By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
Sourcery with Molly O'SheaMercor Raises $750M at $20B Valuation | CEO Brendan Foody
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
Great conversations run long. Their best ideas don't have to.
uListen is free on iPhone, iPad, Android, Chrome, and Claude.
See howCHAPTERS
Jump to publisher timestampsChapter summaries
Mercor’s growth jump and the mission-driven mindset
Brendan Foody and Molly O’Shea open with how quickly Mercor has scaled since they last met, including rapid growth off a “mid nine-figure base.” Brendan frames the company’s motivation as impact over cash, setting up the intensity and ambition that runs through the rest of the conversation.
- •Mercor has grown more than 5x since the prior conversation referenced by the host
- •Brendan describes choosing to work intensely rather than “walk away with a lot of cash”
- •Payouts to experts have accelerated dramatically compared to two years prior
- •The tone is set around urgency, ambition, and building something “important in the world”
Working backward from how humans contribute in an agent-driven economy
Asked how Mercor becomes a trillion-dollar company, Brendan argues the core shift is from humans doing repetitive tasks to training agents to do them. He describes future human contribution as creating evals, updating skills files, and training agents—then positions Mercor as infrastructure for that transition.
- •The future of work is described as training agents once rather than repeating tasks
- •Enterprises are shifting toward eval creation and skill/task definition as core work
- •Mercor’s ambition is to define the category around how humans fit into the evolving economy
- •The company’s strategy is framed as long-term and category-defining from day one
Mercor’s three-product system: talent platform, Studio, and AI project management
Brendan breaks Mercor into three main products that together produce the data types customers need. He outlines the talent platform for sourcing experts, Studio for building RL environments with AI assistance, and an AI project manager that automates operations and quality management.
- •Three products are described as the core system for producing a given data type
- •Talent platform: over five million people looking for work, with strong referral-driven sourcing
- •Studio: experts use an AI copilot plus automated QC to populate RL environment components
- •AI project manager: agents set KPIs, give feedback, hire, assess quality, and own outcomes
Scaling payouts with Stripe Connect and building trust through reliable payments
Molly presses on how Mercor handles multi-million-dollar daily payouts, and Brendan goes straight to infrastructure and operations. He explains why payments reliability became existential after seeing another company’s public payout failures, and why Mercor built a direct relationship with labs to be accountable to experts.
- •Stripe Connect is used as the payments backend
- •A prior experience with another player’s payout issues drove Mercor to over-invest in payments reliability
- •Brendan highlights accountability to experts, including those relying on timely pay for essentials
- •He notes the company has paid out close to a billion dollars historically and scaled a dedicated payments team
Why Mercor is not a marketplace: delivering full RL environments
Brendan challenges the common framing of Mercor as a marketplace, arguing the deliverable is an RL environment rather than a match-and-fee transaction. He decomposes RL environments into apps, worlds, and tasks—positioning experts as only one ingredient in a broader system that must cover the economy’s full distribution.
- •Misconception addressed: Mercor is not like Uber-style matching with a fee on jobs
- •RL environments are described as the product delivered to customers
- •Three RL environment components: apps (app clones), worlds (content), tasks (prompts and rubrics)
- •Core challenge framed as covering the full distribution of apps/worlds/tasks that mirror the economy
Agents, productivity gains, and Jevons paradox expanding demand for human teaching
Molly asks how agent progress changes Mercor’s work, and Brendan argues agents increase expert productivity but also expand demand. He uses Jevons paradox to claim that making human teaching more efficient increases how much teaching the economy does, since much knowledge still lives in people’s heads.
- •Agents and copilots are said to multiply expert productivity in data creation
- •Brendan describes Jevons paradox: efficiency increases total demand for human teaching
- •He argues repetitive work will shift toward training models to automate it
- •He claims most knowledge remains in people’s heads, sustaining demand for extraction into training/evals
Is “human data” durable? Tying the category to superintelligence timelines and feedback loops
Responding to skepticism that the category is temporary, Brendan argues durability depends on how long humans can do things models can’t. He emphasizes that models need human feedback at the frontier—analogizing RL feedback to teachers grading essays—especially outside domains with clear automated signals.
- •Durability is framed as bounded by the gap between human capabilities and model capabilities
- •Examples given of areas requiring humans: persuasive conversations, wet lab work, car repair
- •Models need a “stasis point” and external feedback to learn, similar to human education
- •Math is noted as an exception where checking work can be easier than most real-world domains
Sourcing and evaluating experts: referrals and an AI interviewer at massive scale
Molly asks how Mercor finds and evaluates experts; Brendan highlights referrals as the dominant channel and describes building an AI interviewer early on. He argues labor matching is driven by scalable data collection, and shares the scale of interviews and the problem of talent density across geographies.
- •Referrals account for most expert sourcing, with a focus on quality
- •Brendan describes building an AI interviewer in college shortly after ChatGPT’s release
- •The AI interviewer process includes resume context, model questions, and rubric-based evaluation
- •He says Mercor has done over seven million interviews, enabling sourcing in low talent-density markets
Holding leadership with a young founding team: values, urgency, and non-mimetic strategy
Molly presses on how a young team maintains category leadership amid frequent new competitors and setbacks. Brendan attributes it to operationalized values—high standards, “time travelers” long-term thinking, and customer obsession—plus urgency without falling into competitor-driven mimicry.
- •Three values described: high standards, “time travelers,” and customer obsession
- •He emphasizes not compromising on hiring bar as a differentiator
- •He warns against mimetic competition as a path to being second place
- •Motivation is framed as building a legendary company solving how humans fit into the economy
Founder hardships and leadership as a shock absorber
Brendan describes constant setbacks in company building, including losing early customers and dealing with security incidents and team changes. He then explains how leaders should balance being demanding in good times and encouraging in hard times, and why the company focuses on long-term value drivers rather than weekly revenue graphs.
