Lenny’s Podcast: Product | Career | Growth

Episode: 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

Date: October 9, 2025
Host: Lenny Rachitsky
Guest: Jason Droege, CEO of Scale AI

Episode Overview

Lenny sits down with Jason Droege, the new CEO of Scale AI, for Jason's first public interview since taking over after the massive $14B Meta investment and the departure of Alex Wang. The conversation explores the state and future of data labeling and expert input in AI, the reality of building enterprise AI products, how Scale fits in after the Meta deal, and lessons from Jason’s journey building Uber Eats and other ventures. The wide-ranging talk is packed with tactical insights for founders and product leaders, plus deep reflections on entrepreneurship, team-building, and adapting to technological change.

Key Discussion Points & Insights

1. The Realty of AI in Enterprise (00:00–00:26; 37:35–40:34)

• Hype vs. Reality: Enterprises often find that “on the ground,” AI adoption takes much longer and requires more effort than headlines suggest.
• Quotable:

  "With any of these major tech revolutions, the headlines tell one story and then on the ground, laying broadband means you need to dig up every single road in America to lay it." — Jason Droege (00:06 & 34:21)

• Innovation Timeline: True, impactful AI deployments in enterprises can take up to a year for a single process. The last 10–20% of accuracy is disproportionately difficult to achieve.
• On POCs Failing: Rapid proliferation of pilots is misleading; with focus, robust automation is possible, but hype cycles cloud realistic expectations.

2. Post-Meta Deal: What’s Changed at Scale AI? (10:55–12:26)

• Scale remains a fully independent company.
  • Meta invested $14B for 49% non-voting share; did not get board control, and Alex Wang has left to lead at Meta.
  • Only ~15 employees moved to Meta; rest of company (~1100 employees) continues to operate independently.
• Scale has two major businesses, each in the hundreds of millions in revenue—"two unicorns inside the company."

3. The Evolving World of Data Labeling & Expert Networks (13:27–28:42)

• From Generalists to Experts:
  • Early AI models relied on low-cost, generalist labor for data labeling.
  • Now, advanced models demand hours of work from highly-trained professionals (PhDs, doctors, lawyers, engineers).
  • Stats: 80% of Scale’s expert network have bachelor’s or higher; ~15% hold PhDs.
• Finding Experts:
  • Multi-pronged approach: referrals, campus outreach, traditional recruiting channels.
  • Value for contributors: meaningful work, good pay, and the chance to directly improve AI in their field.
  • Quotable:

    “A lot of times they refer each other. We also have campus programs... But the best ones come from these grassroots and referral networks.” — Jason Droege (17:33)

• AI’s Need for Human Judgment:
  • As AI attempts increasingly complex tasks, the expertise required becomes more individualized (e.g., a specific doctor labeling nuanced healthcare data for their hospital's needs).
  • Human judgment in context—beyond "correct," towards "what does good look like"—is the bottleneck AI can't (yet) transcend.
  • Quotable:

    "Digitizing true subject matter, deep expertise is becoming a bottleneck we’re unblocking for our customers." — Jason Droege (27:01)

4. The Ongoing Need for Human Contribution (28:18–31:13)

• Data labeling for AI is a "history of new beginnings"—as one area requires less, new challenges and domains arise needing human input.
• Prediction: As long as novel human knowledge exists, people will be needed "in the loop" to help AI adapt.
• Optimism about the future:
  • Human adaptability is underestimated—our ability to pivot and find new meaning is a powerful counterweight to “white collar apocalypse” fears.
  • Quotable:

    "If you’re effectively saying no new human skill or knowledge is important enough to put into these models… that feels pretty far out there." — Jason Droege (28:42)

5. How AI Is Being Made Smarter—Concrete Examples (22:03–23:33; 33:48)

• Example tasks:
  • Engineers building and annotating entire websites for a model.
  • Doctor in a specialist hospital labeling what diagnoses are important in a massive patient file for decision-making.
  • AI agents learning to navigate environments like Salesforce.
• Evals—establishing "what good looks like"—are core. Particularly in regulated, high-stakes domains (healthcare, insurance, government), experts define benchmarks that models must meet.

