from, you know, Wix has bought someone." So you at least—

Harry Stebbings

I'm so sorry to interrupt you again. I would just say you need to expand it significantly. It's not 1 in 3. You've got Vercel and Claude Code eating their lunch coming down. You've got, as we mentioned, Replit and Bolt, but then you've got Salesforce, which has its competitor, and Atlassian has its competitor. Figma Make is doing very well. It's 1 in many.

Rory O'Driscoll

You're right about the adjacent competitors. Let's talk about that, because I always think there's a 2-step horse race in all these deals, right? And, by the way, I'm going to argue that what you just said is proof of success. Let me tell you what I mean by that. In any startup, you start off at day 1 and you're like, "There's 3 other startups doing just what we're doing—or maybe 10 other startups doing just what we're doing. I wonder which of us will win."

My experience is, when you go to the first board meeting where you suddenly realize you're scared of the big-company adjacent competition, it probably means you've graduated from the baby class. You're 1 of the 2 or 3 winners in the startup land, and now you've got to worry about the adjacent guy next door.

You might say you're not sure if your Lovable bet is going to be the winner yet, but do you believe that, within the class of venture bets on this space, you've got 1 of the 1 or 2 winners?

Harry Stebbings

Yes, but that doesn't generate enterprise—

Rory O'Driscoll

Yeah, it does—

Harry Stebbings

But that means shit. It doesn't matter if it's 1 of the venture bets. If 1 of the venture bets doesn't win and Salesforce or Atlassian or Canva does, I don't care.

Rory O'Driscoll

Obviously, you're correct, but you've got to think about it as incremental information updating your priors. When someone did the seed, they were like, "This is a good idea. It might not even work." Then you do the, "Oh my God, vibe coding is a thing. There's 5 companies doing it. We're 1 of the 5."

Now you can say, "Vibe coding is a thing. We're 1 of the 5 doing it, and we're 1 of the 2 winners." That's a huge amount of risk reduction. Now, you still have the other risk, which is that the big companies might do it. Something might disrupt you. There's a lot of risk still left, because it turns out that startups are risky.

But you've got to admit that there's been a massive amount of information gleaned and risk reduced, going back to what Jason said, from the 5-on-50 bet, where you don't even know if it's a space and you don't even know if you're going to be a viable player in it. Is that a better or worse risk than doing 2 billion pre when you know both those things? You are the winner. There is a space, you are the winner, and there's still a whole ton of competition to come, which is the Lovable bet you made. Which of those 2 bets are riskier? I don't know.

Harry Stebbings

No, I think about this, which is, at about 4 to 5 million in revenue—it was done at 200 when we first did it, and then when it was about 80 to 100 in revenue, it was done at 2 billion. You can choose your entry price.

Rory O'Driscoll

Totally. What that says to me is that consensus, rightly or wrongly, has said there has been a massive risk reduction. And now, unfortunately, what's happened—and always happens in a bull market—is that valuation has expanded to fill the risk that was reduced operationally.

In other words, at 4 to 5 million, there's still a ton of risk. Fast-forward to when you're doing 50, 150 million, I just have to say it: You can't deny there's been a huge amount of operational risk reduced. Now, the problem is, at 2 billion pre, is there enough upside left in the deal? Separate question.

The point I'm making is it's gone from being, "Is this a category? Are they the winner?" to the third and last question, which is always, "Is the TAM big enough to support a $2 billion valuation?"

Harry Stebbings

Which I think, in this case, it absolutely is. I just want to go back to the fact that we kind of know the winner at the B, and if we don't, and if that certainty has reduced or gotten less, are we overpaying? That was an interesting addition. My statement is, we dramatically know less, and that 1-in-3 has moved to 1 in 7 to 10 significantly across categories. In which case, surely the suggestion is we are dramatically overpaying.

Rory O'Driscoll

There's some level of truth in what you're saying, but I think the variance in these companies is a lot greater than in the 15 boring years of SaaS, where it's obvious and you know what to do for the next 10 or 15 years.

Jason and I would both say we both had a play—Jason as a leader and a founder, and me just as a humble investor—in an e-signature company. When we did those deals in 2008 and 2009, it was e-signature. Fast-forward 10 years, and it's still e-signature. That was a simple world.

We would both agree, I think, that there's way more change in this market in 1 year than in some of those older markets in 5 years. So you're right, there's more risk, Harry. But I still do believe all these investments are riskier than they were in SaaS land, and they're astonishingly priced higher.

I still think there is some significant risk reduction in going from a 1-in-5, 1-in-10, "Will it even work?" to a 1-in-3: "I know who the competitors are."