- •Founding is described as repeatedly getting “punched in the face” and operating amid chaos
- •Early customer churn is tied to starting with seed-stage companies, many of which fail
- •Leadership approach: crack down during highs; rally and encourage during lows
- •He prefers focusing the company on customer and expert health metrics over short-term revenue charts
Role models and a Bezos-related story about incentives and jobs
Molly asks who Brendan looks up to, leading to reflections on entrepreneurs and long-term thinking. Brendan shares a personal story about writing a college paper on incentives around job creation and taxes, then receiving engagement from Jeff Bezos after reposting related ideas.
- •Entrepreneurs he admires are discussed, including emphasis on mission orientation and long-term thinking
- •He describes a college paper thesis around incentives tied to job creation and taxation
- •He recounts Jeff Bezos posting about similar ideas and reposting his response
- •He notes he has not met Bezos in person and wants to eventually
Why Mercor acquired DeepTune: solving the “apps” bottleneck in RL environments
Molly shifts to current business state and asks about the DeepTune acquisition. Brendan explains the acquisition as a way to accelerate building app clones (the “apps” component) to complement Mercor’s strengths in worlds and tasks, noting the partnership was prompted by a customer recommendation and aligned teams in New York.
- •DeepTune acquisition is tied to completing the apps/worlds/tasks stack for RL environments
- •Mercor’s deepest competency is described as worlds and tasks; apps were a bottleneck
- •A customer suggested DeepTune as a top apps partner, catalyzing collaboration
- •Cultural alignment and founder fit are emphasized as key reasons for the deal
M&A strategy and turning down acquisition interest
After confirming this was not Mercor’s first acquisition, Brendan explains how the company thinks about M&A as high leverage. Molly asks about acquisition offers for Mercor; Brendan says there have been cases, but he and his co-founder are motivated by impact and the privilege of building in a great market rather than cashing out.
- •Brendan notes DeepTune was their second acquisition and larger than the first
- •He characterizes M&A as high leverage with both big risks and big upsides
- •He acknowledges there have been attempts to acquire Mercor
- •He frames independence as driven by mission and long-term opportunity, not money
Harvey, legal benchmarks, and “the model is the product” in app-layer companies
Molly brings up Harvey and LAB, asking about Mercor’s role as a partner. Brendan describes Mercor’s work as assembling experts and shaping management processes and data formats to produce high-quality datasets and benchmarks, then argues surviving app-layer companies must build frontier models in their domain because the model becomes the product.
- •Mercor’s contribution is framed as expert curation plus management and data-shape design for quality datasets
- •Public benchmarks mentioned include Apex and Apex Agents
- •He argues app-layer survivors must build frontier models in their domain to stay competitive
- •He links frontier labs to app-layer to enterprise as a cascade of adoption and capability
Moats across the AI stack, why data is mispriced, and the enterprise “Neo lab” future
Brendan lays out a three-layer view (frontier labs, inference providers, app-layer) and argues commoditization pressures make brand and data the most durable moats in the app layer. He then claims data is mispriced because of overly short superintelligence assumptions, and predicts Fortune 500s will become “Neo labs,” ramping evals and post-training as inference spend grows—while emphasizing data sovereignty and the need for task-and-eval-driven RL.
- •He argues software layers and GPU resale face commoditization; switching costs are central
- •For app-layer, he highlights two durable moats: brand and data (learning what customers care about)
- •He says investors have underestimated ongoing human involvement due to Jevons paradox and superintelligence timelines
- •He predicts enterprises will mobilize employees/contractors to build evals and tasks, becoming “Neo labs”
- •He emphasizes data sovereignty: companies won’t allow frontier labs to train on sensitive enterprise data
- •He advises enterprises to start with evals and build task/rubric-driven RL rather than expecting models to learn by reading everything
What’s next: IPO not imminent, enterprise acceleration, and how Brendan runs his days
In closing, Molly asks about the next 12 months and whether Mercor will go public; Brendan says not within that timeframe but it’s on the radar, and he’s most excited about enterprise adoption accelerating. He also describes his daily schedule, how he allocates time across hiring, customers, and research/product, and ends with Mercor hiring across experts and full-time roles.
- •Brendan says an IPO is not expected in the next 12 months, though it’s on the radar
- •He expects rapid acceleration as major non-technology companies recognize existential pressure and opportunity
- •He describes his schedule, long office hours, and time split across hiring, customers, and research/product
- •He discusses focus challenges, the need for strong leaders, and using agents for efficiency
- •Mercor is hiring both experts and full-time roles across engineering, research, and operations
Publisher timestamps
- 0:00Brendan Foody, Co-Founder & CEO Mercor
- 1:00Mercor’s Path to a $1T Company
- 2:22Mercor’s 3 Products
- 4:05$5M a Day in Expert Payouts on Stripe Connect
- 5:40Why Mercor Is Not a Marketplace
- 8:51Is Human Data a Durable Category
- 12:20How Mercor Finds and Evaluates Experts
- 15:45Holding Category Leadership With a Founding Team in Their Early 20s
- 18:00Chips on the Shoulder and Avoiding Mimetic Competition
- 28:25Why Mercor Acquired Deeptune
- 31:22Turning Down Acquisition Offers
- 34:40Harvey, LAB and Why the Model Is the Product
- 41:30Why Data Is Mispriced in the AI Stack
- 43:40Every Fortune 500 Becomes a Neolab
- 49:55IPO Timeline and the Next 12 Months