Lenny RAG MCP Server

{ "episode": { "guest": "Brendan Foody", "expertise_tags": [ "AI model evaluation", "labor marketplaces", "reinforcement learning", "post-training data", "founder/CEO", "hypergrowth scaling", "talent acquisition", "expert networks" ], "summary": "Brendan Foody, CEO of Mercor, discusses the explosive growth of his company from $1 to $400M revenue run rate in 16 months—the fastest ascent in history. Mercor connects AI labs with expert professionals (engineers, lawyers, doctors, analysts) to create evaluations and training data for model improvement. The episode explores why evals have become the new bottleneck in AI development, how the labor market is transforming with AI, what jobs will remain valuable, and the core principles that enabled hypergrowth: leading indicators in fast-moving markets, customer obsession, and three key values: can-do attitude, high standards, and intensity.", "key_frameworks": [ "Evals are the PRD for AI models", "Reinforcement learning from AI feedback (RLHF) over human feedback", "Market bound by capabilities humans can do that models cannot", "Elastic demand industries vs. inelastic demand", "Top 10% of talent drives majority of impact", "Leading indicators in fast-moving markets", "Customer obsession over premature marketing", "Hiring intensity: be patient until product-market fit, then scale aggressively" ] }, "topics": [ { "id": "topic_1", "title": "The Era of Evals and Why It Matters", "summary": "Introduction to evals as the critical bottleneck in AI development. If the model is the product, the eval is the PRD. Labs need evals to measure success and improve capabilities across every domain.", "timestamp_start": "00:00:00", "timestamp_end": "00:09:26", "line_start": 1, "line_end": 87 }, { "id": "topic_2", "title": "Understanding the Landscape: Data Labeling Companies and Market Transition", "summary": "Overview of the three buckets of fast-growing companies: foundational models, vibe coding apps, and data/evals companies. Discussion of the shift from crowdsourcing low-skilled workers to sourcing expert professionals.", "timestamp_start": "00:09:26", "timestamp_end": "00:12:21", "line_start": 88, "line_end": 110 }, { "id": "topic_3", "title": "What Experts Actually Do: The Concrete Work of Model Evaluation", "summary": "Practical examples of what professionals do when creating evals. Case study of lawyers creating rubrics for contract review. Explanation of how humans define success criteria that models learn from.", "timestamp_start": "00:12:21", "timestamp_end": "00:16:24", "line_start": 111, "line_end": 135 }, { "id": "topic_4", "title": "The Future of Work: How Long Will Humans Be Needed?", "summary": "Brendan's perspective that we won't see superintelligence in 3 years. The market is bound by things humans can do that models cannot. Evals will be needed for years as models improve across every domain.", "timestamp_start": "00:16:24", "timestamp_end": "00:20:22", "line_start": 136, "line_end": 171 }, { "id": "topic_5", "title": "Skills Worth Investing In: Elasticity, Leverage, and AI Fluency", "summary": "Advice on what skills remain valuable. Focus on domains with elastic demand (software, product management) rather than inelastic (accounting). Key skill: ability to leverage AI to do 10x more work in your field.", "timestamp_start": "00:20:22", "timestamp_end": "00:24:20", "line_start": 172, "line_end": 198 }, { "id": "topic_6", "title": "Labor Markets and the Inefficiency of Manual Matching", "summary": "Historical context of how labor markets have been inefficient. Mercor's founding thesis: automating resume review and interview processes to create a global unified labor market with perfect information flow.", "timestamp_start": "00:24:20", "timestamp_end": "00:28:56", "line_start": 199, "line_end": 225 }, { "id": "topic_7", "title": "Post-Training vs. Pre-Training: How Humans Shape Model Behavior", "summary": "Explanation of how pre-training loads knowledge into models, while post-training (with human input) teaches reasoning and accuracy. Real-world example of radiologists improving medical diagnosis capabilities.", "timestamp_start": "00:28:56", "timestamp_end": "00:32:35", "line_start": 226, "line_end": 251 }, { "id": "topic_8", "title": "The Competitive Landscape: Scale, Surge, and the Expert Network Transition", "summary": "Comparison between traditional expert networks (AlphaSights, GLG) and new eval-focused companies. Key difference: Mercor focuses on longer-term projects rather than one-off calls, with better retention and incentives.", "timestamp_start": "00:32:35", "timestamp_end": "00:34:51", "line_start": 252, "line_end": 