7. Diversifying Seed Bets

Harry Stebbings

If that's the case, should our B portfolios be more diversified? Jason put it as a great question: Do seeds need to be diversified?

Jason Lemkin

Well, how big a fund do you do, right? Or if seed checks are 5 million, how big a seed fund do you need just to make the minimum diversification work? How big a seed fund do you need with a $5 million seed check?

Harry Stebbings

Well, it depends if you think outcome sizes are expanding with the movement from technology to bank—

Jason Lemkin

Does it matter? It's just some basic math. How many first checks do you want to make—how many do you need to make in that fund for it to work? So if you traditionally needed 20 or 30 checks to work, but risk has gone up, to Harry's point, maybe you need 40. And then there's what reserves you need.

You might need a $500 million seed fund to have sufficient reserves, because I need to do 40 deals at 5 million now. That's 200 million. 200 million for reserves, that's 400 million. 100 million for fees and time and backup. I need at least 500 million for my little seed fund to make the math work.

Harry Stebbings

What you're missing on the maths is, if you think that the outcomes are going to expand, you can have smaller ownership on entry, and so you don't need to increase check size if you assume you're going to have—

Jason Lemkin

4 million out of 50. Does it really matter? These are not massive owners. 1 million on 50 isn't going to work, Harry, in my seed fund, is it? It's not going to be enough ownership, is it?

David Frankel

I'd just try to disaggregate Harry's response, because there was something there. First of all, I agree with Jason's framing. Your response was interesting, because you could have said 1 of 2 things. You could have said, "You don't need to go from 20 deals to 40 because the winners are so much bigger that even if you have fewer winners, you're fine." That would have been 1.

But you didn't say that. Interestingly, you said, implicitly, "Go to 40, but just take less ownership." The mere fact that you made that answer says you are embracing a more diversified story in the face of risk, which is Math 101.

Rory O'Driscoll

Which is a super interesting concept, because we had it down on the agenda to talk about. Some folks are even pulling off these highly focused seed-stage bets, which I find awe-inspiring. All other things being equal, with the amount of variability you're seeing, I would've expected people's deal count to have to creep up slightly.

We had Roger on, who was very much a concentrated bettor. You have the Hummingbird story in Europe, which is astonishing and impressive. Everything in logic says to me that, with the increase in time to exit and the concomitant increase in risk, logically you should be increasing your diversification slightly, probably reasonably. Which probably means either smaller checks and more deals or a bigger fund size to maintain the same ownership.

Fabrice Grinda

Well, I think both work. One fund I'm an LP in has 100 to 150 positions with $100,000 to $150,000 checks. It's a 7X fund. It goes to my point about outcome sizes being so much bigger, and therefore the ability to have lower ownership. Jason, to your point, $1 million on $50 million does work if it's a $100 billion company, not the $3 billion to $5 billion enterprise outcomes that we've been playing with for the last 10 years.

Jason Lemkin

Yeah. It works even better if you own 10% of that company, though.

Rory O'Driscoll

Would you want that? Yes.

Jason Lemkin

The thing with the $100 million to $200 million fund is, I'm not that human. I don't want to meet 500 founders a year. To do 200 deals between your team, you have to meet 500, 600, 700 founders a year, even with your AI agent helping you. I can barely tolerate doing a couple of meetings a week. I can do stuff by email, but I have to carve out an hour for a deal I might not do? I want to blow my brains out after that meeting.

I sold my companies so I wouldn't have to do those meetings anymore, right? I'll do 1 a week.

Harry Stebbings

I can tell you our partnership does 20 meetings per partner per week. With my 4 investing partners, we have 80 net-new companies that we meet in person per week.

Jason Lemkin

80 in person?

Fabrice Grinda

Yeah.

Rory O'Driscoll

Yeah.

Jason Lemkin

I would resign. I would give you all my carry back. Thank you for hiring me, Harry. I'm eternally part of the 20VC team. You make me do 20 in-person meetings? I sold my companies because I didn't want to spend my life in meetings. I'll do 1 a week.

Harry Stebbings

You do the math. We'll do over 3,500 company meetings a year.

Jason Lemkin

Honestly, I think it's great for the LPs and others. I think that is a great playbook. It's just, not to get distracted, you have to match the strategy to who you are as an investor: where you get your leads, where you get your deals, what your brand is. You can do that for a variety of reasons, but it's a lot of meetings, man.

Fabrice Grinda

It's a lot of meetings, 100%.

