Lenny RAG MCP Server

{ "episode": { "guest": "Garrett Lord", "expertise_tags": [ "AI/ML", "Data Labeling", "Marketplace Platforms", "Founder/CEO", "Human Data Infrastructure", "Post-Training Data", "AI Model Development" ], "summary": "Garrett Lord, co-founder and CEO of Handshake, discusses how his decade-old college recruiting platform evolved into a $100M+ ARR data labeling business for frontier AI labs. The company leverages its network of 18 million students, PhDs, and professionals to create high-quality training data for models—a business that went from zero to $50M ARR in four months. Lord explains the data labeling ecosystem, how it shifted from low-cost generalists to expert-driven work, and shares lessons on building a new business inside an established company while managing unlimited market demand.", "key_frameworks": [ "Pre-training vs Post-training", "Supervised Fine-Tuning (SFT)", "Reinforcement Learning from Human Feedback (RLHF)", "Expert Network Moat", "Customer Acquisition Cost (CAC) vs Lifetime Value (LTV)", "Trajectory Data Collection", "Leave Nothing to Chance", "Network Effects in Marketplaces" ] }, "topics": [ { "id": "topic_1", "title": "Introduction to Data Labeling and Model Training Fundamentals", "summary": "Garrett explains what data labeling actually is, distinguishing between pre-training (feeding models the entire internet corpus) and post-training (improving specific capabilities through targeted data). He describes how models have exhausted internet knowledge and now rely on post-training gains through high-quality, domain-specific data collection.", "timestamp_start": "00:05:55", "timestamp_end": "00:08:20", "line_start": 46, "line_end": 55 }, { "id": "topic_2", "title": "Types of Post-Training Data: SFT, RLHF, and Trajectories", "summary": "Deep dive into post-training methodologies including Supervised Fine-Tuning (prompt-response pairs), RLHF (preference ranking), trajectory data (step-by-step problem solving), and reinforcement learning environments. Garrett explains the scientific experimentation process labs use to test hypotheses about data quality improvements.", "timestamp_start": "00:08:41", "timestamp_end": "00:10:52", "line_start": 58, "line_end": 69 }, { "id": "topic_3", "title": "Expert Networks vs Generalist Labor: The Market Shift", "summary": "Garrett articulates the fundamental market shift from low-cost international generalist labor (drawing bounding boxes) to expert-driven data creation. Models have become so capable that only domain experts—PhDs in physics, chemistry, biology, etc.—can effectively identify and fix model failures. This shift defines Handshake's competitive advantage.", "timestamp_start": "00:10:52", "timestamp_end": "00:12:08", "line_start": 65, "line_end": 69 }, { "id": "topic_4", "title": "Practical Examples of Data Labeling Work: GPQA and Educational Design", "summary": "Concrete examples of what experts actually do: using GPQA (a research paper framework) to break models, provide ground truth answers, and identify where step-by-step reasoning fails. Real example of Rachel, a PhD in education, helping improve models' understanding of educational design based on her decade teaching eighth graders.", "timestamp_start": "00:13:22", "timestamp_end": "00:17:12", "line_start": 79, "line_end": 107 }, { "id": "topic_5", "title": "Trajectory Data and Multimodal Work: Screen Recording, Voice, Audio Classification", "summary": "Explanation of trajectory data—collecting entire environments including screen, mouse movements, and voice narration as experts solve problems. Discussion of multimodal work like audio classification with music students and rubric-based evaluation where models act as judges for subjective domains.", "timestamp_start": "00:17:56", "timestamp_end": "00:19:28", "line_start": 112, "line_end": 132 }, { "id": "topic_6", "title": "Quality, Volume, and Speed: The Three Things Model Builders Care About", "summary": "Garrett identifies the hierarchy of concerns for frontier labs: quality is paramount (bad data corrupts learning), volume at scale in advanced domains, and speed to execute on hypotheses quickly. He describes Handshake's infrastructure—their post-training team, GPU rental, and data assessment technology—to deliver on all three.", "timestamp_start": "00:19:52", "timestamp_end": "00:22:03", "line_start": 136, "line_end": 144 }, { "id": "topic_7", "title": "Job Market Impact: AI as Productivity Multiplier, Not Job Destroyer", "summary": "Garrett pushes back on fears that AI training will eliminate entry-level jobs. He argues AI acts as an 'Iron Man suit' enabling young, AI-native workers to be vastly more productive—one person can do work that previously required teams. Examples include social media marketing and engineering productivity improvements.", "timestamp_start": "00:24:42", "timestamp_end": "00:27:08", "line_start": 172, "line_end": 189 }, { "id": "topic_8", "title": "Origins of Handshake AI: From Middleman Disruption to Direct Customer Relationships", "summary": "How Handshake discovered the opportunity: other platforms were recruiting their PhDs and masters students with poor experiences, while frontier labs were reaching out directly trying to bypass middlemen. Garrett realized he could build a superior platform serving experts, labs, and companies simultaneously.", "timestamp_start": "00:33:40", "timestamp_end": "00:35:48", "line_start": 220, "line_end": 227 }, { "id": "topic_9", "title": "Explosive Growth: Zero to $50M ARR in Four Months, Trajectory to $100M+", "summary": "Numbers behind the story: launched in January, hit $50M ARR by May (four months), on pace for $100M+ by end of year one. Working with seven frontier labs. Team grew from 4-5 to 75+ people. Handshake core business is $200M ARR, making this new business enormous relative to company age.", "timestamp_start": "00:36:28", "timestamp_end": "00:37:20", "line_start": 230, "line_end": 251 }, { "id": "topic_10", "title": "The Handshake Core Business: LinkedIn for College Students and Young Professionals", "summary": "Background on Handshake's original business—a social platform helping 18 million students and alumni discover jobs at Fortune 500 companies and 1 million employers. Built over 10 years, powers job matching for colleges and companies. Positioned as an 'unconnected graph' focusing on discovery and career exploration.", "timestamp_start": "00:37:18", "timestamp_end": "00:39:36", "line_start": 244, "line_end": 272 }, { "id": "topic_11", "title": "Competitive Moat: Zero Customer Acquisition Cost and Network Effects", "summary": "Handshake's unfair advantage: zero CAC to acquire experts (built brand with 18M people + 1,600 university partnerships), high conversion rates from trust, and superior LTV through retention and repeat project participation. Competitors spend tens of millions on performance marketing and recruiters; Handshake has trust and brand affinity.", "timestamp_start": "00:29:50", "timestamp_end": "00:31:49", "line_start": 205, "line_end": 210 }, { "id": "topic_12", "title": "Platform Design for Experts: Community, Training, Cohorts, and Quality Assurance", "summary": "How Handshake treats PhDs and experts differently than low-cost platforms: community-first onboarding, cohort-based learning, instructional design teams, assessment teams, and pre-built data packages sold to multiple labs. Separate swim lanes for internal quality projects vs customer projects, creating a pipeline from training to high-value work.", "timestamp_start": "00:42:15", "timestamp_end": "00:43:46", "line_start": 296, "line_end": 303 }, { "id": "topic_13", "title": "Building Business Two Inside Business One: Organizational Separation and Structure", "summary": "Garrett details how he created a separate company-within-company: dedicated teams with no other responsibilities, physically separated office, five-day office weeks, metrics-based cadence, metrics-driven operations, different compensation (tied to hurdles), different expectations around chaos and pace, entrepreneurial hiring, and single DRI accountability regardless of function.", "timestamp_start": "00:48:50", "timestamp_end": "00:52:42", "line_start": 346, "line_end": 357 }, { "id": "topic_14", "title": "Founder-Led Execution and Hiring Top Talent from Core Business", "summary": "Garrett was hands-on as lead (80%+ time allocation), everyone reported directly to him early on, he recruited top senior engineers and principal engineers from core business to lead infrastructure and engineering. Hired people with startup experience, comfortable with ambiguity, willing to work weekends and 2-3 AM nights. Clearly communicated expectations and ownership mentality.", "timestamp_start": "00:53:50", "timestamp_end": "00:55:49", "line_start": 364, "line_end": 369 }, { "id": "topic_15", "title": "Leave Nothing to Chance: Operational Ruthlessness and Urgency", "summary": "Core operating principle for Handshake AI: 'leave nothing to chance, leave it all on the field.' Tracked days in year on whiteboard recognizing unlimited demand window won't last forever. Ethos of getting on planes to talk to customers, making late-night pushes, checking data six times over, shipping extra features. Celebratory culture calling out impact creators.", "timestamp_start": "00:56:39", "timestamp_end": "00:57:44", "line_start": 371, "line_end": 380 }, { "id": "topic_16", "title": "Data Types Evolution and the Future of Training Data", "summary": "Garrett predicts data needs will evolve significantly: CAD files, scientific tool use data for drug discovery, esoteric operating system data, multimodal (video, audio, text), and emphasis on step-by-step instruction following. Synthetic data will have a role in verifiable domains but won't dominate. Billions in value to extract over the next decade.", "timestamp_start": "01:00:50", "timestamp_end": "01:02:24", "line_start": 391, "line_end": 399 }, { "id": "topic_17", "title": "Long-Term Vision: AI-Powered Hiring and Job Matching at Scale", "summary": "Garrett's broader vision for Handshake: AI will completely reinvent job search and hiring process. AI interviewers, work simulations, automated resume review replacing manual review of hundreds of resumes. Human data labeling business serves as foundation for improving matching intelligence. Connects to his founding mission of democratizing opportunity.", "timestamp_start": "00:57:44", "timestamp_end": "00:59:44", "line_start": 379, "line_end": 386 }, { "id": "topic_18", "title": "Advice for Young Entrepreneurs: Building Meaningful AI Companies", "summary": "Garrett's call to action for founders: focus on doing something of meaning that helps people; there are enormous opportunities to improve learning with AI. Committed to making Handshake not just a great business but a solution to a societal problem (labor supply matching). Offers advice to entrepreneurs interested in AI.", "timestamp_start": "01:02:52", "timestamp_end": "01:03:50", "line_start": 409, "line_end": 428 }, { "id": "topic_19", "title": "Lightning Round: Books, TV, Products, Mottoes, and Origin Story", "summary": "Garrett's recommendations: Zero to One by Peter Thiel, Shoe Dog, Hard Things About Hard Things by Ben Horowitz. Recently started Game of Thrones. Favorite product: SNOO bassinet. Life motto: 'Leave nothing to chance.' Origin story: showering in Princeton pool during early Handshake sales trips sleeping in Ford Focus to save money.", "timestamp_start": "01:04:06", "timestamp_end": "01:07:29", "line_start": 439, "line_end": 504 }, { "id": "topic_20", "title": "Hiring and How to Connect with Garrett", "summary": "Handshake is aggressively hiring, particularly engineers, to meet unlimited demand. Offices in San Francisco, New York, London, and Berlin. Ways to reach Garrett: Handshake platform (garrettlord@handshake), Twitter/X, email (garrett@joinhandshake.com). Open to roles across consumer product, employer products, and AI business.", "timestamp_start": "01:08:12", "timestamp_end": "01:09:18", "line_start": 508, "line_end": 521 } ], "insights": [ { "id": "I1", "text": "The models have gotten so good that the generalists are no longer needed. What they really need is experts.", "context": "Fundamental market shift in data labeling: as models improve, only domain experts can identify failures", "topic_id": "topic_3", "line_start": 14, "line_end": 14 }, { "id": "I2", "text": "The only moat in human data is access to an audience.", "context": "Core competitive advantage principle—network effects and scale of expert pool matter more than technology", "topic_id": "topic_11", "line_start": 8, "line_end": 8 }, { "id": "I3", "text": "Pre-training has asymptoted because labs have sucked up all knowledge on the internet; most gains now come from post-training.", "context": "Why frontier labs shifted focus and created massive demand for data labeling", "topic_id": "topic_1", "line_start": 47, "line_end": 47 }, { "id": "I4", "text": "Models don't generalize well yet, so for as long as there are problems to solve, humans will be needed in the loop for the next decade until we reach full ASI.", "context": "Counterargument to AI replacing human labor in data annotation; expertise will remain critical", "topic_id": "topic_1", "line_start": 167, "line_end": 167 }, { "id": "I5", "text": "Being AI native and having your Iron Man suit on, understanding how to leverage these tools is like young people are at a huge advantage.", "context": "Framework for understanding AI impact on employment—tools amplify capability for those who grew up with them", "topic_id": "topic_7", "line_start": 182, "line_end": 182 }, { "id": "I6", "text": "One AI-native, talented young person can build their own videos, produce creative assets, post across platforms, and run analytics—work that used to require teams.", "context": "Concrete example of productivity multiplication from AI tools enabling entry-level workers to do more", "topic_id": "topic_7", "line_start": 173, "line_end": 173 }, { "id": "I7", "text": "Model builders care about three things: quality first and foremost (bad data corrupts learning), volume at scale in advanced domains, and speed to run experiments quickly.", "context": "Hierarchy of concerns that should drive platform design and data collection strategy", "topic_id": "topic_6", "line_start": 137, "line_end": 137 }, { "id": "I8", "text": "The model of today is the worst model you will ever use—it will only get better.", "context": "Kevin Wheel (OpenAI CBO) insight on continuous model improvement, justifying investment in data labeling infrastructure", "topic_id": "topic_6", "line_start": 197, "line_end": 197 }, { "id": "I9", "text": "If you viewed your research and model building teams as the cornerstone of improving, you wouldn't want your latest research being outsourced to a competitor.", "context": "Explains why frontier labs prefer direct