270 }, { "id": "topic_9", "title": "Mercor's Speed and Top-Talent Concentration", "summary": "How Mercor can hire Emmy-winning screenwriters within 24 hours. Data showing top 10% of talent drives majority of impact. Proprietary advantage in identifying and matching top performers.", "timestamp_start": "00:34:51", "timestamp_end": "00:37:08", "line_start": 271, "line_end": 291 }, { "id": "topic_10", "title": "Compensation and Talent Economics: From Crowdsourcing to Premium Labor", "summary": "Mercor's median pay of $95/hour, scaling to $500/hour for deep expertise. Contrast with crowdsourcing companies paying $30/hour. Economics of hiring Goldman Sachs bankers vs. undergrads.", "timestamp_start": "00:37:08", "timestamp_end": "00:38:42", "line_start": 292, "line_end": 314 }, { "id": "topic_11", "title": "Finding the Biggest Business Opportunity: The Origin Story", "summary": "How Brendan discovered the market opportunity. Meeting xAI co-founders while in college. The pivotal moment when a crowdsourcing partner hired 1,000 people who didn't get paid, revealing market gaps.", "timestamp_start": "00:38:42", "timestamp_end": "00:45:50", "line_start": 315, "line_end": 361 }, { "id": "topic_12", "title": "Core Tenets of Hypergrowth: Can-Do Attitude, High Standards, Intensity", "summary": "Three foundational values that drove Mercor's success. Can-do attitude: setting 50x revenue goals and hitting them. High standards: hiring former founders and execs. Intensity: output-oriented culture inspired by Meta and Google.", "timestamp_start": "00:45:50", "timestamp_end": "00:49:30", "line_start": 362, "line_end": 382 }, { "id": "topic_13", "title": "Hiring Philosophy: When to Move Slowly vs. Quickly", "summary": "Extreme patience with first 10 hires, selecting only extraordinary candidates. After proving product-market fit, switch to faster hiring. Brendan acknowledges they may have moved too slowly scaling from 10 to 100.", "timestamp_start": "00:49:30", "timestamp_end": "00:52:20", "line_start": 383, "line_end": 402 }, { "id": "topic_14", "title": "CEO Time Allocation: Hiring, Customers, and Organization Building", "summary": "Where Brendan actually spends his time as CEO: working on hiring and building the team, deep customer relationships, setting compensation levels, and managing organizational processes.", "timestamp_start": "00:52:20", "timestamp_end": "00:53:27", "line_start": 403, "line_end": 408 }, { "id": "topic_15", "title": "Early Entrepreneurship: Donut Dynasty and Lessons Learned", "summary": "Brendan's first business in middle school. Buying Safeway donuts for $5/dozen, selling for $2 each. Competitive response to Chuck's Donuts. Hiring friends and paying them in perceived-value donuts. Core lesson: you can just do things.", "timestamp_start": "00:53:27", "timestamp_end": "00:55:51", "line_start": 409, "line_end": 420 }, { "id": "topic_16", "title": "Evals as Evergreen: The Long Road to AGI and Superintelligence", "summary": "Brendan's perspective: superintelligence isn't coming in 3 years. Models will automate majority of knowledge work in 10 years. The path is paved with thoughtful post-training datasets, not 10x more pre-training data.", "timestamp_start": "00:55:51", "timestamp_end": "01:00:06", "line_start": 421, "line_end": 473 }, { "id": "topic_17", "title": "AI Corner: Personal Use of AI Tools", "summary": "Brendan's personal AI usage: ChatGPT Voice Mode for writing documents and thinking through problems. Uses AI as a thought partner and reasoning tool.", "timestamp_start": "01:00:06", "timestamp_end": "01:02:25", "line_start": 474, "line_end": 525 }, { "id": "topic_18", "title": "Dyslexia as Strength in Entrepreneurship", "summary": "Brendan's dyslexia makes email management challenging but enhances creative thinking and market perception. Management philosophy: leverage people's strengths rather than fixing weaknesses.", "timestamp_start": "01:02:25", "timestamp_end": "01:05:53", "line_start": 526, "line_end": 587 }, { "id": "topic_19", "title": "Lightning Round: Books, Entertainment, Products, and Life Motto", "summary": "Book recommendations: High Output Management, Zero to One, Shoe Dog. Recent entertainment: Oppenheimer. Favorite product: Codex. Life motto: You can just do stuff. Can-do attitude.", "timestamp_start": "01:05:53", "timestamp_end": "01:06:37", "line_start": 588, "line_end": 593 } ], "insights": [ { "id": "i1", "text": "If the model is the product, then the eval is the product requirement document.", "context": "Evals define what success looks like for models, just as PRDs define product requirements for software. Researchers run dozens of experiments