Jason Lemkin

It's a lot of meetings. Can I get you guys a coffee or a drink? Do you want sparkling or still? Which one would you like? Do you have DVI on the Mac? Here, sit here. Harry's running late. Harry's doing the 20VC of sports. He's running a little late. He'll be here soon.

Rory O'Driscoll

No, I love taking meetings. That's how I learn. I'm a meeting junkie. My partners laugh at me. I'll take a meeting with anything, right? Because you can always learn something from it. So I have a bias to meet—in fact, probably an over-bias.

Jason Lemkin

Listen, when I meet a truly great founder—truly great—I always learn a lot. Don't get me wrong. Anyone below that, I don't think I learn enough to be worth the time.

I do my homework. Rory, you're the best homeworker on the team. You could do so much homework on an investment. Harry's the most prepared and the most charismatic, but Rory's the best homeworker, okay?

If they're not great, do you really learn anything in that meeting if you've spent an hour researching the company? I learn nothing. I read your deck, and then I say, "Send me more." Then I say, "Send me your last 5 investor updates." I say, "Send me your financials, and I'll research you on the internet."

Now we have AI. Claude will help me. I'm going to know a lot of stuff unless you're really great. If you're great, you're going to blow my mind, right? But if you're not, I'm going to start yawning about 15 minutes into this meeting.

Rory O'Driscoll

I don't know. I find sometimes you get an insight. I think—and I could be wrong—the beauty of doing this is you make me think, "Should I do things differently?" because you do it so differently.

I do find that even on an okay deal, you learn nuances about a specific market from the one-on-one and the dialogue back and forth that you wouldn't get from the presentation. Now, it does mean I have, as people who've pitched me know, a horrible interrupt-driven style whereby, if you have 20 slides, I'll be like, "We can skip 17 of them. These 3 I care about." It can be annoying at times.

I think in any business, no matter how much you think you know from the outside, the person inside living it every day has a crucial kernel of knowledge that you just can't access any other way, and I believe I get something from most meetings. It's a little like your knowledge of AI agents versus anyone using those words: it's step-function different.

Jason Lemkin

It is. But you have to meet me to learn that. If you meet all the other ones, it's a waste of your time.

Rory O'Driscoll

Yep. Agreed, but you have to meet them to get that experience. I buy into meetings. We've come a long way from fund construction, but that's okay.

Fabrice Grinda

Yeah. This is fantastic. This is what founders like. The number of founders who don't get to hear this, who actually wonder how VCs think, is in the hundreds of thousands.

Rory O'Driscoll

Yeah.

Jason Lemkin

Well, it's related to it. The question a lot of founders have is: How important is it just to get in the room? Just get my foot into the door. Just get the coffee meeting.

I think it varies. I think Rory's saying, "Listen, get in the room at scale"—there's some value to that. I'm saying, "I have no interest. Don't get in the door with me. Just send me a great deck and a great email. I will read it. I will slow down. I will spend time. If it looks good, I will take the meeting, but there's no need to get in the door with me."

There's no value in getting to know me over a coffee meeting. They're like, "Well, just take the meeting, Jason." I'm like, "Dude, I have so much respect for you, but there's a 0% chance I'm going to invest. Don't try to get the meeting. It's not going to help you."

Rory O'Driscoll

I think that's a fair comment. I don't think someone who's doing something I'm totally not going to do should get a meeting. I'm not a huge fan of coffee meetings. I'm like, "You're here to get money. I'm here to give money. Can we just talk about the business rather than a bunch of getting-to-know-you?" But I am always interested in hearing people talk about their business.

Jason Lemkin

That's why you're enduring in the business. That's why you're enduring: because you enjoy the meetings.

8. Running The Fundraising Process

Rory, how do you feel when you hear people say, "No, no, no. I'm not going to meet. I'm waiting to run a process, and I'll run a process on November 19th. I'll email you then"?

For me, I found this really abrasive. I'm just opening up here. I said to the founder, "Listen, if you're running a process, you're either optimizing for price or partner selection. I'm giving you a great price today, a price that you want—a price that you said you wanted—which means that you're not optimizing for price with this process, because I've given you what you said you want. You're just saying that you think you can get better than me.

"That may be the case, in which case, just tell me straight that you think you want Sequoia or Benchmark. There's no harm, no foul, but fine." And they're like, "No, no, no. I just want to run the process." How do you react to, "I want to run the process"?

Rory O'Driscoll

I think from their side they're correct. I don't have to like it, but I think from their side—because I see more failed financings because they didn't run a process than because they did.

You're basically saying, "Mr. Founder, you don't have to run a process because I'm going to give you a term sheet right now at a good price, and you like me. Let's do it." That's not an unreasonable offer.

