May 6, 2024 · 1h 5m · news
Sarah Tavel: Will Foundation Models Be Commoditised? | E1149 · 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, Benchmark General Partner Sarah Tavel discusses the economics of AI application defensibility, the paradigm shift toward "selling the work," and the boutique, high-conviction partnership model that defines Benchmark's investment philosophy.
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 19.3% of the talking time here. How this is scored →
Harry as informed peer 5.9 Guest teaching 5.1 Guest disagreement 2.3 Harry pushing back 5.1
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 6:57 Rejecting premise that founders create markets by sheer will
Sarah directly rejects Harry's suggestion that great founders will markets into existence without a strong underlying catalyst.
Hardest push from Harry ▶ 28:52 Calling out VC FOMO and lack of disciplineHarry aggressively challenges the venture industry's willingness to overpay and accept heavy dilution due to fear of missing out.
Biggest teaching moment ▶ 12:00 Reframing application software as selling outcomes rather than seatsSarah educates Harry on a fundamental mental model shift from seat-based productivity tools to selling complete work outcomes.
Harry holds his own ▶ 42:22 Challenging Benchmark's anti-platform philosophy with specialist quality argumentHarry pushes back firmly against Sarah's critique of platform talent teams, arguing dedicated recruiting experts produce better results than general partners.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Sarah Tavel's Path to Benchmark | 2 | 2 | 1 | 1 | Harry opens with friendly, open-ended questions about Sarah's recruitment path to Benchmark. Sarah recounts her background and initial hesitations without any friction. | |
| The Genius of Peter Fenton | 6 | 6 | 3 | 5 | Harry challenges Sarah's emphasis on 'why now' by asking if elite founders simply will markets into existence. Sarah rejects this premise, and Harry illustrates his point with a self-critical analysis of his seed investment in BeReal. | |
| AI: Sustaining Incumbents vs. Disrupting with 'Selling the Work' | 7 | 6 | 2 | 6 | Harry presses Sarah using previous interview quotes from Sam Altman and Brad Lightcap regarding model readiness and incumbent advantages. Sarah educates on the paradigm shift from per-seat productivity software to 'selling the work' outcomes. | |
| Value Capture and Defensibility in the Application Layer | 7 | 5 | 2 | 6 | Harry questions startup defensibility in the application layer against Microsoft's bundling power and distribution dominance. Sarah breaks down wrapper value vs deep workflow ownership. | |
| Hype vs. Reality: Measuring Genuine Enterprise AI Adoption | 4 | 5 | 1 | 3 | Harry asks how to differentiate genuine enterprise AI adoption from temporary POC budgets. Sarah provides framework criteria including Sean Ellis product-market fit metrics and cohort retention analysis. | |
| Navigating Valuations, Dilution, and FOMO in AI Investing | 7 | 5 | 4 | 7 | Harry confronts the room with a sharp critique of VC FOMO and valuation inflation, stating investors behave like goldfish repeating past mistakes. Sarah concedes the point while explaining how expanded TAMs rationalize higher entry valuations. | |
| Network Effects and Moats in the AI Era | 7 | 5 | 2 | 5 | Harry incorporates specific crowd-sourced questions from 11 Labs founder Mati and challenges Sarah on how fast incumbents reach product parity. Sarah argues deep workflow integrations protect specialized startups. | |
| The Oligopoly of Frontier Models and Market Competitiveness | 6 | 6 | 3 | 6 | Harry challenges the idea that model training must become more expensive over time rather than cheaper. Sarah corrects him by explaining how compute demand growth outpaces technical efficiency gains. | |
| Benchmark's Unique Partnership Philosophy and Board Dynamics | 7 | 6 | 3 | 7 | Harry directly counters Sarah's criticism of venture platform talent teams, arguing domain specialists deliver a superior recruiting product than GPs. Sarah defends Benchmark's hands-on, partner-led model. |
Statements from this episode (24)
Tavel: Frontier AI models currently require closed-source architectures
“Right now it feels like if you need a model that's on the frontier, you're going to be closed source.”
Tavel: AI application layer will capture tremendous long-term value
“I focus on the application layer because I do think that's where a tremendous amount of value ends up being captured and created.”
Tavel: Escalating training costs will produce foundation model oligopoly
“Each successive model is going to be more and more expensive. That suggests a world where you're going to have an oligopoly.”