improving eval scores through reinforcement learning.", "topic_id": "topic_1", "line_start": 70, "line_end": 71 }, { "id": "i2", "text": "Evals are not just benchmarks—they're sales collateral, PRDs, and training signals all at once.", "context": "Labs use evals to show researchers what to build, demonstrate model efficacy to customers, and reward model behavior during post-training. The semantic distinction between 'evals' and 'RL environments' is less important than understanding they measure what good looks like.", "topic_id": "topic_1", "line_start": 82, "line_end": 126 }, { "id": "i3", "text": "The market is bound by the amount of things where humans can do something that models can't.", "context": "This is the ultimate constraint on Mercor's addressable market and the long-term viability of the eval business. As long as there are human capabilities that models lack, there will be demand for experts to define and measure progress toward those capabilities.", "topic_id": "topic_4", "line_start": 115, "line_end": 116 }, { "id": "i4", "text": "The entire economy will likely become an aural environment machine, building worlds and contexts for models to learn from.", "context": "Rather than job displacement being the primary narrative, a new category of jobs is being created: people who design evaluation environments, measure model capabilities, and define what good looks like across every domain.", "topic_id": "topic_4", "line_start": 167, "line_end": 171 }, { "id": "i5", "text": "Jobs with elastic demand will be most valuable. Software development has nearly unlimited demand when productivity increases. Accounting does not.", "context": "When we make people 10x more productive, we'll build 10x more software, but we won't necessarily need 10x more accounting. Individuals should focus careers on domains where increased productivity drives increased demand.", "topic_id": "topic_5", "line_start": 188, "line_end": 191 }, { "id": "i6", "text": "The skill that matters most is not a specific technical ability, but the ability to leverage AI tools to do your work better.", "context": "Just as calculators didn't eliminate math—they changed what math meant—AI won't eliminate jobs in elastic industries. It will eliminate people unwilling to use AI. Use ChatGPT, Claude, Cursor, etc. to build 10x faster and better.", "topic_id": "topic_5", "line_start": 176, "line_end": 182 }, { "id": "i7", "text": "Labor markets have been wildly inefficient because matching candidates to jobs was done manually at scale.", "context": "Companies in Silicon Valley review a fraction of available candidates globally. When you automate the matching problem (resume review, interviews, hiring decisions), you enable a truly global, frictionless labor market.", "topic_id": "topic_6", "line_start": 214, "line_end": 215 }, { "id": "i8", "text": "Pre-training loads knowledge; post-training teaches the model what's correct, what's incorrect, and what to prioritize.", "context": "When ChatGPT diagnoses an X-ray, radiologists aren't just in the training data. They're in the post-training phase, defining ground truth, calibrating rewards, and teaching the model to reason correctly about complex medical cases.", "topic_id": "topic_7", "line_start": 235, "line_end": 236 }, { "id": "i9", "text": "Reinforcement learning from AI feedback (RLHF) is more scalable and data-efficient than supervised fine-tuning or traditional RLHF from human preference labels.", "context": "If you define success criteria (rubric for lawyers, unit tests for code), you can automatically measure whether the model achieved it, scaling evaluation without proportional scaling of human labor.", "topic_id": "topic_7", "line_start": 134, "line_end": 135 }, { "id": "i10", "text": "The top 10% of talent on your team will drive the majority of your impact.", "context": "Just like in a 100-person company, in a 100-person eval cohort, the top performers drive disproportionate model improvement. Building proprietary advantage in identifying and retaining these people is defensible and valuable.", "topic_id": "topic_9", "line_start": 280, "line_end": 284 }, { "id": "i11", "text": "Expert compensation scaling ($95/hr median to $500/hr) reflects the actual value created, not arbitrary market rates.", "context": "Goldman Sachs bankers and McKinsey analysts have different and more valuable capabilities than undergraduates. Labs pay for those capabilities directly. This is a fundamental shift from low-wage crowdsourcing.", "topic_id": "topic_10", "line_start": 295, "line_end": 296 }, { "id": "i12", "text": "Focus on leading indicators in fast-moving markets, not legacy markets