Apr 28, 2023 · 54m · news
Sarah Guo: On Her New $101M Fund; How AI Impacts Inequality; AI Startups vs Incumbents | E1007 · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this deep-dive interview, venture capitalist Sarah Guo details the launch of her $100M AI-focused fund, Conviction, and discusses how technical specialization, startup agility, and proactive security measures will shape the future of the artificial intelligence landscape.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 31.4% of the talking time here. How this is scored →
Harry as informed peer 4.4 Guest teaching 3.5 Guest disagreement 2.7 Harry pushing back 4.1
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 33:02 Seed stage defensibility is a myth
Sarah flatly rejects Harry's premise regarding startup defensibility at early stages, telling him directly that it does not exist and that investors are wrong to search for it.
Hardest push from Harry ▶ 44:45 Forcing guest to answer VC short questionHarry refuses to allow Sarah to dodge his buy/short prompt due to VC social relationships, insisting she commit to a short answer.
Biggest teaching moment ▶ 14:08 Explaining fund size discipline vs asset gatheringWhen Harry pushes on her $100M fund size math, Sarah reframes the argument by detailing how tight fund capital forces strict focus, avoids multi-stage operational bloat, and aligns returns.
Harry holds his own ▶ 14:08 Breaking down $100M fund check mathHarry demonstrates strong technical understanding of venture fund math, running through check sizes, portfolio diversification rules, and fee structures to challenge Sarah's fund model.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Why Sarah Left Greylock to Start Conviction | 1 | 1 | 1 | 1 | Harry warmly welcomes Sarah and asks why she left Greylock to launch Conviction. Sarah explains her desire to operate as a zero-to-one founder/investor and jokes about Harry's workload. | |
| Conviction's Thesis and Bet on AI | 4 | 3 | 2 | 3 | Harry probes Conviction's thesis and raises broader social concerns around wealth inequality and regulatory knowledge gaps. Sarah counters that technological progress creates abundance while acknowledging distribution challenges. | |
| Why a Vertically Centric AI Fund is Necessary | 5 | 4 | 2 | 5 | Harry challenges the need for a vertically centric AI fund by drawing historical parallels to Kleiner's climate fund. Sarah responds that deep domain expertise, community building, and applied research access create strong near-term specialization advantages. | |
| Do Venture Capitalists Understand Technical AI? | 5 | 4 | 3 | 5 | Harry asks candidly whether VCs actually understand AI and questions huge valuation rounds like Character.ai. Sarah distinguishes between media hype and real company traction, explaining model application economics. | |
| The Capital-Efficiency of Applied AI Models | 6 | 5 | 4 | 7 | Harry explicitly questions whether Conviction's $100M fund size is sufficient given check size constraints in AI. Sarah defends her fund sizing, arguing that capital constraints breed discipline and that applied AI startups can be highly capital efficient. | |
| Brand Checks and Hyped Valuation Rounds | 4 | 3 | 3 | 5 | Harry asks if Sarah puts $100k brand checks into hyped rounds just for brand prestige. Sarah rejects this approach, asserting that long-term brand is built on returns, reputation, and meaningful founder partnerships. | |
| Technical Deep Knowledge vs. Domain Expertise | 6 | 4 | 2 | 4 | Harry quotes Alex Rampell regarding startup speed versus incumbent distribution. Sarah shares details about Harvey.ai and notes that software development will increasingly rely on product-focused engineers leveraging AI tools. | |
| Tech Giants in the AI Race | 5 | 4 | 2 | 3 | Harry asks about incumbent AI moves, praising Microsoft and asking who is lagging. Sarah evaluates Big Tech positioning and explains how founders should conduct high-resolution customer discovery rather than chasing generic ideas. | |
| Founder vs. Market: The Force of Nature | 5 | 4 | 3 | 5 | Harry brings up tough markets like EdTech and the UK NHS to test whether great founders can overcome bad markets. Sarah agrees market dynamics matter but argues exceptional founders can navigate or re-architect distribution. | |
