Jun 13, 2026 · 1h 25m · news
OpenAI vs Anthropic vs Open-Source | Token Maxing, AI Hangovers & The Coming ROI Reckoning · 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
Matan Grinberg, co-founder and CEO of Factory, discusses the transition to agent-native software engineering, how enterprises are navigating the 'AI hangover' with cost-conscious routing, and why the future belongs to multidisciplinary, polymath developers rather than raw coders.
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 20.8% of the talking time here. How this is scored →
Harry as informed peer 4.1 Guest teaching 3.8 Guest disagreement 3.2 Harry pushing back 3.2
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:21:30 Matan Slams AI Founders' Job-Loss Rhetoric
Matan delivers his most forceful disagreement of the interview, calling rhetoric about AI taking everyone's jobs disingenuous, selfish, and designed solely to raise unprecedented amounts of venture capital.
Hardest push from Harry ▶ 56:59 Harry Challenges Sequoia Partnership Pitch DynamicsHarry directly refuses the premise that Sequoia would run a full partnership presentation for a modest $1M check at a $5M post-money valuation, openly questioning why a multi-billion dollar firm would hold a partnership meeting for such small dollars.
Biggest teaching moment ▶ 16:30 The Three Phases of Enterprise AI and the AI HangoverMatan educates Harry on enterprise AI adoption dynamics, reframing simple model usage into three distinct structural phases and illustrating the current 'AI hangover' with a story about employees querying expensive models for food macros.
Harry holds his own ▶ 27:10 Harry Hits Back with Bad Product / Great Sales CounterexampleWhen Matan challenges Harry to name a legendary company with a bad sales team, Harry immediately hits back with the counter-observation that many legendary tech companies actually have awful products carried entirely by elite sales teams.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Podcast Title Sequence | 3 | 1 | 1 | 2 | Harry opens the show referencing a previous conversation on macroeconomic GDP growth from AI tools. Matan collaboratively explains how resource allocation takes time to reflect leverage gains. | |
| Redefining the '10x Engineer' as a Load-Bearing Asset | 4 | 4 | 3 | 3 | Harry cites Andrej Karpathy's framing of 100x engineers, but Matan politely reframes the premise around 'load-bearing individuals' rather than raw line-count metrics. | |
| Building vs. Outsourcing: Kirkland's $500M AI Bet | 5 | 5 | 5 | 3 | Harry raises Kirkland's $500M AI commitment and Brendan's thesis on infrastructure value accrual. Matan strongly disagrees with Brendan, using a lunch-pickup analogy to explain core competencies and commoditization dynamics. | |
| The Factory Bear Case and Rapid Model Development | 4 | 3 | 2 | 2 | Harry probes Factory's bear case and cites rapid open-source releases from Chinese providers. Matan explains why multi-model routing mitigates monopoly risk. | |
| The Three Phases of Enterprise AI & The 'AI Hangover' | 3 | 6 | 2 | 2 | Harry asks whether enterprises naturally default to frontier models for convenience. Matan educates Harry on the three enterprise phases—board panic, token-maxing debauchery, and the current 'AI hangover' where workers ask expensive models about food macros. | |
| Uber's Budget Caps & Nuanced Resource Limits | 5 | 4 | 3 | 4 | Harry calculates Benioff's $300M Anthropic spend relative to dev salaries and pushes Matan to predict developer token spend percentages in three years. Matan resists painting with a broad brush before offering an order-of-magnitude estimate. | |
| The 'Planning vs. Execution' Analogy in AI Workloads | 4 | 5 | 5 | 5 | Harry claims that if 80-90% of tasks shift to open source, it presents a major bear case for frontier coding tools. Matan playfully calls Harry's deduction 'nonsense' and uses an organizational hierarchy analogy to explain decision-making tokens. | |
| Matan's Controversial Opinion: Respecting Sales & Marketing | 5 | 3 | 6 | 5 | Matan attacks the Silicon Valley narrative that research and engineering trump sales and marketing, calling it delusional. Harry instantly counters by pointing out legendary companies that have terrible products but world-class sales teams. | |
| The 'Full-Stack' Engineer of the Future | 6 | 4 | 4 | 5 | Harry explains VC psychology and why investors rely on Math Olympiad credentials as a crutch during periods of existential uncertainty. Matan agrees it is a crutch and explains why self-driven agency matters more. | |