But typically, what happens when people, quote, "don't run a process" is that someone comes in and says, "I'm really interested," and they share information serially with someone who's not yet ready to commit on the investor side. So they've run an accidental process. That's a mistake.

Jason Lemkin

But if I'm saying, "Mr. Founder, here's your term sheet. There's nothing to be done"—

Rory O'Driscoll

That's different. At that point, it's not crazy. If I was on the board of a company where that happened, I would take it seriously and think, "Should I hit the bid?"

Conversely, if I was on the board of a company where the founder said, "Hey, I'm meeting with Joe at a mega-firm. They asked if we could share some data. They're not—they just want to get a sense of it," I would shut that down and say, "You share with everyone or you share with no one."

Giving your data to 1 or 2 people when they're not in is just starting a process without meaning to start a process. If they don't move forward, you've had a failed process already without ever doing a process. I'm 100% certain that's a mistake.

Even though, on my side of the table, again, I don't love that, you're raising something, which is why a lot of times we wrestle as investors with—this is my aha—"I want to be able to do what Harry just said."

I wanna be able to come in and commit because it's the only way to get them off that process.” They have to have a good enough relationship with them, they wanna do business with you, and on top of that, you have to bid with probably sparse information.

My big aha on that is the only way you can conceivably do that. No amount of pre-work can do it because you don't know the actual information. If you've seen the prior round, it's probably your best chance because if you've seen the prior round, you have some sense of what's going on. You know what they underwrote, and you had a mental model at the last round. You can probably, in a dialogue, get 1 or 2 pieces of information, calibrate how they're doing, and maybe do that process.

Jason Lemkin

I have a slightly nuanced view in the middle, for what it's worth, just maybe for advice to folks that watch, not to Harry. But when that happens, as Harry said, I think it's a slight founder fail, a slight own goal. Not a total one, because obviously he's got a good company, right? Harry is, in essence, ready to do the deal now. The founder says, “I wanna run the process.”

What many of the best founders do—not all; there are all different types of founders: extroverts, introverts, great fundraisers—but what many of the best do is they're able to cultivate enough interest with enough good VCs that if they hit the number, they just send an email. They just send an email: “Harry, I'm thinking about raising a round before the end of the year.” Harry can say, “I'll give you a term sheet today.”

The right answer is, “I love you, Harry, especially love the one with Rory, but I'm not ready today. Honestly, I'm not ready today. I will be ready at the end of the year.” That is the perfect way to handle this situation. You don't risk losing Harry's term sheet. Founders do overplay their hand—not as often as you might think, but they do overplay their hand. They can say the wrong things.

The best founders, one way or another, build relationships over months. They copy them on their investor updates. They update them. Then 3 or 4 folks are just in when you're ready. It's so casual with so many founders. They're just in: “As long as the deal is reasonably fair, just tell me where to write the check and how much I can buy.”

I think the reason this happens—the Harry scenario—is so many seed investors and so many accelerators hammer into founders that you have to run a process. That is the classic top 3 bits of advice. I think it misses some nuance on the optimal way to run it. The optimal way to run it is for everyone already to want to invest for real, without games, before you open your data room.

The super-optimal way doesn't even require a data room. You don't even need a data room because they already wanna invest. They need diligence. The best-run processes—I know this might be slightly controversial out of context—don't require a data room. Not a traditional one. They only require a file that says “Box-diligence investment 12/21/25.” They don't need any other data room. It's just for diligence.

Rory O'Driscoll

I'm gonna rephrase that. I think what you're actually saying is the best-run processes don't feel like a process, but they are.

I think you're exactly right: if a founder is smartly nurturing relationships, keeping people broadly informed, but then tries to time the interest such that when he's ready to put his or her hand up, there are 3 people who are primed and ready to go, that is the best outcome, as they say. So it is a process, but it doesn't feel like one, and that is perfection itself. I agree with you.

Now, you have to have a very attractive and high-performing company to be able to do that, but—

Jason Lemkin

Maybe, but I also think those are the only ones getting funded in this environment. We talk about all the Gammas and the Schmamas, but everyone below that ain't getting funded anyway, so you might as well run this version of the process.

Rory O'Driscoll

Yes.

Jason Lemkin

Maybe Harry thinks a lot of B folks are getting funded today, but I ain't seeing it. It's the most binary fundraising environment in our lifetimes, isn't it?

Harry Stebbings

Expand on that, Jason.