with incumbents.", "context": "The AI labs market was moving fast and pricing in unlimited value. The legacy hiring market had stagnant economics. By focusing on leading indicators of emerging demand, Mercor captured extraordinary growth.", "topic_id": "topic_12", "line_start": 332, "line_end": 333 }, { "id": "i13", "text": "Customer obsession beats premature marketing. With no sales or marketing team, Mercor relied on word-of-mouth from delighted customers.", "context": "When you build six-star customer experiences and your customers are the smartest AI researchers in the world, they become your distribution channel. This is more efficient than sales teams.", "topic_id": "topic_12", "line_start": 335, "line_end": 336 }, { "id": "i14", "text": "Be patient hiring your first 10, disciplined until you prove product-market fit, then accelerate dramatically.", "context": "Early team density shapes the entire organization. Brendan hired Scale's head of growth as employee 2, but may have moved too slowly from 10 to 100 people when he should have scaled faster.", "topic_id": "topic_13", "line_start": 385, "line_end": 402 }, { "id": "i15", "text": "Don't teach students and professionals to avoid AI. Teach them to leverage it and see how much they can accomplish.", "context": "Just as calculators changed what math education meant (deeper reasoning, less arithmetic), AI should change how we assess talent. Evaluate what people can build with AI tools available, not ability to work around them.", "topic_id": "topic_5", "line_start": 179, "line_end": 182 }, { "id": "i16", "text": "Superintelligence in 3 years is unlikely. We'll automate majority of knowledge work in 10 years, but the path is paved with post-training data sets, not 10x more pre-training.", "context": "Scaling laws have plateaued. Further progress requires thoughtful, high-quality post-training data, not just throwing more pre-training data at the problem. This makes expert evals evergreen.", "topic_id": "topic_16", "line_start": 466, "line_end": 467 }, { "id": "i17", "text": "Founders often try to force product-market fit instead of listening to market signals about what customers actually want.", "context": "Brendan had conviction about labor marketplaces, but was willing to pivot when he saw the bigger opportunity in AI evals. Persistence + openness to market signals beats stubbornness.", "topic_id": "topic_11", "line_start": 352, "line_end": 354 }, { "id": "i18", "text": "The difference between traditional expert networks and modern eval platforms is project length and retention. Labs need ongoing relationships, not one-off calls.", "context": "AlphaSights built a great business on one-off expert consultations. But labs need sustained collaboration to define evals, iterate on rubrics, and improve models over months. Different business model entirely.", "topic_id": "topic_8", "line_start": 256, "line_end": 257 }, { "id": "i19", "text": "AI won't replace you. People good with AI will replace you.", "context": "This isn't about AGI or superintelligence coming. It's about competitive advantage in 2025. If your competitor is 2x faster and 2x better because they leverage AI and you don't, they win.", "topic_id": "topic_5", "line_start": 206, "line_end": 209 }, { "id": "i20", "text": "Models are only as good as their evals.", "context": "This is a quote Brendan attributes to customers. It's the fundamental insight underlying Mercor's entire business: improving evals improves models. Everything else is secondary.", "topic_id": "topic_10", "line_start": 314, "line_end": 314 }, { "id": "i21", "text": "Large enterprises often resist evaluating whether their value chain is being automated because it threatens them. But the companies that lean into it win.", "context": "Fear-based thinking ('if we eval our business, we'll see it's being automated') prevents companies from preparing. Winners ask: if we can do 10-100x more, what will we build?", "topic_id": "topic_5", "line_start": 209, "line_end": 209 }, { "id": "i22", "text": "Hire for strengths, not weaknesses. Some things you'll never be best at, so hire around them.", "context": "Brendan's dyslexia makes email management hard. Instead of trying to get better at email, he hires people who are great at communication and delegation. This philosophy scales across organizations.", "topic_id": "topic_18", "line_start": 584, "line_end": 585 }, { "id": "i23", "text": "ChatGPT Voice Mode is exceptionally useful for thinking through problems as a thought partner.", "context": "Brendan uses voice mode to reason through complex business problems. The conversational interface makes it feel like having an