| Banter: The Goodwill Hunting Side-by-Side | 4 | 3 | 1 | 2 | Harry jokes that Matan looks identical to Matt Damon in Good Will Hunting. The conversation turns to full-stack engineers and historical polymaths like Da Vinci and Newton. | |
| Agent Operations & Eliminating Manual Tedium | 4 | 4 | 2 | 2 | Harry asks about emergent agent operations roles and benchmarks against Stripe's documentation quality. Matan explains how agents scale the return on good developer experience. | |
| Building Factories: The Tesla Analogy of Software | 3 | 5 | 4 | 3 | Matan uses a Tesla automated factory analogy to describe software development's future. He delivers an impassioned argument against AI safety pausers, arguing that slowing AI delays treatments for diseases like dementia. | |
| Free Markets, Incentives, and Government Regulation | 5 | 4 | 2 | 4 | Harry pushes Matan on free-market capital allocation versus Adam Smith's invisible hand and state interventions. Matan offers a balanced perspective on fossil fuels, AI speed, and enterprise sales face-to-face dynamics. | |
| The Obsession with Physics and Existential Crisis | 2 | 2 | 2 | 2 | Matan recounts his transition from purchasing math textbooks on Amazon at age 12 to researching string theory at Princeton and Berkeley. Harry lightheartedly teases Matan for taking 12 years to realize his career pivot. | |
| The Three-Hour Walk and Dropping Out | 2 | 2 | 1 | 1 | Matan describes reading Zero to One, emailing Sequoia partner Shaun Maguire based on a shared physics background, taking a three-hour walk on Sand Hill Road, and dropping out of his PhD program. | |
| Pitching Sequoia and the Value of Trust | 5 | 3 | 3 | 6 | Matan explains pitching Sequoia in early 2023 and receiving $1M at a $5M post-money valuation. Harry expresses disbelief that a multi-billion dollar fund partnership held a formal meeting for a $1M check. | |
| Choosing the Right Board and The Danger of Hype | 4 | 3 | 2 | 3 | Matan warns founders about VC salesmanship when a startup is hot. Harry asks directly whether celebrity investors like Ivanka Trump provide tangible business value beyond branding. | |
| Decoupling Models from Applications | 4 | 5 | 3 | 3 | Harry asks how the coding agent market will mature. Matan argues that model providers shouldn't own application layers because API providers have misaligned incentives to maximize token usage. | |
| Model-Agnostic Enterprise Strategies | 5 | 4 | 4 | 5 | Harry challenges Matan's thesis by pointing out the market dominance of Claude Code and Codex, as well as Replit's multi-model execution. Matan distinguishes enterprise CIO procurement from consumer experimentation. | |
| AI Code Security and Geopolitical Risks | 4 | 4 | 3 | 2 | Harry asks about security risks in AI-generated code and the usage of Chinese open-source models by US startups. Matan demystifies trigger-word fears while expressing patriotic frustration over America's lack of open frontier models. | |
| Geopolitics of Energy and AI Infrastructure | 5 | 4 | 4 | 4 | Harry asks about European infrastructure delays, data center pushback, and Nebius vs. CoreWeave. Harry argues why CoreWeave fits Matan's model better, while Matan reiterates his stance against law firms and non-core tech builds. | |
| The Elite Athlete Model: Sleep, Recovery, and Grind Slop | 4 | 5 | 6 | 3 | In a quickfire segment, Matan bluntly states that companies relying on Forward Deployed Engineers have inferior products. He rejects office-bed 'grind slop' culture in favor of Eight Sleep optimization and elite output. | |
| Work-Life Balance: Robustness vs Extreme Optimization | 5 | 4 | 7 | 4 | Harry contrasts American health optimization with European enjoyment. Matan strongly criticizes Dario Amodei and Sam Altman for disingenuously promoting AI job-loss narratives to raise hundreds of billions of dollars. | |
| AI Adoption in Legacy Orgs and The Future of Frontier Models | 3 | 4 | 2 | 2 | Matan reveals Ernst & Young as one of Factory's most agent-native clients and shares how his mind changed regarding multi-model competition. Harry closes the interview with Good Will Hunting jokes. |
Statements from this episode (59)
Grinberg: Future AI tools will enable anyone to build anything
“The world going forward, there is going to be nothing that no one can build.”