relationships with data providers rather than third-party platforms", "topic_id": "topic_8", "line_start": 205, "line_end": 206 }, { "id": "I10", "text": "We built a decade of trust with 18 million people, and they trust us, with tons of brand affinity and information about their academic performance, so we target people effectively and get to scale faster than anyone else.", "context": "How existing brand, trust, and data create superior network effects and LTV", "topic_id": "topic_11", "line_start": 209, "line_end": 209 }, { "id": "I11", "text": "There's quite a bit of technology we built to assess each unit of data. We have our own post-training teams, we're renting our own GPUs, sitting directly with researchers to share what we're seeing about data quality.", "context": "Vertical integration strategy—Handshake doesn't just source data, they validate and iterate on it", "topic_id": "topic_6", "line_start": 143, "line_end": 143 }, { "id": "I12", "text": "In a marketplace, you have to serve three different sides—it's hard to build a new business inside an existing one, especially with many zero-to-ones.", "context": "Structural complexity of adding new marketplace (labs + experts + companies) within existing platform", "topic_id": "topic_13", "line_start": 350, "line_end": 350 }, { "id": "I13", "text": "Everyone saw the structural advantages we had and the focus was on delivering high-quality data to one customer before expanding to anyone else.", "context": "Discipline of focusing on quality and customer success before growth", "topic_id": "topic_13", "line_start": 350, "line_end": 350 }, { "id": "I14", "text": "For a business operating at hyper-growth with unlimited demand, separate everything: teams, engineering, design, ops, finance—everyone reports to one mission with one job only.", "context": "Organizational structure principle for zero-to-one businesses inside mature companies", "topic_id": "topic_13", "line_start": 364, "line_end": 365 }, { "id": "I15", "text": "Hiring people who've only worked at early-stage companies and are comfortable with ambiguity is critical for fast-moving environments.", "context": "Talent selection principle for startup velocity", "topic_id": "topic_14", "line_start": 365, "line_end": 365 }, { "id": "I16", "text": "Be upfront about the chaos and different expectations from day one: this is 24/7, this is chaotic, this is an early-stage company. Compensation should reflect ownership in the new business.", "context": "Recruiting and retention strategy for high-growth teams", "topic_id": "topic_14", "line_start": 368, "line_end": 368 }, { "id": "I17", "text": "Leave nothing to chance—there will never be a time like this, unlimited demand. How do you make sure in three months you have no regrets? Get on the plane, make the late-night push, check the data six times over again.", "context": "Operating principle for seizing time-limited opportunities with perfect execution", "topic_id": "topic_15", "line_start": 1, "line_end": 2 }, { "id": "I18", "text": "The type of data labs will need is evolving—CAD files, scientific tool use, esoteric operating systems, multimodal (video, audio, text), and step-by-step instruction following.", "context": "Future-proofing insight: data labeling isn't a one-time problem, it's continuously evolving", "topic_id": "topic_16", "line_start": 391, "line_end": 392 }, { "id": "I19", "text": "Synthetic data will play a role in verifiable domains, but it won't dominate—there's billions of dollars of value to extract from human expertise over the next decade.", "context": "Pushback on synthetic data hype; human experts will remain irreplaceable for complex domains", "topic_id": "topic_16", "line_start": 398, "line_end": 398 }, { "id": "I20", "text": "The hiring manager process of reviewing 200 resumes won't exist in five years—AI will completely reinvent job matching and candidate evaluation.", "context": "Vision for how AI improvements (enabled by data labeling) will transform recruiting", "topic_id": "topic_17", "line_start": 380, "line_end": 381 }, { "id": "I21", "text": "Our human data business serves as the foundation for improving matching in our core business—integrating post-training capabilities back into core recruitment products.", "context": "Synergy between two businesses: data labeling improvements enable better job matching", "topic_id": "topic_17", "line_start": 383, "line_end": 383 }, { "id": "I22", "text": "You have to be hands-on as founder when building zero-to-one inside an existing company—everyone reported directly to me early on.", "context": "Leadership principle for entrepreneurial ventures within larger organizations", "topic_id": "topic_14", "line_start": 350, "line_end": 350 }, { "id": "I23", "text": "Pick the best person most capable of driving an initiative forward as DRI regardless of function—organizational structure should be merit-based, not rigid hierarchy.", "context": "Flat organizational principle for