Jun 1, 2026 · 1h 14m · 20vc
Mercor CEO: Why Application Layer Companies Have No Moat & The Cost of Hiring AI Researchers · 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 20VC podcast episode, Mercor CEO Brandon Foody addresses public controversies surrounding his multi-billion dollar startup while outlining his perspective on AI labor dynamics, the lack of defensibility in the application layer, and the future of enterprise AI integration.
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 23.3% of the talking time here. How this is scored →
Harry as informed peer 5.4 Guest teaching 6.2 Guest disagreement 2.9 Harry pushing back 5.3
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 55:52 Brandon stands firm on tax policy despite host reaction
Brandon refuses to back down on taxing capital gains over labor income, openly poking fun at Harry's emotional reaction as an investor.
Hardest push from Harry ▶ 55:32 Harry forcefully rejects capital gains tax hikeHarry aggressively pushes back against Brandon's tax policy idea, arguing forcefully that taxing capital gains disincentivizes innovation risk and causes capital flight.
Biggest teaching moment ▶ 42:45 Brandon reveals token spend surpasses employee payrollBrandon stuns Harry by revealing that Mercor spends more money on AI agent inference tokens than on total employee headcount.
Harry holds his own ▶ 35:09 Harry defends application layer moatsHarry leverages deep domain knowledge from his own portfolio companies to mount a detailed defense of specialized enterprise workflows against foundation model takeover.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Myth #1: Addressing the Hack and Net ARR Claims | 2 | 5 | 3 | 4 | Harry directly addresses rumors regarding a security breach and flat revenue. Brandon corrects the record with specific data, stating Mercor added $300 million in net new ARR over the past 60 days. Harry expresses shock at the numbers and asks Brandon how he handles crisis moments. | |
| The Twitter Echo Chamber vs. Reality and Managing Noise | 1 | 6 | 1 | 2 | Harry openly admits ignorance regarding AI cybersecurity threats and asks Brandon to explain how agent swarms work. Brandon explains how automated coding agents exhaustively scan codebases faster than human teams. | |
| Myth #2: Debunking Customer Loss Rumors (OpenAI and Meta) | 5 | 6 | 5 | 6 | Harry presses Brandon on rumors that OpenAI and Meta dropped Mercor, specifically citing Handshake's revenue growth. Brandon firmly denies losing OpenAI, clarifies Meta's status due to Scale AI dynamics, and debunked press reports about poaching Micro One staff. | |
| Myth #4: Amazon’s $13 Billion Acquisition Offer vs. Independence | 6 | 6 | 3 | 5 | Harry asks about a rumored $13B Amazon buy-out offer and brings up high-profile tech layoffs to challenge Brandon's optimism about human labor. Brandon rejects the offer rumor and counters job displacement fears using historical economic productivity data and the lump of labor fallacy. | |
| The Unprecedented Speed of AI Transition and Labor Demands | 6 | 5 | 2 | 7 | Harry explicitly rejects Brandon's historical timeline comparison, arguing that AI adoption accelerates job replacement far faster than previous industrial shifts. Brandon acknowledges displacement speed but points to rapid growth in new agent-training job categories. | |
| Tacit Knowledge: The True Barrier to Enterprise AI Adoption | 4 | 7 | 3 | 5 | Harry asks if data structure and cleanliness are the primary barriers to enterprise AI. Brandon reframes the premise, explaining that AI reasoning handles data cleaning, making unwritten tacit human knowledge the real bottleneck. | |
| Profitability, Cash Assets, and Managing Market Corrections | 5 | 6 | 2 | 5 | Harry probes Mercor's financials, questioning whether reported figures represent real revenue or GMV marketplace volume. Brandon clarifies that Mercor operates at 30-40% gross margins on full-stack deliverables, making it true revenue. | |
| The Power Law of Data Quality and Pricing Moats | 4 | 5 | 1 | 2 | Harry summarizes Mercor's shift toward vertical integration. Brandon details the power law in data quality, where a small fraction of high-quality complex tasks drives the majority of model performance improvement. | |