Grinberg: Usage of frontier AI models may contract in the short term
“We might see a short-term contraction of usage of the very frontier models.”
Grinberg: Lacking frontier open-source AI models in the US is embarrassing
“I think it's pretty embarrassing that we don't have frontier open models in the United States.”
Grinberg: Top companies will manage teams like Seal Team Six
“We will see the best companies treat teams more and more like seal team six or like professional athletes.”
Grinberg: Token, dollar, and headcount allocation will dominate C-suites within 24 months
“This resource allocation problem of token, it's not just tokens, it's like dollars, it's tokens, it's people. This is, I think, going to be the thing that over the next 24 months, every C-suite is going to be thinking about.”
Grinberg: Corporate bloat was caused by evaluating teams on intermediate metrics
“I think part of the reason why so many organizations got so bloated is because we were in a period of time where everyone was focusing on intermediate metrics.”
Grinberg: Kirkland's $500M AI push will ultimately benefit legal AI vendors
“Now, I actually think this is good for Harvey because 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. …”
Grinberg: Value Accrual in AI Is Dynamic, Not Static
“The reality is value accrual is a time dependent phenomenon. So like, it's not like there is one person who steady state gets all of the value. That's not how it works.”
Grinberg: AI frontier models will remain roughly equivalent to each other
“The bare case against factory is if one model provider gets significantly better than all of the others. So basically I think a key thing for us is that all the models are going to be roughly as good as each other.”
Grinberg: Discrete AI model releases will be replaced by continuous updates
“I think eventually we'll stop seeing them as model releases, and they'll feel more continuous.”
Grinberg: Enterprises will realize frontier AI models are unnecessary for most tasks
“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, and we can do it much faster, much cheaper with these open models.”
Grinberg: User ego drives unnecessary reliance on frontier AI models
“There's kind of an ego thing, where, oh no, no, the work that I'm doing, only a frontier model could handle. Oh, this mere open model can't deal with the work that I'm dealing with. And this is like, even admittedly, when I first started switching over, I'd be…”
Grinberg: Enterprises Entering 'AI Hangover' Over High Costs and Unclear ROI
“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. Does this, like, is this helping our business? That's where a lot of these companies are at now”
Grinberg: Enterprise wasted hundreds of thousands monthly on trivial Opus queries
“One of the CIOs I was speaking with realized we've been spending hundreds of thousands of dollars per month on people asking Opus 4.8 questions like, Hey, how's it going? Like, what should, what are my macros from the food I ate today? Like, what's the weather…”
Stebbings: Uber placed a $1,500 individual budget cap on employee AI spending
“Uber announced last night, I think it was, or yesterday, that they were having like a 1500 dollar budget. Per individual.”
Grinberg: Dozens of Factory clients privately capped employee AI spend
“What's happened with Uber publicly has happened privately with a lot of customers of ours.”
Grinberg: Individual AI token spend will range from 0% to tens of thousands % of salary
“I actually think it can be as low as zero percent for some individuals, and it can be as high as like 1010 of thousands of percent for some individuals.”
Grinberg: Standardizing AI token spend targets across all engineers is a mistake
“In fact, I would argue that 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.”
Grinberg: Median developer AI token spend will match salary within three years
“I would say order of magnitude will probably be comparable to salary.”
Grinberg: Open-source models can execute 80% to 90% of frontier AI tasks
“Probably, like, 80 to 90%. It's typically the planning that really needs the frontier models.”
