Matan Grinberg | 20VC Summary

1. Key Themes

The ROI Reckoning: Three Phases of Enterprise AI Adoption

Enterprises are moving through a predictable arc: board pressure to adopt AI, reckless "token maxing," and now a hangover phase of cost accountability. This has direct implications for how AI vendors need to evolve their sales and customer success motions.

"Phase one, this was a couple months ago, was board yells at CEO, hey Mr. CEO, what's your AI strategy? ... Phase two was AI at all costs, token maxing. Part of your performance reviews, we're gonna measure how much you guys use AI... Phase three is the hangover, where you go and look at the bill and it's like, oh my God, we are spending so much. I have no idea what the ROI is." 00:18:16

Token Economics Will Rival Headcount Economics Within Three Years

The ratio of token spend to developer salary is not a minor line item — Matan believes it will reach salary-comparable levels within three years, making model routing and token governance among the most strategically important decisions a CTO will make.

"I would say order of magnitude it will probably be comparable to salary. Comparable to salary? Yeah, like on the same order of magnitude. Within the three year timeline? Yeah." 00:23:36

Value Accrual in AI Is Time-Dependent, Not Structural

Everyone across the AI stack — models, applications, infrastructure — is actively trying to commoditize the others. There is no permanent winner; pricing power shifts based on who has leverage at a given moment.

"The reality is value accrual is a time dependent phenomenon. It's not like there is one person whose steady state gets all of the value... everyone is trying to commoditize the people that are not them." 00:13:05

Open Source as a Strategic Counterbalance to Frontier Model Spend

Approximately 80-90% of coding tasks can be handled by open source models, with frontier models only truly necessary for planning/decision-making steps. This creates a natural routing architecture and puts a ceiling on how much frontier model providers can charge.

"Probably 80 to 90% [of tasks could be done with open source]. It's typically the planning that really needs the frontier models." 00:23:36

"A lot of enterprises will realize so many of the tasks that we're doing we don't need the very frontier to do it. Like we can do it much faster, much cheaper with these open models." 00:16:54

The Age of the Polymath Has Returned

AI is compressing the time required to reach the frontier of multiple disciplines simultaneously — reversing centuries of increasing specialization. This has direct implications for who to hire and how to structure teams.

"With AI we're now completely the opposite. These tools can get you up to speed to the frontier... And so now if you're someone that's good at thinking around constraints, thinking about systems, holding uncertainty in your head and being okay with that... you can be a polymath." 00:33:19

Sales and Marketing Are First-Class Product Functions, Not Afterthoughts

Matan argues that Silicon Valley's cultural hierarchy — research > engineering > sales — is a dangerous delusion that will haunt companies when competition intensifies. The entire customer journey, from first brand impression to tenth renewal, is the product.

"The product at Factory is the entire journey from the very first time they hear our name till their tenth renewal after a decade of being a happy customer... Name a legendary company that has a shit sales or marketing team. You can't." 00:26:39

Engineers Must Become Full-Stack Business Owners, Not Code Custodians

The role of the engineer is shifting from shipping features to owning end-to-end business outcomes — including marketing copy, sales enablement, and customer behavior change. The skills that made engineers valuable (syntax memorization, competitive coding) are becoming irrelevant.

"The best engineers are going to be the ones that don't see sales and marketing as dirty work... You are owning full end-to-end outcomes of here's the way a customer is behaving. Here's how maybe we can change that behavior... This is like a full stack engineer that goes way beyond just engineering but into sales, into marketing, into enablement." 00:28:32

Security Is the Invisible Ticking Clock of AI-Generated Code

Code generation is growing exponentially while security efforts are not keeping pace. Matan expects significant incidents in the next couple of years, many of which will likely be attributed to AI-generated vulnerabilities.

"Code generated is growing exponentially. The security efforts aren't growing in kind. And so I think there's kind of a lag there. I think there're probably going to be in the next couple years some pretty big incidents that occurred because of AI." 01:03:56

Engineers Build Factories, Not Code

The most important reframe of Factory's entire thesis: future engineers don't write software, they design and maintain the automated systems that produce software — analogous to Tesla engineers who design robotic assembly lines rather than hand-assembling cars.