| Biggest Investing Mistakes and Regrets | 5 | 5 | 4 | 4 | Harry shares his own past emerging markets mistake and questions startup defensibility. Sarah candidly tells him seed defensibility does not exist and recounts her missed investments in Benchling and Rippling. | |
| Multistage vs. Boutique VC Firm Dynamics | 4 | 4 | 2 | 3 | Harry asks how founders should evaluate multistage giants versus boutique VC firms. Sarah outlines firm structural incentives and warns founders against internal fund dynamics like groupthink and seniority overriding. | |
| Is the Generalist Seed VC Dead? | 5 | 4 | 4 | 4 | Harry quotes Hunter Walk's claim that the generalist seed VC is dead. Sarah breaks down Hunter's argument but dissents with counterexamples like Jim Getz and Elad Gil. | |
| Learnings from Podcasting & 2023 Macro Outlook | 3 | 2 | 2 | 2 | Harry and Sarah share podcasting growth stories and banter about media production. Harry transitions into a quickfire round asking for a 2023 macro outlook. | |
| AI Opportunities in Services vs. Software | 5 | 4 | 5 | 7 | Sarah points out that AI services present a larger market opportunity than software alone. When Harry presses her to pick a VC firm to short, Sarah resists making enemies before identifying subscale seed funds and aggressive growth models like Tiger. | |
| Unknowable Outcomes, Reserves, and LP Landscape | 6 | 3 | 3 | 6 | Harry strongly criticizes standard VC practices like pro-rata reserve modeling and GP commits as inefficient. Sarah reflects on the unpredictability of market outcomes and LP dynamics. | |
| Defining Success for Conviction 20 Years Out | 2 | 2 | 1 | 1 | Harry asks Sarah to define 20-year success for Conviction. Sarah outlines her goals of top-tier fund multiples, market relevance, and founder trust. |
Statements from this episode (49)
Guo: Nation states and hackers will use AI tools for malicious code
“If you can do code generation, you can do malicious code generation, right? And just like other nation states and hackers are going to use every other tool out there. If they write code, they're going to write code with AI tooling.”
Sarah Guo: AI Is the Largest Value Creation Opportunity in Our Lifetimes
“So AI is the biggest value creation opportunity in our lifetimes.”
Guo: 10- to 20-person teams will build billion-dollar AI businesses
“Like I'm quite confident that we're going to have 10 and 20 person teams building billion dollar businesses.”
Stebbings: Wealth inequality has never been worse
“I'm really worried about wealth inequality. I think it's never been worse.”
Stebbings: AI knowledge gap between regulators and tech is unprecedented
“Like, and I think you've never seen such a big chasm between regulatory body knowledge and, like, the actual technology itself. This worries me that they're not in a position to actually regulate with the main knowledge.”
Guo: Advising national security bodies and visiting DC on AI risk
“I've engaged with like, you know national security bodies on policymaking in this area. And I'm spending time in DC two weeks from now on thinking about AI risk as well.”
Guo: Society will not have decades to adjust to AI capabilities
“And so I think the thing that is different today may be the speed of change, right? Like, I don't think we have decades to adjust to these capabilities in society.”
Guo: AI funds will eventually become general horizontal software funds
“I think it's eventually just a horizontal software fund, but right now, I think it's also useful for us to be specialists, right?”
Sarah Guo: Applied CS research matters now more than ever before
“Applied research matters in a way that has never before, right?”
Guo: Average VCs do not spend time studying computer science research
“I think your average venture capitalist does not spend a lot of time in computer science research”
Sarah Guo: AI driving biggest UX redesign in two decades
“We're rethinking a lot of user experience in a way that hasn't been true in two decades.”
Sarah Guo: Investing in AI Without Understanding Is Perilous
“Buying access to investments without some level of understanding sounds perilous.”