fast-moving teams", "topic_id": "topic_13", "line_start": 356, "line_end": 356 }, { "id": "I24", "text": "Set up metrics-driven operating cadence from the start—weekly, monthly, quarterly metrics and data rigor from early stage, not after scaling.", "context": "Building discipline and visibility into execution from day one", "topic_id": "topic_13", "line_start": 356, "line_end": 356 }, { "id": "I25", "text": "Celebratory culture calling out people putting up impact points makes fast-moving environments more sustainable and fun.", "context": "Cultural practice for maintaining momentum and team morale", "topic_id": "topic_15", "line_start": 374, "line_end": 374 }, { "id": "I26", "text": "In data labeling, LTV is calculated simply: retention of person plus number of projects they participate in—so treat people really well and retention compounds.", "context": "Economic model showing why network effects and retention matter more than CAC", "topic_id": "topic_11", "line_start": 305, "line_end": 305 }, { "id": "I27", "text": "Universities not tolerating bad treatment of their students/alumni means platforms must build genuinely good experiences—external accountability for ethics.", "context": "Market mechanism that enforces quality and fairness", "topic_id": "topic_12", "line_start": 305, "line_end": 305 }, { "id": "I28", "text": "Create instructional design and assessment teams teaching experts how to use tools—the cost of acquisition is high, so training front-load matters.", "context": "Quality assurance practice differentiating from competitor platforms", "topic_id": "topic_12", "line_start": 83, "line_end": 84 }, { "id": "I29", "text": "Building pre-made data packages sold to multiple labs lets you control quality, test post-training impact, then graduate top performers to customer projects.", "context": "Portfolio approach to data projects with quality testing and tiered expert development", "topic_id": "topic_12", "line_start": 299, "line_end": 302 } ], "examples": [ { "id": "EX1", "explicit_text": "At my previous company, there's a PhD student, Rachel on the network, she got her PhD from the University of Miami, spent two decades as a teacher teaching students in the eighth grade. And she was an adjunct professor at a local community college in the field of education.", "inferred_identity": "Rachel (Example Handshake user, verified by direct naming)", "confidence": "high", "tags": [ "education", "PhD", "teacher", "data-labeling", "educational-design", "model-improvement", "domain-expertise" ], "lesson": "PhD in education with 10+ years teaching experience can break frontier models on educational design—combining credentials with lived practice creates irreplaceable expertise", "topic_id": "topic_4", "line_start": 104, "line_end": 104 }, { "id": "EX2", "explicit_text": "Take a non-verifiable domain like education. Rachel is interacting with the state-of-the-art models in educational design. She is trying to understand what is the best way to teach people, and how do you spot incorrect issues in a model in the way that they're training people, and help the models understand the forefront of educational design.", "inferred_identity": "Frontier AI labs (OpenAI, Anthropic, Google DeepMind, Meta, etc. - 7 labs mentioned but not named)", "confidence": "high", "tags": [ "educational-design", "model-improvement", "frontier-labs", "non-verifiable-domains", "human-feedback", "step-by-step-reasoning" ], "lesson": "For subjective domains without ground truth answers, experts use rubric-based evaluation—models learn what good looks like from experienced practitioners", "topic_id": "topic_4", "line_start": 104, "line_end": 104 }, { "id": "EX3", "explicit_text": "There's a public paper called GPQA. Essentially the crux of the paper is you break the model, you provide a ground truth, the right answer to the question, you provide the step-by-step reasoning steps.", "inferred_identity": "GPQA (Google Research paper/benchmark on expert question answering)", "confidence": "high", "tags": [ "GPQA", "benchmark", "reasoning", "expert-questions", "model-failure", "step-by-step", "data-labeling" ], "lesson": "Models often get the right answer for wrong reasons—high-quality training data must capture correct step-by-step reasoning, not just correct outputs", "topic_id": "topic_4", "line_start": 79, "line_end": 80 }, { "id": "EX4", "explicit_text": "We have 18 million professionals across, we have 500,000 PhDs, we have 3 million master students, we're a global platform.", "inferred_identity": "Handshake (Garrett's core platform with massive expert network)", "confidence": "high", "tags": [ "Handshake", "PhDs", "master-students", "platform", "network-effects", "competitive-advantage", "expert-pool" ], "lesson": "Built network over 10 years becomes irreplaceable asset when market suddenly values expert labor—timing and prior preparation create breakthrough opportunities", "topic_id": "topic_11", "line_start": 65, "line_end": 