"The future of software development is where these organizations, instead of having engineers that build the software, they're going to have engineers that build the factories that build their software." 00:38:57

Vendor Lock-In Trauma Is Reshaping Enterprise AI Procurement

CIOs have fresh scars from cloud vendor lock-in and are deliberately architecting AI procurement to be model-agnostic from day one — creating structural demand for application-layer companies that offer model routing rather than single-provider tools.

"Everyone has scars from that. So now every CIO I speak to is really keenly aware of we cannot throw our lot in with just one model provider. We're going to need to be agnostic." 01:00:50

2. Contrarian Perspectives

Labor Displacement Fears Are Overblown — The Real Problem Is Engineering Talent Distribution

Against the prevailing narrative of AI causing mass unemployment, Matan argues there is a near-infinite backlog of unsolved problems addressable with software. The real issue is that engineering talent has been concentrated at a handful of large tech companies rather than distributed across the economy.

"Very few of those problems that can be solved with software are we currently solving with software. And so if we are going to be flooding the job market with tons of engineers, that means that we can now allocate them on the broader economy to solve more of these problems in the world." 00:40:16

The Kirkland & Ellis $500M AI Build Is Good for Harvey, Not a Competitive Threat

Most observers read a law firm building its own AI as a threat to AI legal vendors. Matan reads it as validation of how hard it is — and predicts Kirkland will ultimately become a better Harvey customer after experiencing the difficulty firsthand.

"I think as an example... it's nothing like trying to do something yourself to make you realize, oh shit this is actually really difficult. This doesn't actually matter for us to have the in-house ability to build this ourselves. Let's go and have someone who is an expert in this to go and build this for us." 00:11:09

Credentialism (Math Olympiads, AIME, IMO) Is Becoming Anti-Signal for the Best Future Engineers

While VCs use competitive math credentials as a proxy for talent in the absence of other certainty, Matan argues that following the prescribed credential path is actually evidence of low agency — the opposite of what matters in the agent-native world.

"There's some high schools that like really focus on like you must do the Math Olympiad... like you must do this, the AIME to then go to the IMO and like this is the path to success. Where actually that's kind of anti-signal. Because there it's like you're not owning your fate or choosing your agency. You're kind of going through the funnel." 00:30:15

Dario Amodei's Public Statements About AI Taking Jobs Are Both Wrong and Harmful

Despite Matan leaning positively toward Anthropic as a business, he is sharply critical of Dario's repeated public messaging about AI job displacement — calling it disingenuous and damaging to the ecosystem.

"Yes it actually this like really upsets me so on one hand I maybe just implied Anthropic there but on the other hand I think that has been not only like disingenuine and wrong but..." 01:15:15

FDEs as a Sales Tool Are a Sign of a Bad Product

The industry norm of deploying field deployment engineers to close or retain enterprise deals is widely accepted. Matan argues that if FDEs are necessary to make a product work rather than to accelerate adoption, you simply have a bad product.

"If I'm sending in FDEs as services like I'm not Accenture here... If we need FDEs to make the product work, we have a shit product. Like the point of FDEs should be accelerate and get them consuming faster. If you're putting in FDEs because that's the only way you'll get a deal done, I'm sorry my friend you have a shit product." 01:09:18

3. Companies Identified

Factory AI-native software development platform; model-agnostic coding agents for enterprises. Raised at $1.5B valuation. Works with some of the largest enterprises in the world. Matan's framing of "engineers building factories, not code" is the core product thesis. The company emphasizes model routing across frontier and open-source models to optimize for cost, quality, and speed.

"We want to give our customers the best pricing, the best performance, the best speed for whatever task they want to do in their software development. And we want to make sure that OpenAI, Anthropic, Google, Microsoft are all under pressure to make sure they give the best models for as cheap, as quick as they can." 00:13:32

Harvey AI legal platform. Mentioned as analogous to Factory in the legal vertical — a specialized application layer on top of models that enterprises should use rather than building their own tools.