Jason Lemkin

You're either YC, Neo, or South Park Commons, you've got something, and you get funded, or who the hell's gonna find you in your pre-seed round, right? You better be whatever the hell AI-native is—I mean, we know what it means. You better be hot AI-native with top-quartile venture growth, or you ain't getting funded. It's pretty simple. There's not a lot of gray zone anymore.

Harry Stebbings

But I do think even with that, you say YC, Neo, fine, but they have a huge number of companies per batch at YC. If you wanna stick to the religious “we're gonna run a process,” that does work only if you have stellar numbers. If you haven't built relationships before and you start on Monday the 19th, to expect that you're gonna come in, hit the ground running with first meetings on Monday the 19th, and get term sheets super freaking fast with average-to-middling numbers—

Jason Lemkin

No, of course. I was making an assumption in the story you told me, the anecdote of the term sheet, that it had top-decile venture numbers. The very fact that they were flippant, I assumed, implied it. Now, if they didn't—

Harry Stebbings

Zero to $1 million ARR in 7 months?

Jason Lemkin

I would take that offer. I would sign the term sheet and send it right back, and ask if he wanted to meet at Selfridges over the holidays to see the Disney exhibit or something, like we're doing with our team trip. I would say, “Thank you, Harry.”

Rory O'Driscoll

Guys, I think you were saying the same thing. Jason's throwaway comment is worth pausing on: he was basically saying you should run this kind of—let's call it the light process—which is very FOMO-driven. I had said, “Hey, you can only do that if you're a good company.” His comment was, “Only the amazing ones are getting funded.”

So implicitly what he's saying is, if there's 100 companies and only 20 of them have these kinds of numbers, they should run that kind of process, and the other 80 are screwed no matter what, so he wouldn't even invest. So 100% of successful deals will be this kind of FOMO-driven, non-process process. That was the implicit statement in what you're saying, Jason, right?

Jason Lemkin

I mean, Harry's right. There's some untruth to it, but I think there's a lot of truth to it: we're in the Captain Obvious era of investing. It only takes 1 term sheet, so, to Harry's point, be careful. But I'm not seeing the non-obvious ones get funded. I just don't see it, right?

The most brutal one is, again, the classic SaaS company—triple, triple, double, double—which has been discussed ad nauseam on every 20VC channel, including 20VC Cricket and 20VC Sales. Those ones you could always find someone to fund if you met enough people back in the day. Now you can do better than that, and only 20% of people wanna take a meeting, and they still might not do it. I think every month that goes by, those deals are harder to do. Every podcast, everything makes them harder. Harry's disagreeing with me?

Harry Stebbings

No, I'm 100% with you. We had a company that went from $400K to $3M ARR, a classic enterprise SaaS business. Bread-and-butter enterprise SaaS that would've had 5 term sheets from 5 good firms, 120 meetings, 1 term sheet, and it was a $10M round on a $40M post, 12X revenue for a 10X grower.

Jason Lemkin

But have you seen the market comps, Harry? Have you seen what the average public company is trading for? That's still a fine valuation.

Rory O'Driscoll

Yeah, but they're not growing 10X, as Harry's point. But the real truth implicitly in that is it's a combination of some element of it's just not fashionable, but also the implicit statement is that growth rate's going to attenuate.

I might have 2 or 3 years further on that same investment. We might have a company doing $25M in revenues; when we invested, it was doing $5M, but its growth rate's now down to 60%. It's still only burning $10M. But those are incredibly hard deals to get funded.

9. Datadog Captures AI Spend

Harry Stebbings

Okay. We're gonna switch tack slightly, but I wanna discuss 2 kind of crazy results in the last few days: 1 good, 1 bad. Datadog absolutely freaking crushed. Stock up 23%, $15 million plus AI native customers. Wow. How do we think about Datadog?

Rory O'Driscoll

Jason has covered this so well last time. His basic comment was, even if you're not AI-first, co-attach to the AI trend and you'll be fine. These guys co-attached to the AI trend and they're fine. Sell shit to the people who are making AI, and if they grow, you'll sell more shit too. And they did it.

Jason Lemkin

No, no, it's true. The irony as we go into next year is that the AI leaders—the hyperscalers and the hyper-this and the hyper-that—they're starting to buy like classic B2B companies. In fact, they're recycling the same people in procurement and the same people in GTM, and they're buying the same stuff.

If you've attached to the AI budget and you're a Datadog, and you're in that, you're actually gonna have a great 2026 because these are normal B2B companies.

Like, OpenAI is buying like a normal company, buying like an Adobe or a Microsoft now. But if you're not in that, man, you're just dead, right? And the other one was Clio, which raised at $5 billion in legal tech.