Tavel: Peter Fenton is the single best VC at developing founders
“I've learned from Peter because his, EQ around people, what motivates us, what drives us what the, what holds people back, and how important that is in the, you know, the arc of any company that we end up investing in, this watching, helping those founders bec…”
Tavel: A strong 'why now' allows startups to survive mistakes
“When you have a strong why now, to me, it's like this strong current that just pushes the company forward. And it means that there are so many things that as a founder, There's, you can, you know, we always say that being a founder and going through hyperscale…”
Tavel: BeReal failed because it couldn't compete with short-form video dopamine
“What happened with Be Real in my estimation is that they did capture that why now of the authenticity But they could never earn the right to enough minutes for a consumer because the consumer was then having, you know, the dopamine hits of, you know, the most …”
Tavel: AI for existing workflows will drive tremendous market cap for incumbents
“If it's existing employees and their existing workflows that, you know, AI technology is going to be sustaining and it's going to drive tremendous market cap for the incumbents.”
Tavel: AI enables startups to sell completed work rather than software seats
“What AI enables is actually a very different unit of work that you sell, which is doing the work. And so you're almost a software company that looks like a services business that is able to sell like the full work product that the outcome, as opposed to sellin…”
Tavel: Early-adopting AI startups will take market share from traditional service firms
“New technologies, more often than not, are best served for the group of hungry up-and-comers that want to become the new incumbents. And so, you're going to see a class of companies that realize that if they're early adopters of this technology, they're able t…”
Sarah Tavel: AI application startups will improve and expand margins as models advance
“I think all the companies that we invest in, they will only get better as the underlying models get better. And they'll be able to take on more work over time. They'll be able to, you know, take, you know, have better and better margins, better because they wo…”
Sarah Tavel: First-wave AI wrappers derived 90% of product value from OpenAI
“The first wave of these AI companies, you could, they, people would talk about them As a, like, you know, wrappers around an LLM. But what that essentially meant is that if you have like this hundred percent distribution, 90% of the value that the startup was …”
Tavel: AI Adoption Is Currently Driven by SMBs and Mid-Market
“Right now we're just not, it's too early to know for that enterprise use case. What you're seeing, like where most of this adoption is happening is like the mid-market and down, this kind of digital native company, the SMB, the, these, like the hungry companie…”
Tavel: Selling AI work expands target markets by 10x-50x over software
“Part of the opportunity with selling work as an AI startup is that the market for that could actually be 10 or 50 X bigger than if you were selling software. Why is that? You're selling a 95% productivity improvement instead of a 10% productivity improvement. …”
Tavel: Benchmark partners make only 1-2 new investments annually
“Our way of thinking of things that benchmark is that each of us makes one or two new commitments a year.”
Tavel: B2B software relies on economies of scale, not network effects
“B to B software has very rarely had network effects, right? What it has had instead is more economies of scale.”
Tavel: OpenAI will not match specialized audio and video startups' focus
“Those are going to progress, but they're not going to progress with the same level of focus and ambition as the core foundation model that OpenAI is working on. They might progress as beneficiaries of that core work, but it doesn't have the same level of focus…”
Tavel: OpenAI will not build deep enterprise workflow integrations
“So that something coming out, you know, an open AI, they're just not going to build that same level of workflow, the same level of integration. And so it's going to be, there's, there are, there's plenty of work to be done. That's beyond the model itself.”
Tavel: End-customer AI model costs will continue falling due to competition
“The cost to the end customer, I think that should continue to go down. And that's what we're seeing is that it is just getting more and more competitive. The price continues to go down.”
Tavel: Frontier AI models will remain closed source unless Meta shifts market
“If you want a model that's on the frontier, that's going to be closed source. Now, look, every, as I mentioned, every month it feels different with what Meta's doing with Llama, that may actually fundamentally change the game. If Meta is willing to make that h…”
Tavel: Benchmark waived board seats only once in seven years
“There, in my seven years at Benchmark, I think I've seen one example where we broke the model on, on board, and it was for a small crypto related company.”
Tavel: Pay high startup prices if you like everything else
“In those cases, someone said this to me once, and it really hit me, is like, if you like everything but the price, you pay the price.”
Tavel: Benchmark barely reserves capital for follow-on investments
“We don't think about reserves. Period. Barely. We're very oriented towards our initial investment, and in that moment, you know, we are more ownership sensitive than probably most other firms. But then after we invest, we are 100% aligned with whatever objecti…”
Tavel: Benchmark's model ensures 100% post-investment alignment with founders
“But more often than not, it creates conflict that's unnecessary, and it's just very different than our model, which is like, okay, now that we've invested, we are 100% on the side of your side of the table, and let's make every round as successful for you as p…”
Sarah Tavel names passing on Ethereum as her biggest investment miss
“Without question, Ethereum. I, you weren't expecting that. I can see, you know, what, well, Ethereum, I mean, I remember I was at Greylock. And the ICO had just happened, I believe, or just closed. And I was reading more about what's possible with smart contra…”