"That is my sense... I think as an example, we're so used to a world where moat in software was, I know how to do this and you don't... I think this is good for Harvey." 00:10:11

Sequoia Venture capital firm. First check investor in Factory ($1M at $5M post). Alfred, Pat, and Roloff were in the room for Factory's first pitch in April 2023 — before agents were a mainstream concept.

"A million dollars. And you know what he gives me shit for? You know what the terms? Five posts." 00:53:49

Anthropic Frontier AI model lab. Mentioned as one of the two dominant frontier model providers. Harry notes Salesforce (Mark Benioff) spends $300M on Anthropic for their devs, representing 3.8% of dev salary spend today.

"In my mind the answer here is I think they're approximately equivalent... from the business perspective they're both very well suited and well positioned." 01:14:39

OpenAI Frontier AI model lab. Noted for higher organizational turbulence than Anthropic, but equivalent business positioning. Released a competitive product to Lovable/Replit the night before recording.

"Probably past is an indicator of the future and like there's just been more like random chaotic turbulent events at OpenAI but like from a business perspective to me that's like they're both great choices." 01:14:39

Stripe Payments infrastructure. Mentioned as the gold standard for developer documentation — a competitive advantage that will be equalized by AI but that Stripe has other ways to differentiate.

"Stripe has a really great reputation. They had incredible documentation... Five years from now it's going to be like oh my God, I cannot imagine, cannot believe that these people that get paid so much money spent hours of their time doing this." 00:35:30

Nebius GPU cloud / AI infrastructure company. Mentioned in the context of the pace of model releases (founder noted new models every few days, especially Chinese open source). Also compared to Coalesce in a quickfire round; Matan prefers Coalesce given Nebius's more full-stack ambitions potentially conflicting with application-layer companies.

"To me and this is speaking from strongly biased as an application person... I would want Coalesce to be bigger. Why? Because Nebius I think have more ambitious plans to be full stack which will eat into some of your plans." 01:08:09

The Chainsmokers / Mantis VC The Chainsmokers' investment vehicle. Described as "incredibly good investors" — often underestimated. Brought Francesca (key Factory hire) into the orbit of Factory.

"Alex Paul, who's one half of the Chainsmokers. And obviously people know them as the Chainsmokers. They're also incredibly good investors. Incredibly good investors. Which sometimes people are surprised by." 00:56:29

Tesla Manufacturing company. Used as a conceptual model for the future of software development — engineers designing robotic factories rather than doing manual assembly.

"Visually, whenever I say this, I always think of Tesla's factories... It's all these robotic arms going and you have the assembly line going through and there might not be as many humans in that assembly line but you know damn well that humans designed this process." 00:38:57

Eight Sleep Smart mattress company. Factory bought Eight Sleeps for their entire team (~30 people) during a two-week sprint. Matan advocates for treating engineers like professional athletes — investing in recovery as a productivity multiplier.

"When we were 30 people we had like a surge like a pretty aggressive like two week sprint and as part of it I got everyone on the team Eight Sleeps like fully free whatever $3,000 per person." 01:11:30

Kirkland & Ellis Law firm. Announced $500M spend over five years to build internal AI tools. Matan uses this as a case study in companies pursuing non-core competency work — predicts they'll become a better Harvey customer as a result.

"My understanding is that building AI technology is not a core competency of that firm. So I was surprised to see it." 00:10:11

Y Combinator Startup accelerator. Matan credits YC YouTube videos and Peter Thiel's Zero to One as his introduction to startup culture before founding Factory.

"Was watching these videos, a lot of them like Y Combinator, you know, videos and all this stuff." 00:50:05

4. People Identified

Matan Grinberg CEO and co-founder of Factory. Former theoretical physicist; worked with Juan Maldacena at Princeton as first undergraduate co-author. 12 years in physics before founding Factory. First job ever. Raised Factory from a $1M seed at $5M post (April 2023) to a $1.5B valuation round. Described by Harry as one of the best founders he's met in the last year — invested millions after a ~5-minute walk in Hyde Park.

"No one else would have believed in me except him. No one else would have understood. Like, I literally had never had a job before. Like, no other partner I would have met." 00:00:29

Sean (Sequoia Partner — implied) Sequoia partner with a physics background who sold a company for $1B and became an investor. Wrote the first $1M check to Factory after a three-hour walk. The connection was established because Matan recognized his name from having cited his academic paper at Princeton.