always-available advisor. This is a profound use case beyond document generation.", "topic_id": "topic_17", "line_start": 485, "line_end": 497 }, { "id": "i24", "text": "The narrative around AI has been entirely job displacement, but very few people talk about the new category of jobs being created.", "context": "The eval economy, the training data economy, the AI infrastructure economy—these are creating massive new job categories. Reframing from 'jobs lost' to 'jobs created' is both more accurate and more helpful.", "topic_id": "topic_4", "line_start": 170, "line_end": 171 } ], "examples": [ { "id": "e1", "explicit_text": "We grew from 1 to 400 million in revenue run rate in 16 months, fastest ascent in history.", "inferred_identity": "Mercor (founded January 2023, exited to $2B valuation with $100M funding in 17 months)", "confidence": "explicit", "tags": [ "Mercor", "hypergrowth", "B2B SaaS", "AI evals", "fastest-growing company", "revenue growth", "series A funding", "founder achievement" ], "lesson": "Massive market opportunities exist at the intersection of emerging technology bottlenecks and shifting customer needs. By focusing entirely on solving the biggest pain point for the most sophisticated customers, companies can grow orders of magnitude faster than traditional models.", "topic_id": "topic_2", "line_start": 14, "line_end": 14 }, { "id": "e2", "explicit_text": "We started the company together when we were 19 initially, in January 2023, initially hiring people internationally, matching them with our friends and automating all the processes of how we did that.", "inferred_identity": "Mercor, co-founded by Brendan Foody with college friends", "confidence": "explicit", "tags": [ "Mercor", "young founders", "labor marketplace", "automation", "bootstrapped", "college dropout", "founder story" ], "lesson": "Experienced founders often automate manual processes they've done themselves. Brendan and his friends manually hired and matched international talent, realized the bottleneck, and built a business to solve it at scale.", "topic_id": "topic_2", "line_start": 92, "line_end": 92 }, { "id": "e3", "explicit_text": "We met OpenAI and we saw that there was this enormous transition in the human data market where it was moving away from this crowdsourcing problem...to this sourcing and vetting problem. How do we source and assess the best professionals, the experienced?", "inferred_identity": "OpenAI (and implicit meeting with other AI labs)", "confidence": "explicit - customer interaction", "tags": [ "OpenAI", "AI labs", "market transition", "data labeling", "crowdsourcing vs expert networks", "customer insights", "product pivot" ], "lesson": "Meeting with your future customers before you've fully built the product teaches you what they actually need. OpenAI taught Mercor that the bottleneck was quality and expertise, not volume.", "topic_id": "topic_2", "line_start": 95, "line_end": 95 }, { "id": "e4", "explicit_text": "Say you have a model that you want to write a red line for a contract in the way that a lawyer would, and it makes a handful of mistakes, misses a bunch of key points in doing so. What you could do is have a lawyer create a rubric similar to how a professor might create a rubric.", "inferred_identity": "Composite example (could be from Anthropic, OpenAI, or other labs working on legal AI)", "confidence": "composite/illustrative", "tags": [ "legal AI", "contract review", "rubric creation", "eval definition", "professional expertise", "model evaluation", "domain-specific evals" ], "lesson": "Experts don't need to write perfect training data. They need to define clear success criteria (rubrics) that models can optimize toward. This is more scalable than generating perfect outputs.", "topic_id": "topic_3", "line_start": 116, "line_end": 119 }, { "id": "e5", "explicit_text": "We hired all the people from the Harvard Lampoon a couple of months ago, their comedy club, to help with making models funnier.", "inferred_identity": "Mercor hiring for unnamed AI lab (likely OpenAI given context of conversation)", "confidence": "explicit - Brendan personally hired them", "tags": [ "Mercor", "Harvard Lampoon", "comedy writers", "creative AI", "humor generation", "talent recruiting", "non-technical evals", "soft skills" ], "lesson": "AI model improvement isn't just about technical tasks. Creative capabilities like humor, voice, and entertainment value require experts in those domains, not just engineers.", "topic_id": "topic_8", "line_start": 269, "line_end": 269 }, { "id": "e6", "explicit_text": "We hired Emmy award-winning screenwriters and everything across the board on creative capabilities