| Underserved Domains and Recruiting AI-Native Experts | 7 | 5 | 3 | 6 | Harry walks through Mercor's funding rounds, questioning high valuation multiples relative to revenue run-rates. Brandon defends the valuations by showing how rapid revenue growth repeatedly surpassed investor projections. | |
| The Defensibility Dilemma: Application Layer vs. Infrastructure Layer | 6 | 6 | 2 | 4 | Harry cites Brandon's tweet predicting infrastructure superiority over application layer startups and brings up Nebius price increases. Brandon elaborates on why software layers lack moats when frontier models easily absorb application logic. | |
| The Debate: SaaS Moats, Re-Created Software, and Network Effects | 8 | 7 | 5 | 8 | Harry strongly defends application layer startups using his legaltech investments, citing specialized workflows and go-to-market defensibility. Brandon counters that models are the primary product and will clone end-to-end SaaS applications within 12 months. | |
| Go-To-Market vs. Forward Deployed Operations & AI-Enabled Services | 7 | 7 | 4 | 8 | Harry challenges Brandon again by contending that go-to-market relationships form the core product moat. Brandon reframes GTM into forward-deployed operations and reveals Mercor is already replacing operational delivery teams with autonomous AI project managers. | |
| Inference Economics: Jevons’ Paradox & Enterprise Token Spend | 4 | 8 | 2 | 4 | Harry asks about unit economics and token costs. Brandon shares a striking stat: Mercor now spends more money on internal agent tokens than on entire human payroll, illustrating Jevons' paradox in real time. | |
| The Commoditization of the API Layer and Model Evaluations | 7 | 7 | 3 | 4 | Harry brings enterprise data points on Salesforce's Anthropic spend. Brandon argues that zero switching costs will completely commoditize model API layers, forcing enterprises to rely on custom evals to hot-swap providers. | |
| Redefining Evaluations: Measuring Real-World Workflows | 6 | 6 | 1 | 7 | Harry criticizes present-day AI evaluation benchmarks as impractical and disconnected from reality. Brandon agrees, explaining the industry shift toward evaluating real-world complex workflows. | |
| The Future of NVIDIA and Chip Hardware Competition | 8 | 6 | 6 | 9 | Brandon advocates eliminating income taxes for the bottom 50% of earners and raising capital gains tax. Harry forcefully pushes back against raising capital gains tax, arguing it penalizes risk-taking investors and drives capital abroad. | |
| Meeting Tech Giants and Overcoming Imposter Syndrome | 6 | 6 | 3 | 5 | Harry asks about Europe's position in foundation models and sovereign AI arguments. Brandon candidly advises Europe to accept defeat in foundation models due to talent cluster network effects in the US. | |
| The Escalating Cost of Hiring Top-Tier AI Researchers | 5 | 7 | 2 | 4 | Harry probes the talent war for top AI talent. Brandon reveals Meta's superintelligence group is extending liquid annual compensation offers exceeding $20 million to top researchers. | |
| Scaling Operations, HR Challenges, and Work Culture | 6 | 6 | 4 | 6 | Harry brings up his controversial tweet stating great CEOs dislike HR departments. Brandon dissents, explaining that proper HR infrastructure is vital when scaling headcount rapidly from 40 to 400 employees. |
Statements from this episode (62)
Mercor spends more on AI agent tokens than employee headcount
“Right now, we're spending more on tokens for our internal agents than we are on employee headcount.”
Foody: Top AI researchers cost tens of millions in stock annually
“Oftentimes, it would be in the tens of millions of stock per year.”
Foody: Twitter perception of Mercor breach was far worse than internal reality
“There was this broad perception on Twitter that was much more exaggerated than what actually happened within the business.”
Foody: Investor Spread False Rumors of China Accessing Mercor Data
“Think of one person that's very prominent who's invested in multiple competitors and just, like, made this tweet about how all of our data was getting accessed by China when it was totally untrue.”