It attached to AI in a different way. This is a company founded in 2008 that found its way in the AI era, right? It added fintech. Payments got up to $3 billion, now at $5 billion. So find your way, right? Datadog, it helps that Datadog, even though it got expensive, it was the darling, right? It was the darling product. But, man, they capture that revenue. Go find it.

Rory O'Driscoll

They did, because I'm gonna leave Clio out because I think it's just so different. I think Datadog is a core piece of compute infrastructure, and these hyperscalers are the most compute-intensive companies that have ever been known. So if you're selling compute stuff, you should be having, as Jason said, a great quarter.

If you're selling routers, if you're selling switches, if you're selling little interconnects, whatever it takes to stand up Stargate, and observability is a key part of that, you're gonna be golden because there's more compute than you've ever seen, which I think is a very different dynamic than some of the others.

Jason Lemkin

Fair, fair enough. If you're not attaching to that compute, something's off. If you're compute-adjacent, you better be growing quickly if you're compute-adjacent.

Rory O'Driscoll

Yes.

10. Duolingo Faces The AI Test

Harry Stebbings

I'm a Duolingo shareholder, down 25%. Rory, you know, the age-old thing which I just love that you describe me with is, “That's great, but what about me?”

Rory O'Driscoll

Yes, Harry. Harry, that's you.

Harry Stebbings

Duolingo plunges 25%. I'm going, “Why? What the fuck happened to make it plunge 25%?”

Rory O'Driscoll

I don't think there's a mega story here. They're still 80% up on their IPO 4 or 5 years ago. They had a period when it was, “Oh my God, AI's gonna kill them.” Then the CEO very wisely got ahead of that and said, “We're using AI,” so the stock got way ahead of itself.

And now it's like AI's not gonna kill you, nor is it gonna make you enormously rich. You just guide it down for next quarter slightly for business fundamental reasons, and the stock was overvalued and went down. I don't think there's a big story here. It's like the graph is still, as I say, up over 5 years. It's still significantly up on its lows. It just got ahead of itself. They had a good quarter. The revenue guidance was slightly less. Life goes on.

Jason Lemkin

I'd say it has the wrong kind of AI. And what I mean is, Duolingo is using AI to make its product better, right? Hooray. Every single portfolio company better be using AI by this point to make your product better. This is not 2023. You don't get any kudos for sprinkling AI dust on your product.

Going to Rory's point, what you get kudos for is attaching to that compute budget. That's the only kudos you get, and Duolingo didn't earn any.

Rory O'Driscoll

Well, they couldn't because, to be fair, as an apps investor, I'm gonna step up and defend the poor little apps companies, right? If you're an infrastructure company, you can co-attach to compute. If you're a new AI apps company, you are using that compute and using that AI.

If you are like Duolingo and a lot of our companies that have been around since pre-AI, I don't think you're gonna co-attach to the spend, but what you can do is co-adopt the technology. And I'm gonna give Duolingo credit. They've done a decent job of saying, “We're gonna be AI-forward, you're gonna be AI-leaning.”

But in the end, this is my zoom-out point: you're still selling a subscription product to help people, at a very modest level, learn another language. There's no AI compute to attach to there, Jason.

Jason Lemkin

Well, there is. They have to find a way. Listen, I'm not a total Duolingo expert, okay? But Duolingo took money from Berlitz and all these language schools and stuff online, right? Hooray, you did that. Now, where are you gonna disrupt humans? This is your job. You already disrupted those humans, unfortunately. They're gone, right? Where is the next level of human disruption?

Rory O'Driscoll

Zooming out, I will give you that. That is true. I do think the next generation of companies that are going to be, quote, teaching a foreign language will obviously be LLM-based, and there's a lot you can do. We've seen some of those companies, like what I call more professional, more interactive teaching using LLMs.

I do believe there's a whole ton to be done in terms of one-on-one instruction in language and a whole bunch of other things. And Duolingo should be getting on its skates to do more of that versus just using AI to obviate the need for humans, which is the thing the CEO got into a little bit of trouble for.

The real question is, can you build a more compelling set of products using AI to do more immersive learning? And I think it's companies like Speak and companies that we've talked to. I think there's a whole bunch to be done in AI-enabled learning, and it's actually a super interesting space, including language learning.

Harry Stebbings

A really interesting question to ask for all investors, which is, if you are not removing humans from the equation, you are going to be heavily discounted.

Jason Lemkin

Rory's got a good point. You're either getting money from compute, right? This massive spend. Or you're getting money because you're using AI to replace humans. Otherwise, you're not gonna grow. Otherwise, hooray, congratulations on your 14% growth.