Guo: Very few AI startups can wisely spend $100M upfront
“I think that a vanishingly small number of AI companies can spend a hundred million dollars up front well, and, like, constraint is, is the name of the game in startups, right? It, like, breeds discipline and creativity”
Guo: Fewer than 10 AI startups should train foundation models
“The thing that is really expensive I mean, there are many things that can be expensive, but one of the things that is really expensive is I want to train a model from scratch that is very large and it's going to take me you know, low tens of people, probably 2…”
Guo: Most companies will apply existing AI models rather than train custom ones
“The vast majority of companies are going to figure out actually how to apply these models that other people have built that are offered by APIs or in the open source or fine tune them or like build some other part of the stack.”
Guo: Validating demand matters more for AI startups than training capability
“I think a much bigger question than, like, can you train a large model is, like, does anybody want it, right? Like, is it going to be useful? And, like, you could answer some of those questions up front.”
Guo: Conviction capped debut fund at $100M despite $500M demand
“I could have raised half a billion dollars, and I thought that doing this was, if I believe constraints breed discipline and creativity for founders, like, They also do for investors.”
Guo: Conviction will deploy $1M-$10M checks with high portfolio concentration
“They're seed and series A investments, and we're putting a million to like eight or ten million dollars to work, and we'll, we're gonna be more concentrated than most seed funds”
Guo: Raising larger funds for higher fees does not drive returns
“I think the pressure to like raise bigger funds and like collect larger fees and do more things is real. It's just like, doesn't matter. Right? Like, will that generate returns? I don't think so.”
Guo: $100M-$200M rounds at $1B valuations will not move the needle for a $100M fund
“I think it is unlikely that if somebody wants to raise a hundred or two hundred million dollars at a billion dollars out of the gate, like, that is hard for us to, it's hard for me to imagine that making an impact on the fund.”
Guo: AI can perform first-year legal associate work end-to-end
“We can do a lot of the work that a first year legal associate does end to end with these large models, and it's like, that's very valuable, right?”
Guo: Robotic process automation software will be reborn with AI
“RPA, big category of software, is gonna be obviously reborn.”
Guo: Product engineers, not researchers, will build most AI software
“I think over time, actually, we spent, you know, we spent a lot of time with the research community, but I think the vast majority of companies, like, software is going to end up being built by, with love and respect, like, run-of-the-mill product-oriented eng…”
Guo: AI coding tools will soon handle end-to-end software creation
“Like, I think we're gonna get much more end-to-end stuff soon.”
Guo: AI will not eliminate software engineering as a profession
“That doesn't mean I think engineering as a profession goes away. I think they just, like, leverage these tools much better.”
Guo: Incumbent data moats are overrated against creative founders
“Much ado has been made about this idea of like a data moat. I'm sure you've heard this term, but honestly, there's a lot of data out there and entrepreneurs are incredibly creative about collecting it and increasingly about generating it.”
Guo: Satya Nadella and Kevin Scott led Microsoft's AI transition brilliantly
“I mean, how could you not, right? I think Satya has done an amazing job making a, making Satya and Kevin Scott have done an amazing job championing, like really believing in this set of technologies, taking a bunch of bets, open AI and otherwise, using it as a…”
Guo: Google, Amazon, and Apple are not leading the AI race
“I think you'd be hard pressed to say that, like, Google or Amazon or Apple is, like, leading the field here.”
Guo: Amazon and Apple must invest heavily to catch up in AI
“I think that Amazon and Apple will have to make more investments over time in these technologies than, like, they don't have labs doing cutting edge work when, like, They have the skill to support them.”
Stebbings: Restaurant analytics is a terrible startup idea
“We know that restaurant analytics is not a good idea. Like, it's just not. It's, and if it does work, it's gonna be so much harder than anything else you brilliant entrepreneur could do.”