"It was supposed to be a 30 minute meeting. But we end up going on this walk and ends up being a three hour walk. And on this walk, it turns out we had very similar reasons for getting interested in physics, very similar reasons for leaving physics." 00:50:57

Juan Maldacena Famous theoretical physicist and Princeton professor. Matan was the first undergraduate to work with him and co-author a paper — the paper citation is what led to Matan's email to the Sequoia partner and ultimately the founding of Factory.

"I ended up going to Princeton because I had a great physics professor I wanted to work with. He's this famous professor named Juan Maldacena. And I was like the first undergrad to work with him and write a paper with him." 00:46:50

Eno (Factory Co-Founder) Princeton alum. Met Matan at a San Francisco hackathon the day after the Sequoia meeting. Described as "a thousand X better of an engineer" than Matan. Together they built a better demo in 72 hours that led to the Sequoia check.

"Eno is a thousand X better of an engineer than I ever will be. And so he and I for the next like 72 hours put together this better demo." 00:52:26

Francesca (Factory GTM/Business hire) Former investor, connected to the Chainsmokers' firm Mantis. Met Matan at a random conference. Described as a "killer" — was so relentless in pursuing more allocation that Matan jokingly offered her a job, which she took. Became one of Matan's best hires.

"She was fucking relentless... I was like look Francesca like if you want more ownership Factory you could just join us... She was a killer." 00:56:57

Ivanka Trump Investor in Factory. Brought in through Francesca's network. Described as genuinely intelligent, kind, and possessing a significant network — does "dirty work" investor help that more prominent investors avoid.

"She is first of all she is one of the kindest and smartest people that I've met... She's so generous with her time. There is kind of dirty work investor help that she helps out with that some other investors who are more known as investors do not do." 00:58:11

Alex Paul (The Chainsmokers) Co-founder of the Chainsmokers music act and Mantis VC. Described as a surprisingly excellent investor. Was Francesca's close collaborator before Factory hired her away.

"Alex Paul, who's one half of the Chainsmokers. And obviously people know them as the Chainsmokers. They're also incredibly good investors. Incredibly good investors. Which sometimes people are surprised by." 00:56:29

Andrej Karpathy Former Tesla/OpenAI AI researcher. Referenced by Harry as having recently discussed the concept of the 100X engineer and the bifurcation of engineering talent.

"I was watching Andrej Karpathy and he was talking recently about, you know, the 10X engineer actually is wildly misunderstood and you won't see the 10X engineer, you'll actually see a smaller number of 100X engineers." 00:06:44

Dario Amodei CEO of Anthropic. Criticized for repeated public statements claiming AI will take jobs — Matan calls this disingenuous and harmful to the ecosystem even while respecting Anthropic as a business.

"Has Dario done a disservice to the ecosystem by saying we're going to take your jobs, we're going to take your jobs, we're going to take your jobs? Yes it actually this like really upsets me." 01:15:08

Mark Benioff CEO of Salesforce. Referenced as spending $300M on Anthropic for developer tooling — 3.8% of dev salary. Harry uses this as the baseline for projecting how token spend ratios will evolve.

"If Mark Benioff says that he spends 300 million on Anthropic, okay, for his devs, that is 3.8% of salaries." 00:21:29

Winston (Harvey) Founder/CEO of Harvey. Described as a "fantastic guy" and personal friend of Matan's. Referenced in context of Kirkland & Ellis's $500M spend and enterprise AI legal tooling.

"You're friends with Winston from Harvey, fantastic guy." 00:09:52

Peter Thiel Author of Zero to One. Credited as Matan's introduction to startup culture — his first Amazon order after deciding to start a company was the book.

"Read Zero to One, incredible book. I know it's so cliche, but like to someone who didn't know, growing up in the Bay Area, shockingly, I just like did not care about any of that. And reading this, it was like so concise, beautifully written." 00:49:36

5. Operating Insights

Treat Token Budgets Like Headcount Budgets — But Never Uniformly

The mistake enterprises are making is either spending with no governance or imposing uniform per-person caps. The right model is differential allocation based on leverage: some engineers should spend orders of magnitude more tokens than others based on what generates business outcomes.

