Lenny's Podcast: Product | Career | Growth

First interview with Scale AI’s CEO: $14B Meta deal, what’s working in enterprise AI, and what frontier labs are building next | Jason Droege

October 9, 2025

Key Takeaways Copied to clipboard!

  • The general trend in AI is shifting from models that know things to models that can actively do things within complex, real-world environments. 
  • Scale AI's business has evolved from generalist data labeling to highly specialized, expert data labeling and creating evals, with 80% of their expert network holding a bachelor's degree or higher. 
  • Successful new businesses require founders to possess a unique, independent insight (alpha) and focus on solving the customer's most urgent daily problems, not just valuable occasional ones. 
  • Pushing back against initial offers, like the one from McDonald's, can sometimes lead to securing a significantly better deal later on, as demonstrated in the building of Uber Eats. 
  • High gross margins combined with healthy churn curves are a strong indicator of business health, serving as a crucial litmus test for assessing value addition and differentiation, though exceptions like Costco exist. 
  • For long-term success, especially in volatile environments like tech, survival is a precursor to winning, emphasizing the importance of making asymmetrically positive decisions and avoiding unnecessary existential risk. 

Segments

Scour Co-founding Lesson
Copied to clipboard!
(00:06:30)
  • Key Takeaway: In business and startups, everything is negotiable, a lesson learned early while navigating aggressive financing terms.
  • Summary: Co-founding Scour taught Jason Droege that there is no fixed way to conduct business; one must negotiate through the world by aligning incentives. This experience, including being sued for a quarter of a trillion dollars, imprinted the reality that established players may invent rules during conflict. Learning this at a young age influenced a mindset where possibilities are defined by negotiation rather than rigid structures.
Scale AI Post-Meta Deal
Copied to clipboard!
(00:10:27)
  • Key Takeaway: Scale AI remains a fully independent company following the Meta investment, which secured a 49% non-voting stake for $14 billion.
  • Summary: The transaction involved Meta investing over $14 billion for a non-voting stake, with only about 15 employees transferring to Meta. Scale AI maintains its independent governance structure and has two major business lines, each generating hundreds of millions in revenue. The company has experienced month-over-month growth since the deal closed.
Evolution of Data Labeling
Copied to clipboard!
(00:12:40)
  • Key Takeaway: AI training data needs have rapidly shifted from basic preference rankings to complex tasks requiring PhDs and professionals, such as building entire websites.
  • Summary: Scale AI has continuously adapted its data focus, moving from computer vision labeling to Gen AI needs, which now demand expert input. Tasks have escalated from simple comparisons (e.g., which short story is better) to multi-hour efforts like explaining nuanced cancer topics or developing full websites. Currently, 80% of Scale’s expert network holds a bachelor’s degree or higher, contrary to some market narratives.
Expert Network Acquisition Strategy
Copied to clipboard!
(00:17:29)
  • Key Takeaway: The most valuable experts for AI contribution are acquired primarily through grassroots and referral networks, driven by a positive experience and the interest in contributing to AI.
  • Summary: Finding experts is difficult, requiring multiple tactics beyond traditional sourcing like LinkedIn. The largest source of experts comes from referrals because they enjoy using their expertise to contribute to AI development. Providing a great experience is crucial for retention, as experts are motivated by both compensation and the intellectual interest in solving AI model gaps.
AI Agents and RL Environments
Copied to clipboard!
(00:19:09)
  • Key Takeaway: The future of AI involves agents learning in Reinforcement Learning (RL) environments, requiring data that is generalizable across vast permutations of software systems and user goals.
  • Summary: RL environments are sandboxes where AI agents learn to accomplish goals, such as navigating a complex, configurable system like Salesforce. Scale’s research focuses on creating data that is generalizable across many use cases to avoid needing to train on every possible combination of environment variables. The goal is to provide model builders with the most valuable data to make agents useful for end-users.
AI Adoption Timeline and Hype
Copied to clipboard!
(00:35:26)
  • Key Takeaway: The next major frontier for AI models in the next two to three years will be the shift from models knowing things to models actively doing things, which will then push change management policy.
  • Summary: The current hype cycle often overlooks the operational reality: getting sophisticated AI systems robust enough for important enterprise automation takes six to twelve months of dedicated work. While POCs are easy to launch (leading to high failure rates in initial studies), achieving the necessary reliability for critical workflows requires significant time investment. The technology is approaching a point where its capabilities will force policy makers to address adoption challenges.
Uber Eats Economics Insight
Copied to clipboard!
(00:41:46)
  • Key Takeaway: When building new products, focus on the customer’s urgent daily incentives, as solving a high-value but occasional problem may fail if it doesn’t address immediate operational concerns.