Mercor Attacker Used Swarm of AI Coding Agents to Breach Systems
“Because in our incident, it was the attacker that used a swarm of coding agents to help get access to the system as is happening in a lot of these.”
Foody: Enormous boom coming in AI security engineering tools
“And so I think there's going to be An enormous boom in AI security engineering tools and various forms of defense that are able to help protect companies against all of the increasing waves of cyber incidents that are just getting started.”
Foody: Mercor did not lose OpenAI as a customer after hack
“False. Our relationship with OpenAI is stronger than ever.”
Meta Paused Mercor Relationship After Hack While Other Labs Expanded
“I mean, I think that Meta you know, like, currently the relationship is still paused. Every other one of the Frontier Labs has grown their relationship with us since and the company has been crushing it but they're the only one that is paused, which is public.”
Foody Rejects $30B Sale Despite Receiving Significant Acquisition Interest
“No, I wouldn't. I mean, ultimately we've gotten a lot of acquisition interest and we could walk away with, like I could walk away with billions of dollars in cash.”
Foody: There will be significantly more jobs in 10 years
“I believe there's certainly going to be many more jobs in 10 years than there are today.”
Frontier AI Task Automation Rose From 1% to 40% in 12 Months
“On Apex, the Frontier model right now is at about 40%, and 12 months ago the Frontier model was a one, which was scoring one percent.”
Foody: People systematically underestimate demand elasticity from productivity gains
“Everyone underestimates the elasticity for demand and increased productivity in the economy.”
Foody: Mercor is paying out over $3 million daily
“Now we're paying out over three million dollars a day and the fastest job category ever created in history.”
Foody: Mercor daily payouts will triple to $9M in 12 months
“In 12 months time, that's probably about triple that.”
Foody: Training AI agents will become a massive new job category
“One of the largest things that people underestimate both in the context of AI labs as well as within the enterprise is how significant of a job category it is going to be to train agents.”
Foody: All knowledge work is converging on training AI agents
“All knowledge work is converging on training agents because it is structurally more efficient to do something once.”
Foody: AI models will clean enterprise data as reasoning capabilities rise
“They'll be able to clean the data themselves fairly effectively as reasoning capabilities go up.”
Foody: Unwritten human context is the key bottleneck to AI agent automation at Mercor
“When I try to get agents to do all of these workflows throughout Mercure, there's just an enormous amount of context that lives in people's heads that the agents need to have access to perform effectively.”
Mercor Straps Cameras to Electricians and Scientists for AI Training Data
“We're doing a ton of data collection in the physical world as well, especially across skilled domains where you have electricians and mechanics and scientists strapping cameras to their head to record things.”
Foody: Mercor Has a Talent Network of Over Five Million People
“When we have this talent network of over five million people that are able to refer their friends, it's just so much easier for us to find the marginal doctor because we have that enormous talent network that can refer us to their friends.”
Foody: AI Labs Prefer Horizontal Data Vendors Over Niche Vertical Players
“We are finding that the labs tend to prefer partnering with a very horizontally capable vendor that is able to flex across all of the different Verticals and scale extremely quickly rather than working with a hundred different vendors that they have to train f…”
Foody discloses Mercor holds over $500M in cash and is highly profitable
“And so we view having over five hundred million in cash and a super profitable business as a significant asset and allowing us to be prepared for when there is a market correction to make sure that we consolidate market share.”
Mercor Burned Only $500K Post-Seed and Holds More Cash Than Raised
“The, we burnt a half a million dollars after our seed round. And then from there, we've pretty much been profitable ever since. We have more cash than we've ever raised.”
Foody: Mercor operates at 30% to 40% gross margin
“So the revenue is between a 30 and 40% gross margin, but the key distinction and why it's not GMV, but it's revenue, is that the experts are actually only one part of the broader value chain that we deliver to customers.”
Foody: Top 20 Percent of AI Tasks Generate Majority of Model Value
“And there's oftentimes this very power law nature of data that drives model improvement in that out of a data set of 10,000 tasks, the top 2000 tasks will create majority of the value.”