65 }, { "id": "EX5", "explicit_text": "Scale was just acquired for $30 billion. A lot of the market before the model started to get better was leveraging talented international lower cost labor to do basic generalist tasks.", "inferred_identity": "Scale (formerly Scale AI, largest data labeling competitor, acquired by Meta for $30B+ reported)", "confidence": "high", "tags": [ "Scale", "data-labeling", "acquisition", "Meta", "generalist-labor", "international", "model-scaling" ], "lesson": "Scale optimized for generalist labor and bounding-box annotation just as market shifted to expert-driven work—business model misalignment with evolving market needs", "topic_id": "topic_3", "line_start": 191, "line_end": 191 }, { "id": "EX6", "explicit_text": "I hired three expert firms, AlphaSights and GLG, and started doing a bunch of calls with the latest researchers, because we had resources. One of the cool things about being a larger company is our core business is $200 million ARR.", "inferred_identity": "AlphaSights and GLG (expert networks used to access frontier researchers for due diligence calls)", "confidence": "high", "tags": [ "expert-networks", "AlphaSights", "GLG", "due-diligence", "frontier-labs", "research", "learning" ], "lesson": "Larger companies can afford to invest in learning and exploration—hiring expert networks to interview researchers accelerated understanding of market", "topic_id": "topic_8", "line_start": 329, "line_end": 329 }, { "id": "EX7", "explicit_text": "And then, we started working with arguably the number one lab about five months ago... now we're working with seven of the Frontier Labs, basically every lab that's doing work and building the best large language models.", "inferred_identity": "Frontier Labs (OpenAI, Anthropic, Google DeepMind, Meta/LLaMA, Mistral, xAI, etc. - 7 unnamed labs)", "confidence": "high", "tags": [ "frontier-labs", "OpenAI", "Anthropic", "DeepMind", "Meta", "LLM", "model-training", "customers" ], "lesson": "Went from zero to contracted with 7 frontier labs in less than a year—network effects and competitive advantage made inbound sales effortless", "topic_id": "topic_9", "line_start": 236, "line_end": 237 }, { "id": "EX8", "explicit_text": "100% of the Fortune 500 uses Handshake, so we basically power the vast majority of how young people find jobs, and a lot of people are hyperbolic at saying that all young people won't have jobs, and that's not what we're hearing from our employers.", "inferred_identity": "Fortune 500 companies using Handshake for recruiting", "confidence": "high", "tags": [ "Fortune-500", "recruiting", "employer-base", "scale", "distribution", "moat", "competitive-advantage" ], "lesson": "Distribution to employers creates indirect distribution to candidates—platform effects compound when one side of marketplace is locked in", "topic_id": "topic_10", "line_start": 172, "line_end": 173 }, { "id": "EX9", "explicit_text": "What we're hearing is pick social media marketing, before you needed somebody that could do Photoshop, and take pictures, and create the videos. Then you needed somebody that understood marketing analytics platforms. It's like now one person, one young, talented, AI native, Iron Man suit enabled young person can do it all.", "inferred_identity": "Social media marketing industry (implicit example of job expansion through AI tools)", "confidence": "medium", "tags": [ "social-media", "marketing", "productivity", "AI-native", "entry-level", "job-evolution", "tools" ], "lesson": "AI tools don't eliminate jobs; they expand the scope of individual capability—one social media marketer can now do work that required a team", "topic_id": "topic_7", "line_start": 173, "line_end": 173 }, { "id": "EX10", "explicit_text": "Or take an intern in our company, he had his first PR up I think the afternoon he started. You were a PM, you realize how challenging that would've been historically to get your dev environment set up and figure out where to add value.", "inferred_identity": "Handshake AI intern (anonymous example from company)", "confidence": "medium", "tags": [ "engineering", "intern", "productivity", "AI-tools", "developer-experience", "onboarding" ], "lesson": "Entry-level engineers using AI can ship code on day one—removing setup friction enables immediate contribution and learning", "topic_id": "topic_7", "line_start": 175, "line_end": 175 }, { "id": "EX11", "explicit_text": "We're talking about PhDs at the top institutions in the country. They can make 100, 150, $200 an hour in their area, in their field of expertise. It's pretty sweet. You can make 25 bucks an hour being a teacher's assistant, or you can actually make $150 an hour breaking the latest models.", "inferred_identity": "Stanford, MIT, Berkeley, and other top physics schools (implied by earlier context of reaching out to top GPA students)", "confidence": "high", "tags": [ "PhDs", "compensation", "top-schools", "economic-opportunity", "research", "data-labeling", "income" ], "lesson": "Handshake