Where are you replacing humans for real? Whatever vertical, whatever industry, where are you gonna go in and reduce the head count that vendor needs by half?

Rory O'Driscoll

Most of the time, I'm in the camp that replacing humans is a story. Interestingly enough, in education, I actually think it's doing it better. I think the data, the stunning data, shows that if the human today is someone in a class with 20 people, there's a whole bunch of examples that say LLM learning is equivalent to one-on-one human learning, tutoring one-on-one.

Let me say it more positively. What you should be doing with AI in education is allowing everyone to get something they haven't had, which is one-on-one learning instead of group-based learning, because all the data says one-on-one-based learning constructed specifically to your needs is a far more efficient way of teaching anyone a foreign language, and frankly, most concepts.

Jason Lemkin

I admit I'm not a total expert in education, but where does the budget come from for that software? If it's not replacing humans, then you better steal it from a legacy incumbent. Which is fine, but Duolingo's a legacy platform now. You gotta steal it. You gotta steal or you end up Chegg. At least Harry didn't do Chegg.

Harry Stebbings

I didn't do Chegg.

Jason Lemkin

No, what I mean is, listen. Okay, so let's say your AI does not replace humans or attach to compute. You have a third option. It's captain obvious. Your third option is to use AI to massively displace an incumbent and steal all their revenue.

In fact, that's the history of B2B software mostly, right? It's stealing the revenue. I just don't know how many of our public leaders are in a place to steal their own revenue, right? They're in a tough spot with it, with all their seats, right?

And we can fund those deals as investors, but I think the first 2 categories are much, much easier: attach to the compute or replace humans, rather than just steal. I mean, there's like 400 AI CRM startups out there all saying they're gonna eat HubSpot's and Salesforce's lunch. That's not exciting to me as an investment.

I mean, maybe you will, but that's much less exciting than truly replacing 90% of your GTM team, right?

Harry Stebbings

It's almost like doing customer support to me.

Rory O'Driscoll

Jason, that was a really good framing. The fact that getting the new budget is just so much better than trying to slug it out with the existing provider and say, “Our new thing is better.” It's not to say you can't in those spaces—

Jason Lemkin

I'd have to think—education is an interesting market, right? It's large. But getting incremental budget has to be close to impossible. The public school district's not gonna come up with another $10 million for your software. It's impossible.

Rory O'Driscoll

It's not. Sadly, this is not a public school comment, right? But you're right. I think the positive-sum comment is, I think Duolingo is actually what I call light learning. You're learning another language, but you're just learning a few words.

I think a really interesting space we've seen with LLMs is replacing the spend that an adult wanting to learn a second language, typically for business purposes, would heretofore do with a one-on-one coach, right? So you're replacing that human coach with a very intensive, immersive LLM-enabled learning program, which is actually quite a compelling market.

But you're right: it's education, but you have to find an existing spend. And the existing spend is pretty niche, which is people who can afford to spend money on a one-on-one tutor to learn a foreign language. There's not a whole ton of budget, unfortunately, in K–12 to give every kid a one-on-one customized tutor.

Harry Stebbings

I love the way Jason's just getting better and better with every show.

Rory O'Driscoll

Just get him out.

Harry Stebbings

So could you...

Aren't you, Rory? We're like, “Geez, he was here for the entertainment, and now he's become so wise.”

Rory O'Driscoll

He's getting hard and mean.

Harry Stebbings

He's so wise.

Jason Lemkin

No.

Harry Stebbings

What is going on?

Jason Lemkin

I'm not mean; I just don't want to live in the past. Here's the thing. If you're in software today—B2B software—wherever you are in the org chart or as an investor, honestly, if this isn't the most exciting time of your lifetime, going back to Sarah's conversation, you're doing it wrong. It should be one of the most stressful times of your lifetime, and we've talked about this.

If you're not truly excited—truly excited—I mean, this is the first time software has gotten better since the 3 of us met. It hasn't gotten any better since all of us met. It's the same crap. So, if you're not incredibly excited, again, going back to the beginning of the Sequoia conversation, I would retire. Retire from your VC fund, retire from your company. No shame in that. You had a great run, right? Just put the rest into Nasdaq, and you're going to make more than most VC funds anyway, right?

Rory O'Driscoll

I think it's a great point because, as I think about it, the '90s were Unix and client-server, and the last 2 decades were effectively, I would say, Salesforce: We built Siebel in the cloud. It was fun, and we made a lot of money, but you're right, somewhat boring.