Grinberg: High-value AI tasks mirror executive decision-making, consuming few tokens
“Oftentimes, leadership makes very key decisions that determine the fate of the company, and they don't spend the most hours. Like, if you look at the human hours of a company, most human hours are not spent on making the decisions. They're on gathering data or…”
Grinberg: Believing a good product sells itself is delusional
“There's a very common Silicon Valley fallacy, which is there's like research is like the pinnacle. And then there's engineers who implement the research, you know, they're not quite there, but you know, they're still great. And then there's sales and marketing…”
Grinberg: AI startups ignoring sales will fail when market gravity returns
“I think that the reality is it will come to haunt some of these companies one day, because I think right now where there's a gold rush and everyone's like desperate to sign, you know, and get more tokens from these people, it's easy. They're kind of like, in m…”
Grinberg: Top AI-era engineers will embrace sales and marketing
“The best engineers are going to be the ones that don't see sales and marketing as dirty work, but as again, an important part of the product, because as an engineer, you're no longer, you know, just your job is ship feature.”
Grinberg: Coding speed and syntax memorization no longer matter for engineers
“And the parts of engineering that become less important are funny enough, the things that the Silicon Valley has really bragged about a lot, which is like, Competition winning or like Olympiad type. Are you like as fast as possible at coding? Are you, do you m…”
Stebbings: VCs use elite founder credentials as a crutch amid AI uncertainty
“Fundamentally, there is intense uncertainty around what Anthropic and OpenAI will do and who they will kill at the application layer. And so in a world where we desperately seek certainty, we look for validators, and the validators of someone being a math Olym…”
Grinberg: Institutional Math Olympiad track records can be an anti-signal in hiring
“Especially there's some high schools that, like, really focus on, like, you must do the math olympic, like, you must do this, the Amy, 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,…”
Grinberg: Software engineers will increasingly transition into full business outcome GM roles
“I think, so it's starting to exist more and more, but I think it's kind of this like GM or general manager Like role for someone who used to be an engineer, where basically you own end to end an outcome that is not just a shipped feature, but like a business o…”
Stebbings: Modern companies now expect every professional function to be full stack
“We're expecting everyone to be full stack in every function.”
Grinberg: Engineers won't manually write documentation or release notes in five years
“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. I think that's something that, you know, we definitely won't do.”
Grinberg: AI documentation equalizes software docs and reduces Stripe's documentation edge
“Yes. But I think Stripe has plenty of places that they can differentiate. And I think it's a better world where everyone has documentation as good as Stripe's.”
Grinberg: Developer experience investments yield 10x to 100x impact with AI agents
“These are all things that the best organizations at like developer experience would invest a lot of resources in, but they would do it because it makes it easier for engineers to work, easier for them to onboard. But the impact of doing that well is just like …”
Grinberg: Software engineers will build automated factories rather than write code
“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.”
Matan Grinberg: AI Labor Displacement Is Short-Term; Long-Term Reallocation Is Net Positive
“Short term, yes. Long term, no. Short term, yes, because it's just a shock to the system where, you know, there are all these big layoffs that are happening that are pretty aggressive, and, you know, these are 1010 of thousands of people that had a job that no…”
Grinberg: Dementia will eventually be solved using AI and software
“Dementia is kind of a go-to example where everyone understands how big of a deal that is. That is something that can be solved with better AI and better software. Like it's a matter of time. Like we will solve it and we can solve it.”
Grinberg: Advocating to slow down AI development is harmful and selfish
“And by saying you want to slow down AI, that's saying like, These people who have relationships with loved ones who have dementia, you're like, no, no, no, sorry, you guys, you gotta maintain that relationship for a little bit longer. We're scared. We don't kn…”
Grinberg: AI infrastructure is not in a long-term bubble
“Maybe there's like some short term blips, but like long term, absolutely not. Like not even close. I think there might be similar corrections to like this thing at Uber, Where, oh, we were going a little haywire. We weren't allocating it appropriately. And the…”
Grinberg: Human behavior change is the top bottleneck for enterprise AI
“I think the biggest bottleneck by far working with all these organizations is the human side of it. It's just like behavior change.”
Matan Grinberg discloses leading Factory is his first-ever job
“So this is the first job I've ever had, which I think is always a funny thing to say. Cause prior to this, I was a theoretical physicist, literally never Never, like, coffee shop, any of that. Literally never have had a job. Like, never have been paid to do an…”
Grinberg: Solving AI code generation requires starting a company, not academia
“Eventually I realized that the way to actually solve this problem was not in academia but in the industry. And to properly solve it in the industry, you'd have to start a company.”