"If your org has a standard number where it's like we want every engineer to be at this percent of their salary and token use, you're probably painting with way too wide a brush." 00:23:16

The Post-Sales Motion Needs a Proactive Token Review, Not Just Onboarding

Factory learned the hard way that giving customers unlimited access leads to bill shock and churn risk. The fix is a structured conversation during onboarding about conscious token allocation — which parts of the codebase warrant which level of model spend — rather than waiting for the customer to discover runaway costs.

"We need to make sure with every customer we are having a very clear conversation with them of, you know, it looks like you guys are spending a lot of tokens on some of these things. Have you thought about consciously, yes, we want to do this. Sometimes we'll proactively set in those user limits." 00:20:45

Invest in Developer Experience Infrastructure — Its ROI Multiplies With Agent Count

Setting up CICD, linters, pre-commit hooks, and remote execution environments has always been best practice — but the return on that investment now scales with the number of agents running, not just the number of humans. This creates a new business case for DevEx investment in agent-heavy organizations.

"The impact of doing that well is just like one-to-one kind of correlated to how many engineers you have. With agents though, the impact of that is now like 10X or 100X depending on how many agents you're using." 00:37:16

Sell by Genuine Curiosity About Customer Problems, Not Product Features

Particularly important when selling to engineers, who will immediately disengage from a pitch. The most effective enterprise sales motion is deep curiosity about how each organization uniquely builds software — the variety itself is compelling — which organically surfaces whether the product is a genuine fit.

"If you go in a conversation trying to sell something, especially to engineers, don't waste your time. If you go in trying to have genuine curiosity... these organizations do their engineering so differently and I find it fascinating how all of these different banks, consulting firms, pharmaceutical companies, they have the most different ways of building software." 00:45:13

Hire for Agency Over Credentials — Especially Anti-Credential Agency

The most predictive signal for the new generation of engineering talent is not what competitions they won but whether they pursued those interests independently, without institutional pressure. Self-directed obsession from "the middle of nowhere" is stronger signal than top-of-funnel credential achievement.

"There are people on our team who are from the middle of nowhere where no one else in their high school ever did this stuff and they kind of took the agency of like I think this is really fun, I'm really competitive, I want to compete at this and they kind of go and do those competitions on their own. Those are when the signal is still positive for this kind of engineer of the future." 00:30:44

6. Overlooked Insights

The "Trigger Word" Risk in Chinese Open Source Models Is Theoretically Self-Defeating — But the US Has No Frontier Open Model Alternative

Matan dismisses the national security concern about Chinese open source models containing hidden adversarial triggers — not because the risk is zero, but because any nation attempting this would rationally wait until maximum deployment before activating, making early-model deployment relatively safe. However, buried inside this reassurance is a pointed observation that has received almost no attention: the United States has no competitive frontier open source model, which Matan calls "pretty embarrassing." This is a significant gap — if enterprises need open source for cost routing (the 80-90% of tasks that don't need frontier), and the only frontier-quality open models are Chinese, US companies face a strategic dependency that the current policy debate has largely ignored.

"I do think just from a... I'm quite patriotic... I think it's pretty embarrassing that we don't have frontier open models in the United States. So I do hope to see us reclaim superiority there." 01:05:29

The "Agent-Native" DevEx Investment Thesis: A Hidden Infrastructure Trade

Almost no one in the conversation paused on the compounding implication of Matan's point about developer experience infrastructure. When DevEx ROI scaled 1:1 with human engineers, only the most disciplined organizations invested in it. Now that it scales 10-100X with agent count, every hour spent improving CICD pipelines, linters, documentation freshness, and remote execution environments has asymmetric return. This creates a stealth investment category: companies and tools that improve the "factory floor" of agent-native software shops will capture outsized value — and the customer who understands this will compound their agent productivity advantage dramatically faster than competitors who treat DevEx as overhead.

"The better your DevEx, the better your agent ends up adhering to your standards, which means there's less time that that poor staff engineer has to go through reviewing your PR, which means you're kind of faster throughput in your software development." 00:37:41