Guo: Generative 3D modeling carries zero market risk and high value
“The world is built on three models from everything from entertainment to the physical world around us. It is an open research question as to whether or not you can generate three D models that are usable in these use cases, but it's like, ah, there's no market…”
Stebbings: I pass on great founders in edtech due to churn
“I have turned down companies before because there is an amazing founder, but it's in ed tech, which just hell or high water. The churn is a complete bitch and Duolingo is the leader, but it's still not a hugely valuable company”
Guo: Don't dismiss founders operating in tough markets
“My view is you don't just turn away from all problems because, like, the space has traditionally been hard, but it should, like, it should educate you as to, like, does the founder recognize that, and can they navigate out? Are they doing something to uniquely…”
Harry Stebbings: Investing in Emerging Markets Was My Biggest Mistake
“Well, my biggest mistake was investing in emerging markets.”
Sarah Guo Regrets Passing on Benchling and Rippling at Series A
“One of the investments I, like, actually, multiple investments I regret not making, so Benchling, Rippling at the A, right?”
Guo: Looking for defensibility at the seed stage is a mistake
“Yeah, it doesn't exist, right? Like quite literally you're starting with nothing. And so I think that it is a like I think investors are wrong to look for it. What you are investing in is trajectory, right? And the ability for founders to navigate a market and…”
Guo: $5B funds lack skin in the game on $5M checks
“To me, fundamentally, I think it's like very hard for a five billion dollar fund to have skin in the game on a five million dollar investment.”
Guo: VC Returns Are Dominated by Few Individual Investors
“Returns in most firms are dominated by a few good investors, even when the partner group might be 10 or more.”
Guo: Large Investment Committees Risk Poor Decisions From Groupthink and Politics
“And so I think there's real risk to good investment decision making in big groups. Groupthink, seniority overriding, like, positioning in politics.”
Guo: Only one in five founders conducts investor reference checks
“I'm shocked that maybe one in five founders, like, does investor references.”
Stebbings: $250 Diet Coke promotion pushed 20VC to #2 in UK
“I remember going with three friends to a football stadium, and having 500 Diet Coke cans strapped in, like, drinks, rucksacks, and saying, we'll give you a free Diet Coke, if you'll give us your phone for a review. And we spent, like, the 50 P each, so like, a…”
Guo: Most tech startup macro pain is yet to come in 2023
“Assuming you're talking about the macro, like, it's, you know, I think most of the pain is yet to come. We'll still be ugly.”
Guo: AI's primary opportunity lies in services, not software
“It may be it may not be very well understood that a significant part of the opportunity for AI is services, not software markets, right?”
Guo: Subscale, undifferentiated seed funds will struggle to persist
“I think that it will be a hard time for subscale seed stage funds without a differentiated strategy to persist. I think like it was not hard to raise 10 to fifty million dollars for a couple years and it just will become more because LPs are going to become mo…”
Guo: Tiger Global's index-style early-stage strategy will perform poorly
“Short on a multi-stage fund. I think early in growth investing is more different than it appears. And so I think there are firms that like, you know, tried the early stage investing route with like a blanket based approach. And like, I don't know if that's goi…”
Guo: Early-stage startup outcomes and market trajectories are inherently unknowable
“I am increasingly convinced that, like, it's not knowable what the outcomes are from, for companies, like, at the very beginning, and specifically, like, how markets play out is unknowable, because, like, there are actors with agency determining how the market…”
Guo: Conviction will not reserve capital for follow-on investments
“Like we are going to invest in the early rounds of a company and we're going to leave money on the table in the later rounds. And it's an explicit decision, but what it means, it's like very focusing, right? Like we make the bets we make with them. We're align…”
Guo: Most LPs lack individual conviction and blindly follow big brands
“But the LP landscape overall is very clubby, right? And I'm sure you experienced this. Much like venture, a lot of investors they lack individual conviction. They simply follow bigger brands.”
Sarah Guo: Conviction targets best-in-class venture multiples over absolute dollar returns
“For me, like I define it on a multiple basis, not an absolute dollar basis. Otherwise like I should have raised a larger fund. And like, you know, let's put best in class venture multiples on the board, right?”