Matan Grinberg Dropped Out of PhD to Pitch Sequoia Partnership
“There was so much momentum, I'm, you know, I, he answered my email, we got along well, I met the co-founder the next day, and I was like, you know what, fuck it. Dropped out, sent him a screenshot, and he was like, alright, you have a meeting with the Sequoia …”
Matan Grinberg: Founders should accept valuation discounts for Sequoia Capital
“So generally, yes. I mean, they're the best firm.”
Grinberg: The Chainsmokers are surprisingly great investors
“They're also incredibly good investors. Incredibly good investors, which sometimes people are surprised by.”
Grinberg: Ivanka Trump provides better hands-on help than traditional VCs
“She is genuinely so kind, so intelligent, and like, people just in, throughout tech, throughout the world, really love her, and for good reason, and she has an incredible network. She's so generous with her time. Like, there is, like, kind of dirty work invest…”
Grinberg: AI models must be decoupled from applications for optimal incentives
“So I think what is necessary for the best outcome for the consumers Is going to be models that are separate from the applications. You, as a consumer, do not want to use applications that are provided for you by the same people that are giving you the model, b…”
Grinberg: Every Enterprise CIO Demands Model-Agnostic AI to Avoid Lock-In
“Every CIO I speak to is really keenly aware of, we cannot, you know, throw our lot in with just one model provider. We're gonna need to be agnostic.”
Grinberg: Enterprise AI Coding Governance Will Be Controlled by Engineers, Not Non-Technical Staff
“I think it would be ill-advised if they were to try and go to the niche of non-technical people writing code for code sake, because I think that is going to be run by, like, if you're going to need enterprise controls over who has access to what databases and …”
Grinberg: Major security incidents from AI code will hit within two years
“I think there are probably going to be in the next couple years some pretty big incidents that occur because of,”
Matan Grinberg: Europe is unlikely to catch up on frontier AI labs
“Probably on the, like, frontier model lab side.”
Grinberg: US states permitting data centers will prosper economically
“There won't be as many jobs that get created there. Whereas the states that do allow for data centers to be good, to be created, you know, people will prosper. They're gonna have great jobs. They'll, you know, see the downstream benefits of it.”
Grinberg: Nuclear power would have given Europe an AI energy advantage
“In Europe, I mean, it's a, it's tough. I think that there were some, there was some good positioning that Europe had, you know, a few years ago, a few decades ago with nuclear that I think hasn't been, you know, delivered on as much as of late, but that would …”
Grinberg: Software requiring forward-deployed engineers to close deals is a bad product
“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 p…”
Grinberg: Mandating extreme hours or office beds signals poor hiring
“I think you are doing a bad job on hiring if you need to, like, mandate certain crazy hours or you need a bed in the office.”
Factory bought $3,000 Eight Sleep Pods for all 30 employees during sprint
“When we were 30, we had, like, a 30 people. We had like a, what we call 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, 3000 dollars per person, like, you know, the…”
Matan Grinberg proposes managing founder performance like athletic seasons
“Looking to athletes, what they do is they have in season and out of season. Maybe it's like, when you're in season, you're fucking locked in, you're not drinking, you're like optimizing all this stuff with your eight sleep, and then, you know, take a week off,…”
Grinberg: Anthropic is a safer IPO investment than OpenAI due to volatility
“In my mind, the answer here is, I think they're approximately equivalent. Like, to me, it doesn't really matter. The biggest reason that affects like the EV is like volatility of the company. That's the only, like, cause I think from the business perspective, …”
Grinberg: AI labor displacement warnings are disingenuous hype designed for mega-fundraising
“I think that has been not only like disingenuine and wrong, but it's like really hurt the psychology of a lot of people, developers, like just people in the world. Does AI a disservice? Does the world a disservice? Because this is again, talking about the use …”
Grinberg predicts at least four AI companies will achieve frontier model parity
“There was a brief period of time where I thought it might be just one or two companies that run away with being kind of the frontier and the best. What seems pretty clear to me is it's probably going to be at least four that are going to probably be approximat…”