What's really exciting now is you're not just talking about, “We build Salesforce exactly the same, but with a slightly modern UI.” You are talking about something much more fundamental here in terms of eating the work and doing a lot more with the app. So, yes, it is exciting. But, looping right back to the first conversation about Sequoia, it is also clearly stressful for even the best of firms.

Harry Stebbings

Boys, is there any other topic that we haven't covered that you think we should cover? I think one thing that wasn't on the schedule, and that I just think is incredible, is freaking Hummingbird. Hummingbird, the fund that does not get credit and is not talked about in the same way that many other great firms are, made their first biotech deal in BillionToOne. They have an $800 million position at the IPO. You want freaking great venture returns in whatever $150 million fund that is. Credit due. Amazing.

Rory O'Driscoll

Agreed. Good for them.

Jason Lemkin

It was obviously incredible to see all the success. What I would like to know—and maybe we're out of time, and I should have done my research—is how they collected the capital as a traditional seed manager to deploy enough to maintain the ownership. I'm just obsessed with ownership now. It took me a long time to be obsessed with it, and now I've given up. But I don't see how I'll ever own 18% of something at IPO ever again, right?

Rory O'Driscoll

I think they were fairly capital-efficient as a company.

Jason Lemkin

It could be that answer. It could be the Veeva of biotech, where they were one and done. I'm looking for that dream.

Harry Stebbings

They definitely did put in subsequent checks, but I think it was a combination of doing subsequent checks, concentrating cash, and being a capital-efficient business.

Jason Lemkin

Yeah, I'm impressed with the deals they've gotten into. Maintaining the ownership with 9 figures of AUM—I don't know the exact numbers—that, to me, is just as S-tier, right? Figuring that out. Whether it's a bunch of side funds and SPVs and VPSs or VPNs, just how you do that is, I think, the elite game today. It's hard enough to get into the deals, but with a 9-figure or 8-figure fund, maintaining ownership is God-tier, I think.

Rory O'Driscoll

There's always the option of accepting—provided the follow-on rounds are at a high enough price—some dilution and optimizing, as you do, for multiple rather than ownership. You can say, “Look, I've established a 20% ownership as a seed fund.” I mean, those early funds those guys had were sub-$100 million. Okay, I'll get diluted from 20% to maybe 12% by the time we exit, but I've put in $4 million. I'm a hero.

Jason Lemkin

Yes, but this was a VC fund. I don't think they started with 20%, did they? But keep going.

Rory O'Driscoll

But fundamentally, the way they've been able to produce 8x and 10x funds has been in part by keeping them at $40 million, $60 million, or $100 million. It's hard to simultaneously keep your ownership and keep your multiple. You can decide which one you want to do. I mean, we talked last week: If you have Hummingbird, they've kept their multiple and gotten a 10x, 20x, whatever it is. If you talk about someone like Lightspeed, they've kept their ownership in the van and gotten a 5x, but on $250 million. You know, both of them are great outcomes. They're just different ways to play the game.

The interesting thing is that both of these different outcomes have great outcomes for the GP—the most important LP fact. If you only have $1 to play with, then obviously you want to do the one in the small fund that's going to give you the 10x, because if you have to deploy $100, then obviously you have to do the big fund. To some extent, the high-MOIC small fund—accept the follow-on dilution, but just make a marvelous return—is the compelling product for the marginal dollar.

Harry Stebbings

I think the outcome sizes are also interesting: Nirvana at $4.5 billion, now public, and BillionToOne at $5 billion. The power of capital efficiency and, bluntly, running lean—you see the difference as an investor. The benefits of investing in capital-efficient businesses: obvious statement.

Jason Lemkin

Yeah.

Harry Stebbings

Guys, I'm glad. It's nice to finish on a positive note, with a venture outlier that actually returns a huge amount of money to investors.

Jason Lemkin

Yeah.

Harry Stebbings

How nice. Thank you so much for joining me. I'm glad that you both approve of my shirt. This isn't my favorite.

Jason Lemkin

I am struggling with it, but I approve of it. And listen, how long are the pants?

Harry Stebbings

He's got pants on. He hasn't got shorts on.

Jason Lemkin

Well, they are jeans.

Harry Stebbings

But let's not push too hard. At least he's covering his knees. We'll take it as a win. Fantastic. You guys are stars.

20VC: Sequoia's Leadership Transition | Michael Burry Shorts NVIDIA and Palantir | Gamma Raises $100M at $2BN | Has Defensibility Died in a World of AI | Datadog Surges as Duolingo Plummets: What is Happening | BidClub