Foody: Data Quality Gives AI Data Vendors Strong Pricing Power
“And so it allows vendors that are extremely high quality to be super differentiated in so far as pricing power, because quality is the X factor that becomes dramatically more valuable than any other dimension.”
Foody: AI models will execute complex multi-week long-horizon tasks within 6-12 months
“Those are the kinds of tasks that we need to be building to push the frontier of research and evaluation so that those are the capabilities that people are able to use in the models in six to 12 months.”
Foody: AI data training requires domain experts who are AI power users
“And it's more about people that actually are very acclimated to the frontier of AI, because it's the people that understand, that both have the expertise in oncology, but also are power users of ChatGPT or Claude that are able to find where the model makes mis…”
Foody: Millions of people will soon be hired to train AI on workplace tasks
“And so there's this enormous mobilization of hundreds of thousands and soon millions of people to build out the full distribution of everything that you could pass into Google Workspace and everything that you could want out on the other side in every job cate…”
General Catalyst issued Mercor's seed term sheet at $23M valuation
“They gave us a term sheet within 36 hours for 2.3 million dollars at a twenty three million dollar post-money valuation.”
Felicis Offered Mercor $2B Valuation at $20M ARR
“We're at twenty million in revenue. They ask us, what valuation do we think makes most sense? And I say one to two billion dollars. So they give us a term sheet at a two billion dollar valuation.”
Felicis Offered Mercor $10B Valuation at $400M ARR
“By September of 20, 25 or say October, we were at called four hundred million in revenue run rate. And then Felisa's was like, we want to invest more. And so they gave us a term sheet at a ten billion dollar valuation.”
Application Layer AI Startups Will Struggle to Build Defensible Moats
“And so I feel like building defensibility in the software layer On top of the models is going to be incredibly difficult. Whereas on the other side of things and the infrastructure side, it feels like there are meaningful moats that are getting built.”
Foody: AI infrastructure companies will achieve higher margins than application startups
“And so I think that There are going to be high margins that get achieved at the infrastructure layer and sort of sustainable, profitable businesses in a way that it's less immediately clear at the application layer.”
Foody: Mercor has demand to double overnight but lacks capacity
“Like we have the demand to Double overnight. We just don't have the capacity.”
Foody: Mercor could raise prices 30% without impacting customer demand
“We maybe can't double prices. We could double capacity. We could probably increase prices by 30% without much of an impact.”
Foody: Mercor building benchmark to test AI cloning of SaaS apps
“We're building out an eval set that measures how effectively agents can build end-to-end SaaS applications, where twenty-twenty-five was the year of how do you get a model to make a PR and a code base, and twenty-twenty-six is the year of how do you get the mo…”
AI Models Will Clone End-to-End SaaS Apps Within 12 Months
“Those capabilities are going to exist in the models in the next 12 months.”
Foody: Software companies lacking network effects will struggle due to AI
“The companies that don't have network effects are going to struggle very significantly because then there's not really a defensible moat in the pure software associated with the products that they build.”
Foody: Software moats are eroding while layered services create defensibility
“These software boats are whittling away and it's the ability to layer services on top of software to meet the customer where they're at and go the last mile that is creating stronger defensibility.”
Mercor AI Agent Autonomously Managed an Entire Data Annotation Project
“Now we have an AI project manager that just completed its first project managing that entire thing end to end, where it's able to hire the experts. It's able to answer their questions. It's able to build the annotation tool using its coding tools within our pl…”
Foody: Most businesses will spend more on AI tokens than headcount
“And I think most businesses are going to look like that.”
Foody: Mercor's AI agent has conducted over 5M interviews
“We have our interview question agent that where we've done over five million interviews and ask all the questions in the interviews.”
Foody: Enterprises will build evaluation systems to commoditize AI model providers
“And I believe that over time, this is going to develop to look very similar across every fortune, where they'll need to have this system of record for evaluating and specifying agent behavior across every workflow in their business. And they're going to use th…”
Foody: The AI model API layer will get commoditized
“So I think the key distinction is that I think the API layer will get commoditized.”