  • Summary: Droege discovered restaurant economics by independently weighing sandwich ingredients to establish ground truth, finding that incremental demand offers 70-80% incremental gross margin for restaurants. This insight gave confidence to charge a high commission rate because the primary customer incentive was securing new, incremental sales volume. Urgency is key: if a problem isn’t top-of-mind daily for the buyer, adoption will be slow, regardless of the solution’s long-term value.
Independent Thinking and Founding
Copied to clipboard!
(00:48:19)
  • Key Takeaway: Founding a successful business requires seeking market alpha through independent insight, which means questioning external consensus and having a burning desire to solve the specific problem for 5-10 years.
  • Summary: To gain alpha in a crowded market, entrepreneurs must rely on insights others do not possess, often requiring contrarian thinking. Founders must be passionate about the problem itself, not just the customer’s stated issue, to sustain the long effort required for success. Independent thinking also demands the ability to discard one’s own ideas when they conflict with the mission of serving the customer.
Setting High Bar for New Ventures
Copied to clipboard!
(00:50:45)
  • Key Takeaway: New businesses have the best chance of success when the founder is a persistent force of nature, coupled with an understanding of fundamentally good business models like marketplaces or sticky SaaS.
  • Summary: Successful ventures often rely on the founder’s sustained energy to pivot over many hard years, but filtering ideas based on proven business models increases the odds. Good models typically exhibit network effects, lock-in, and increasing value at scale, which VCs favor. Eliminating bad ideas early via this filter allows founders to focus their passion on the remaining viable opportunities.
Uber Eats and McDonald’s Deal
Copied to clipboard!
(00:56:39)
  • Key Takeaway: Stubbornly pushing back on McDonald’s initial approach helped Uber Eats secure an exclusive relationship and a better deal.
  • Summary: Jason Droege left his role before COVID, noting the massive growth in ride-sharing and food delivery. He recounted how McDonald’s approached Uber Eats for delivery, but he initially refused for several months, which his team thought was insane. This delay ultimately resulted in an exclusive relationship and an ‘insane number of customers’ when they finally agreed to partner.
Gross Margins Feasibility Filter
Copied to clipboard!
(01:00:13)
  • Key Takeaway: Gross margin analysis is a coarse but quick filter to test if a business idea adds sufficient value or is differentiated.
  • Summary: Founders should be obsessed with gross margins as a litmus test for value addition, though low margins are acceptable for businesses like Costco or Walmart that leverage scale or membership fees. If a proposed margin seems low, asking why a 60% margin doesn’t work reveals the true competitive constraint, often pointing to alternatives with lower margins like offshoring companies.
Not Losing Precursor to Winning
Copied to clipboard!
(01:04:51)
  • Key Takeaway: The best entrepreneurs focus on making asymmetrically positive risk decisions rather than blindly ‘going for it’ in hype cycles.
  • Summary: The concept of ’not losing is a precursor to winning’ counters the tech culture narrative that encourages reckless risk-taking. Survival is essential for long-term success, meaning founders must calculate risk profiles to ensure they remain viable to solve customer problems years down the line. An example of failing to think through risk was selling used golf clubs online, where hubris led to ignoring margin compression from easy entry.
Hiring for Team Composition
Copied to clipboard!
(01:09:12)
  • Key Takeaway: Hiring should prioritize curiosity, collaboration, and leadership over specific scale experience for most roles, focusing on team synergy.
  • Summary: While specialized roles (like AI researchers) require immediate expertise, most hiring should focus on three core traits: being a curious problem solver, humility for collaboration, and good leadership. The management team for Uber Eats remained consistent from zero to $20 billion because their mutual understanding of each other’s strengths minimized conflict, proving team synergy outweighs lacking prior scale experience.
AI Use in Daily Work
Copied to clipboard!
(01:12:11)
  • Key Takeaway: AI is effectively used as a personal tutor to rapidly learn complex, fast-moving technical concepts outside one’s core expertise.
  • Summary: Jason Droege uses AI as a tutor to keep up with the rapidly evolving technical nuances of the AI space, which is crucial for a CEO managing an AI company. Another impactful use is summarizing internal documents to quickly identify the most important information, cutting through organizational noise and individual agendas.
Lightning Round Insights
Copied to clipboard!
(01:15:36)
  • Key Takeaway: The motto ‘The end is never the end’ helps unlock perseverance during impassable entrepreneurial challenges.
  • Summary: Recommended books include The Selfish Gene, The Road Less Traveled, Good to Great, and Thinking, Fast and Slow, emphasizing psychology and proven business analysis. Jason Droege’s favorite motto, ‘The end is never the end,’ reminds him that survival is possible even when facing seemingly impassable decisions, allowing him to seek imperfect solutions rather than giving up.