that you'd look for.", "inferred_identity": "Mercor hiring for AI labs (unnamed but likely OpenAI or Anthropic)", "confidence": "explicit - hiring by Mercor", "tags": [ "Mercor", "screenwriters", "Emmy winners", "creative professionals", "entertainment", "content generation", "talent networks" ], "lesson": "The future of AI training requires experts across every human domain—not just engineers and scientists, but artists, writers, and creators.", "topic_id": "topic_8", "line_start": 269, "line_end": 269 }, { "id": "e7", "explicit_text": "We met xAI...and xAI was just getting started at that point and they were super excited about our focus on the quality of the experts.", "inferred_identity": "xAI (Elon Musk's AI company, founded 2023)", "confidence": "explicit", "tags": [ "xAI", "Elon Musk", "early stage AI labs", "customer acquisition", "founder network", "product-market fit signals", "Tesla office" ], "lesson": "When frontier AI labs are excited about your solution, that's the strongest market validation. It's a leading indicator of massive demand.", "topic_id": "topic_11", "line_start": 341, "line_end": 341 }, { "id": "e8", "explicit_text": "One of the crowdsourcing players came to us and actually used our platform to hire over 1,000 people where this is very interesting experience because we started getting flooded with support tickets about how those people weren't getting paid.", "inferred_identity": "Scale AI or Surge (major crowdsourcing companies)", "confidence": "inferred - Brendan mentions 'Scale and Surge were the primary companies'", "tags": [ "Scale AI", "Surge", "crowdsourcing", "data labeling", "labor exploitation", "market pain point", "payment issues", "competitive advantage" ], "lesson": "Incumbents cut corners on talent treatment. This became Mercor's competitive advantage: treating experts well, paying them fairly, and building a premium marketplace instead of a discount one.", "topic_id": "topic_11", "line_start": 344, "line_end": 344 }, { "id": "e9", "explicit_text": "When I was in eighth grade, I started Donut Dynasty where I saw that Safeway Donuts were selling for $5 a dozen, and I was amazed because I felt like as an eighth grader, this was such an incredible deal. And I started to bike down to Safeway, buy Safeway Donuts for $5 a dozen, and then go back to my middle school and then sell them for $2 each.", "inferred_identity": "Brendan Foody's personal early business (age ~13-14, likely Bay Area)", "confidence": "explicit - personal anecdote", "tags": [ "Brendan Foody", "early entrepreneurship", "retail arbitrage", "pricing", "unit economics", "youth entrepreneur", "initiative", "business fundamentals" ], "lesson": "Entrepreneurship starts with identifying inefficiencies and capturing them. A middle schooler identifying a 3x margin opportunity and executing it is the essence of founder mentality.", "topic_id": "topic_15", "line_start": 413, "line_end": 414 }, { "id": "e10", "explicit_text": "The school tried to shut me down because I was selling food on school campus, which they didn't like. So they had me in the principal's office asking me to not do that. And then I moved my donut stand over 50 feet, so it was off school campus, saying that they could no longer police me.", "inferred_identity": "Brendan Foody's Donut Dynasty at middle school campus", "confidence": "explicit - personal story", "tags": [ "regulatory constraints", "creative problem-solving", "jurisdiction", "founder resilience", "rule-bending", "persistence", "anti-authority" ], "lesson": "When faced with arbitrary restrictions, find creative workarounds. Legal boundaries often have unexpected gaps that clever entrepreneurs can exploit.", "topic_id": "topic_15", "line_start": 417, "line_end": 417 }, { "id": "e11", "explicit_text": "I remember we had competitors pop up where the competitors were charging...Chuck's Donuts, which if anyone in the Bay Area knows, are higher end donuts than Safeway Donuts, but they have a higher cost basis. They cost a dollar per. And so I dropped my prices to $1 for two weeks to run them out of business before I knew what anti-competitive practices were.", "inferred_identity": "Brendan Foody's Donut Dynasty (competitive response to Chuck's Donuts competitors)", "confidence": "explicit", "tags": [ "competitive strategy", "pricing war", "predatory pricing", "market dynamics", "youth entrepreneur", "unit economics", "aggressive tactics" ], "lesson": "Young founders learn competitive tactics intuitively. Dropping price to kill competitors is a basic strategy, though often illegal at scale. Context matters.", "topic_id": "topic_15", "line_start": 416, "line_end": 