pays 6-8x teacher assistant salaries, making expert data labeling economically attractive to advanced students while keeping costs far below market rates for specialized expertise", "topic_id": "topic_7", "line_start": 188, "line_end": 188 }, { "id": "EX12", "explicit_text": "We bring in a lot of those insights into the classroom to help them be more effective at teaching. More importantly, they're starting to learn how to leverage these tools to actually advance their area of research.", "inferred_identity": "Handshake AI fellows (PhDs and grad students earning money while advancing their research)", "confidence": "high", "tags": [ "PhD-fellows", "research", "education", "learning", "knowledge-transfer", "career-development" ], "lesson": "Data labeling work becomes a learning tool for PhD students—they gain exposure to frontier models and can apply insights to their research, creating positive externality", "topic_id": "topic_7", "line_start": 188, "line_end": 188 }, { "id": "EX13", "explicit_text": "One of the leading players, they have 200 recruiters. It's unsustainable. There are like 200 people on LinkedIn sending individual messages to acquire these people, because there's no brand, there's no trust.", "inferred_identity": "Scale or another competitor with 200 recruiters on staff", "confidence": "medium", "tags": [ "competitor", "recruitment", "unsustainable", "cost-structure", "brand", "network-effects" ], "lesson": "Competitors without brand pay extreme customer acquisition costs (200 recruiters + performance marketing)—Handshake's 10-year brand advantage creates natural moat", "topic_id": "topic_11", "line_start": 284, "line_end": 284 }, { "id": "EX14", "explicit_text": "They're spending tens of millions of dollars a month on performance advertising, Google Ads, YouTube ads. Somebody's scrolling their Instagram feed that's a physics PhD of which you can't target them that well.", "inferred_identity": "Competitor data labeling platforms (multiple players, unnamed)", "confidence": "medium", "tags": [ "competitor", "marketing-spend", "customer-acquisition", "inefficient", "performance-marketing", "waste" ], "lesson": "Targeting experts on social media through ads is inherently inefficient—Handshake's owned network is vastly more efficient", "topic_id": "topic_11", "line_start": 206, "line_end": 206 }, { "id": "EX15", "explicit_text": "At Palantir as an intern, it totally changed my life, and I started Handshake because I wanted to make it easier for anyone regardless of who you knew, what your parents did, what school you went to, to find a great opportunity.", "inferred_identity": "Palantir (company where Garrett interned; foundational experience for Handshake mission)", "confidence": "high", "tags": [ "Palantir", "internship", "career-change", "mission", "opportunity", "founder-origin" ], "lesson": "Garrett's Palantir internship showed him the power of opportunity access—became founding mission for Handshake, now applies to AI era", "topic_id": "topic_17", "line_start": 383, "line_end": 383 }, { "id": "EX16", "explicit_text": "I went to community college, I paid my way through school. I went to a no name school in Upper Peninsula of Michigan.", "inferred_identity": "Garrett Lord (personal background/origin story)", "confidence": "high", "tags": [ "founder", "community-college", "working-class", "background", "scrappy", "determination" ], "lesson": "Garrett's non-traditional background informs mission to democratize opportunity—founder lived the problem Handshake solves", "topic_id": "topic_17", "line_start": 383, "line_end": 383 }, { "id": "EX17", "explicit_text": "I mean, I just think in this moment right now with AI, for young entrepreneurs that listen... I just focus on doing something of meaning that really helps people.", "inferred_identity": "Target audience: Lenny's podcast listeners (entrepreneurs and product builders)", "confidence": "high", "tags": [ "entrepreneurs", "advice", "AI-opportunity", "purpose", "impact", "meaning" ], "lesson": "Timing creates opportunities for founders to build meaningful businesses solving real problems—focus on impact, not just capture", "topic_id": "topic_18", "line_start": 410, "line_end": 410 }, { "id": "EX18", "explicit_text": "We were sleeping out of our car. We had this Ford focus, we would put 20, 30,000 miles on it, sleep in the back of like McDonald's parking lots because they're well lit and had good wifi back in the day.", "inferred_identity": "Garrett Lord and co-founder(s) during early Handshake sales phase (2014-2015 era)", "confidence": "high", "tags": [ "founder", "scrappy", "early-stage", "sales", "hustle", "determination", "hardship" ], "lesson": "Founders willing to endure hardship (sleeping in cars, showering in campus pools) demonstrate commitment that wins over customers", "topic_id": "topic_19", "line_start": 500, "line_end": 500 }, { "id": "EX19", "explicit_text": "Every pool, what