Foody: Enterprise AI API switching costs are zero due to continuous benchmarking
“Because the switching costs are zero. Like when the switching costs are zero, that means that, and there's a new frontier model every two months. That means that we very quickly are going to swap them out.”
Enterprises Will Spend More on Compute Than Headcount Within Five Years
“I don't know about 24 months time, but I would bet that in five years the average enterprise spends more on compute than headcount.”
Foody: Academic AI Benchmarks Are Disconnected From Enterprise Needs
“We used to have this paradigm of all of the academic benchmarks that were totally disconnected from the outcomes that enterprises actually care about, where people were building everything ranging from GPQA for PhD level reasoning to IMO for Olympiad math to H…”
Foody: AI Model Evaluation Focus Is Shifting to End-to-End Workflows
“And now they're focused on how do we get the model to do this end-to-end workflow, coordinating with multiple colleagues for a financial model or a slide deck like we were discussing. How do we get the model to build an entire SaaS application end-to-end? And …”
Majority of AI Inference Will Run on Open-Source or Distilled Models
“I think that majority of inference in five years is going to be using a open source or custom fine-tuned or distilled model, not using a frontier model.”
OpenAI or Anthropic Will Exceed $10 Trillion Valuation Within Five Years
“I could definitely see one of them being a 10 trillion dollar company maybe even significantly higher. It feels Like the opportunity associated with being the frontier model is so large that it will just like eat up so much of the other demand within the econo…”
Foody: Most frontier AI labs are building in-house chips
“Most of the labs are building in-house chips.”
Foody: NVIDIA will lose its monopoly in five years but remain most valuable
“And so I would guess that in five years, it doesn't feel like Nvidia has quite the same monopoly, but that's okay, because even if they only have 30 or 40% market share in the largest market in the world by far, that is the world's most valuable company.”
Foody: Jeff Bezos Retweeted Proposal to Eliminate Bottom-Half Income Tax
“Jeff Bezos retweeted me, which I was ecstatic about.”
Foody: US AI talent aggregation will continue driving major geopolitical advantage
“I think that it's going to be difficult to change because there's just so many strong network effects around talent, right? When we have the best talent, even I know so many Brilliant French researchers that go to work at OpenAI and dropping in DeepMind, right…”
Foody: Europe should concede foundation AI model race to US tech giants
“I would accept that, yeah. I think that maybe it's worth having some post training capabilities because there is going to be value to distillation and some of the work that happens after foundation models are built, and there's definitely going to be some valu…”
US AI Labs Will Hire 10,000 French Lawyers to Defeat Sovereign Models
“That said, the labs are just going to hire 10,000 people in France to teach the models how to be better at French law. And I don't think that there's so much that others are going to be able to do to stop that because The transfer learning capabilities from al…”
Foody: Three former Mercor employees founded $100M+ startups
“We've had Three employees that have founded companies worth in excess of a hundred million dollars.”
Meta Offered an AI Candidate $20 Million Per Year in Cash
“There was someone I was hiring the other day, and he had an offer for twenty million dollars in cash per year from TBD, and like that's the kind of stuff we run into on a regular basis.”
Foody: AI researcher compensation will normalize as training skills spread
“So I think that it'll probably continue to escalate for the people that, for a smaller group of people. But I also suspect that as more people gain knowledge of how these labs operate and what the capabilities of how to train a frontier model That means that t…”
Shorts cut from this episode
▶ "The Model is the Product" · 20VC with Harry Stebbings (@0:21)
▶ The $10B Startup Running on AI Agents · 20VC with Harry Steb (@0:43)
▶ Why AI Won't Take Your Job · 20VC with Harry Stebbings (@12:59)
▶ AI Is Creating Jobs Faster Than Ever · 20VC with Harry Stebb (@14:15)
▶ Why we should increase capital gains tax · 20VC with Harry S (@54:54)
▶ "We spend more on tokens than salaries" · 20VC with Harry St (@0:00)