416 }, { "id": "e12", "explicit_text": "And I'd hire all my friends, paying my friends in donuts because they perceived the donuts as $2 each where they could sell them throughout the school and I could have a lower cost basis on them.", "inferred_identity": "Brendan Foody's Donut Dynasty compensation structure", "confidence": "explicit", "tags": [ "creative compensation", "barter economy", "perceived value", "incentive alignment", "team building", "economics", "founder lessons" ], "lesson": "Compensation doesn't have to be cash. If people perceive value in what you're offering, they'll work for it. This is similar to early startup equity dynamics.", "topic_id": "topic_15", "line_start": 416, "line_end": 416 }, { "id": "e13", "explicit_text": "And then Benchmark before they led our Series A...I said we'd be at 50 million in run rate by the end of the year. And they said we were absolutely insane, right, as anyone would. And plus or minus two weeks, we hit it.", "inferred_identity": "Benchmark Capital (led Mercor Series A), Mercor milestone", "confidence": "explicit", "tags": [ "Benchmark", "Series A fundraising", "growth targets", "can-do attitude", "execution", "VC expectations", "founder credibility", "revenue growth" ], "lesson": "Setting ambitious goals and hitting them (even when investors think you're insane) builds credibility and momentum. It also becomes the culture.", "topic_id": "topic_12", "line_start": 368, "line_end": 368 }, { "id": "e14", "explicit_text": "We have an incredibly high hiring bar where we hire tons of former founders, people that have incredible experiences. We just hired or partnered with Sundeep Jain who joined us as president. He was previously the chief product officer and chief technology officer at Uber.", "inferred_identity": "Sundeep Jain (Uber's former CPO/CTO), Mercor hire as President", "confidence": "explicit", "tags": [ "Sundeep Jain", "Uber", "senior hires", "executive recruitment", "high standards", "labor marketplace expertise", "scaling leadership" ], "lesson": "Hiring executives from category-defining companies (like Uber for labor marketplaces) is a strategic bet. It signals confidence to the market and brings proven expertise.", "topic_id": "topic_12", "line_start": 371, "line_end": 371 }, { "id": "e15", "explicit_text": "Our second employee, Sid, as an example, our second employee in the US, Sid was previously the head of growth at Scale who joined us when we were a seed stage company.", "inferred_identity": "Sid (Scale's head of growth), Mercor employee #2", "confidence": "explicit", "tags": [ "Scale AI", "head of growth", "early hires", "talent density", "competitive hire", "operator recruitment" ], "lesson": "Hiring experienced operators from competitors at the seed stage requires conviction and equity, but creates disproportionate value.", "topic_id": "topic_13", "line_start": 386, "line_end": 386 }, { "id": "e16", "explicit_text": "Daniel who joined us was previously scaled to consumer apps to over 100,000 users and all sorts of just extraordinary backgrounds of our first 10 hires.", "inferred_identity": "Daniel (Mercor early hire with scaling experience)", "confidence": "explicit", "tags": [ "consumer scaling", "growth experience", "early hires", "operational excellence", "founder background" ], "lesson": "First 10 hires should be people who've already succeeded at previous scales. This creates a high-performance culture from the start.", "topic_id": "topic_13", "line_start": 386, "line_end": 386 }, { "id": "e17", "explicit_text": "I remember when I used to think... Everyone in venture talks about the why now, and I used to think about the why now of how from a product standpoint, less from a market standpoint of now we can automate the way that we review resumes or the way that we conduct interviews, et cetera. But ultimately there is this legacy market that's has all these incumbents and it's relatively stagnant.", "inferred_identity": "Mercor's historical product positioning vs. market positioning", "confidence": "explicit", "tags": [ "Mercor", "product-market fit", "market timing", "legacy vs emerging", "strategic pivot", "founder lessons" ], "lesson": "The difference between a good market and a great market isn't product innovation—it's whether the market is expanding or stagnant. Mercor's real breakthrough was recognizing the AI evals market was emerging while the hiring market was mature.", "topic_id": "topic_12", "line_start": 332, "line_end": 332 }, { "id": "e18", "explicit_text": "I remember when we were talking to Benchmark before they led our Series A, we were at 1.5 million in run rate. And I said we'd be at 50 million in