do they have? They have a shower. So, you could go to any pool at any university in the country, and you can get a free shower and freshen up. So, the Princeton campus security did not appreciate me showering as a non-student, but I think it meaningfully helped us.", "inferred_identity": "Princeton University (where Garrett showered in pool during sales pitch, almost arrested)", "confidence": "high", "tags": [ "Princeton", "campus", "scrappy-tactics", "hustle", "story", "unconventional", "memorable" ], "lesson": "Unconventional tactics and genuine commitment leave memorable impressions—Princeton security called career services director, generating buzz and proving seriousness", "topic_id": "topic_19", "line_start": 500, "line_end": 503 }, { "id": "EX20", "explicit_text": "The Princeton campus security called the career service center director we were selling to, being like, 'Who's Garrett Lord? Is he really here to pitch you software for your career center?' And it made the start of the meeting with the career center really stimulating and exciting, because they were like, 'You showered in our pool, you drove here?'", "inferred_identity": "Princeton University career services director (customer decision maker)", "confidence": "high", "tags": [ "sales", "persistence", "commitment", "unusual-tactics", "customer-acquisition", "early-stage" ], "lesson": "Extreme commitment and persistence create stories that customers remember and amplify—Princeton security call validated dedication and generated curiosity", "topic_id": "topic_19", "line_start": 503, "line_end": 503 }, { "id": "EX21", "explicit_text": "We started to see all what I would call middleman companies reaching out to us saying, 'Can we recruit your PhDs and master's students?' And like any great marketplace we started sending them to these different platforms, and started to really realize that from hearing from our users that the experience was really frustrating.", "inferred_identity": "Middleman platforms (Scale, Appen, Outlier, CoreWeave, Modal Labs, etc. - unnamed competitors)", "confidence": "high", "tags": [ "middleman-platforms", "competitors", "user-experience", "frustration", "opportunity" ], "lesson": "Customer dissatisfaction with competitor experiences revealed business opportunity—users were ready for a better platform", "topic_id": "topic_8", "line_start": 221, "line_end": 221 }, { "id": "EX22", "explicit_text": "Training was very transactional, it was very amorphous how you could get paid. There was immense amount of drop-off in the process to actual project like completion on these other platforms.", "inferred_identity": "Competitor data labeling platforms (user pain points)", "confidence": "high", "tags": [ "competitor-weakness", "user-experience", "retention", "payment", "onboarding" ], "lesson": "Competitors had poor UX around training, payment clarity, and project completion—Handshake could win by treating users as customers, not labor", "topic_id": "topic_8", "line_start": 221, "line_end": 221 }, { "id": "EX23", "explicit_text": "We started working with arguably the number one lab about five months ago. I mean, we've gone from four or five people working on this to 75 plus people working on it. I think we had 12 people start last Monday.", "inferred_identity": "Handshake AI team (rapid scaling from initial 4-5 to 75+ in 8 months)", "confidence": "high", "tags": [ "team-scaling", "growth", "hiring", "momentum", "founder-led" ], "lesson": "Successful product-market fit with frontier labs required extreme scaling velocity—12 people hired in one week", "topic_id": "topic_9", "line_start": 340, "line_end": 341 }, { "id": "EX24", "explicit_text": "In this market there's essentially unlimited demand. If you can produce high quality volumes of data, you most likely will be able to sell whatever you produce.", "inferred_identity": "Frontier AI market (general observation of market conditions)", "confidence": "high", "tags": [ "market-demand", "unlimited", "supply-constraint", "economics", "opportunity" ], "lesson": "Quality supply is the constraint, not demand—execution and scaling ability determine success in data labeling market", "topic_id": "topic_9", "line_start": 341, "line_end": 341 }, { "id": "EX25", "explicit_text": "I'm really a believer this is just enabling human beings to be even more productive and create more impact. And yeah, of course, hundreds of millions of jobs, the jobs will evolve. People will become displaced, they'll have to upscale and rescale, and I think Handshake has a huge role to play.", "inferred_identity": "Handshake's mission (retraining and reskilling platform for job evolution)", "confidence": "high", "tags": [ "mission", "displacement", "retraining", "opportunity", "impact", "societal" ], "lesson": "Handshake's original mission (connecting talent to opportunity) becomes more critical during AI-driven job market disruption", "topic_id": "topic_7", "line_start": 176, "line_end": 176 } ] }