run rate by the end of the year...Plus or minus two weeks, we hit it. And then we've now well blown past the tracking to 500 million in run rate, which was initially our goal for this year.", "inferred_identity": "Mercor's revenue trajectory", "confidence": "explicit", "tags": [ "Mercor", "revenue growth", "Series A", "execution", "ambitious targets", "founder prediction", "hypergrowth" ], "lesson": "Setting public, ambitious targets and hitting them becomes a form of communication to the market about your capabilities and certainty.", "topic_id": "topic_12", "line_start": 368, "line_end": 368 }, { "id": "e19", "explicit_text": "How can they leverage this technology to do so much more? We'll give people interviews where we say, 'Use whatever tools are available to build a website and let's see what product you're able to build in an hour.'", "inferred_identity": "Mercor's hiring assessment methodology (implicit example from Mercor interviews)", "confidence": "explicit", "tags": [ "Mercor", "hiring practices", "AI tool usage", "assessment", "practical skills", "modern recruiting", "tool-assisted evaluation" ], "lesson": "Future hiring assessments should measure output with AI tools available, not ability to work without them. This is how you identify who'll be productive in the AI era.", "topic_id": "topic_5", "line_start": 32, "line_end": 32 }, { "id": "e20", "explicit_text": "I have had a couple conversations with teachers where they get my thoughts on how they should be assessing their students because we originally started out curating all of these AI interviews and assessments for people.", "inferred_identity": "Mercor's educational assessment work (unnamed teachers)", "confidence": "explicit", "tags": [ "Mercor", "education", "assessment", "AI in learning", "teacher interviews", "curriculum design" ], "lesson": "The same principles that apply to evaluating AI systems apply to evaluating students: measure what they can accomplish with tools available, not what they can do from first principles.", "topic_id": "topic_5", "line_start": 179, "line_end": 179 }, { "id": "e21", "explicit_text": "My parents would give me a hard time for not going to classes in college and I'd be like, well, there's way better lectures on YouTube. Why not just listen there?", "inferred_identity": "Brendan Foody's college experience (unnamed university, likely Stanford or Berkeley based on context)", "confidence": "explicit", "tags": [ "Brendan Foody", "college dropout", "online learning", "educational disruption", "prioritization", "founder mindset" ], "lesson": "Founders often optimize for high-value activities over social expectations. Skipping classes to learn better elsewhere is a small signal of founder mentality.", "topic_id": "topic_16", "line_start": 479, "line_end": 479 }, { "id": "e22", "explicit_text": "Anthropic...Claude has been so good at coding so much better historically than other models.", "inferred_identity": "Anthropic, Claude (AI model)", "confidence": "explicit", "tags": [ "Anthropic", "Claude", "coding capability", "model quality", "competitive advantage", "AI labs" ], "lesson": "Claude's coding superiority likely comes from thoughtful post-training with expert software engineers defining rubrics and evaluating reasoning quality, not just more pre-training data.", "topic_id": "topic_10", "line_start": 299, "line_end": 299 }, { "id": "e23", "explicit_text": "I really liked Oppenheimer. My favorite TV show of all time is Suits, so I know not recent, but if I had to choose a recent one, probably Oppenheimer.", "inferred_identity": "Brendan Foody's media preferences", "confidence": "explicit", "tags": [ "Brendan Foody", "personal preferences", "Oppenheimer", "Suits", "entertainment", "cultural references" ], "lesson": "Leaders consume media across different genres. Oppenheimer (complex historical drama about nuclear physics) and Suits (legal procedural) suggest interest in both technical and interpersonal complexity.", "topic_id": "topic_19", "line_start": 560, "line_end": 560 }, { "id": "e24", "explicit_text": "I love using Codex, like the new version. I know it's sort of new in terms of version. Yeah, I think it's incredible and just a huge, huge improvement.", "inferred_identity": "GitHub Copilot Codex (or Claude Code based on context)", "confidence": "explicit", "tags": [ "Codex", "GitHub Copilot", "AI coding tools", "developer tools", "productivity", "Brendan Foody preferences" ], "lesson": "Even CEO of an AI company uses external AI coding tools. This suggests multi-model diversity is valuable even for AI practitioners.", "topic_id": "topic_19", "line_start": 566, "line_end": 566 } ] }