6. AI Models: From Knowing to Doing (35:25–37:35)

• The Coming Era: Models are shifting from knowledge retrieval to decision-making and action.
• AI agents will increasingly learn to interact with complex business systems, making operational decisions for users.
• Adoption will depend as much on organizational change management as technological progress.
  • Quotable:

    "The general trend right now is going from models knowing things to models doing things." — Jason Droege (35:47)

7. Lessons in Product and Company Building

a. Customer Obsession & Incentive Alignment (42:33–48:19)

• Direct customer insight is key, but not always obvious or spoken. Understand incentives (often not just financial).
• Uber Eats example: Rather than take restaurateurs at their word, Jason’s team reverse-engineered restaurant economics to find real “incremental demand” value.
• Focus on customer urgency and day-to-day pain and align your offering accordingly.
  • Quotable:

    “Show me the incentive and I’ll show you the outcome.” — Jason Droege (43:12)

b. Independent Thinking and Alpha (48:19–50:43)

• The founder’s most important job: cultivating and acting on unique, independent insights.
• Don’t start a business or project just based on customer problem—have a “burning desire” for the solution and be willing to challenge your own assumptions continually.

c. Setting a High Bar for New Ventures (51:07–53:04)

• Surviving and thriving as a founder requires either being a "force of nature" or deliberately picking the right business model and market from the start.
• Look for business models with recurring revenue, network effects, and significant opportunity for differentiation.

d. Gross Margins as a Litmus Test (60:33–64:49)

• High gross margins + healthy customer retention indicate differentiated value.
• Use gross margin as a filter, not a definitive answer—ask “why can’t it be higher?” to surface unseen competitive pressures or business risks.

e. Survival Before Thriving—On Not Losing (64:49–68:35)

• "Tech loves the just-go-for-it mentality, but if I only have a finite number of shots, sometimes I want to survive first."
• Founders who endure are often those who minimize downside enough to "be around for success."
  • Quotable:

    "Survival is a precursor to thriving." — Jason Droege (65:02, 79:57)

8. Reflection on Team Building & Hiring (68:51–72:11)

• Right team composition trumps hiring the most “optimal” individual talent.
• Focus on:
  1. Curiosity and problem-solving ability.
  2. Humility and collaboration.
  3. Leadership.
• The best teams scale together and complement each other’s strengths and weaknesses.
• Quotable:

  "The team knowing each other's strengths and weaknesses and being able to compensate for each other was more important than all the classic advice." — Jason Droege (71:21)

9. Personal Use of AI & Continuous Learning (72:11–74:39)

• Jason uses AI as a daily tutor—voice mode commuting, internal doc summaries—to keep up with the fast-evolving space.
• Sign of workplace change: At Perplexity, employees must “ask AI first” before asking a colleague.

10. Lightning Round & Final Thoughts (75:39–81:56)

Highlights:

• Book Recommendations:
  • The Selfish Gene (Richard Dawkins)
  • The Road Less Traveled (M. Scott Peck)
  • Good to Great (Jim Collins)
  • Thinking, Fast and Slow (Daniel Kahneman)
• Recent Product:
  • VO3—AI turning a photographed script into a scene.
• Life Motto:
  • "The end is never the end… Surviving is the precursor to thriving." (79:57)
• Favorite Uber Eats Order:
  • McDonald’s (for the family treat), with a plug for mixed/tender greens for daily eating.

Notable Quotes & Memorable Moments

• "Everything in business is negotiable." – Jason Droege on early startup lessons (06:43)
• "AI’s need for expert human data is not going away anytime soon." (28:42)
• "Digitizing true subject matter, deep expertise is becoming a bottleneck we’re unblocking." (27:01)
• "Survival is a precursor to thriving." (65:02, 79:57)

Timestamps for Important Segments

• 00:00 – AI’s real progress in enterprise
• 10:55 – The reality behind the Meta–Scale AI deal
• 13:27 – Evolution from generalist to expert data labeling
• 22:03 – Real-life examples of expert data for AI
• 28:18 – How long will humans be needed to train models?
• 35:25 – Where models are heading in the next 2–3 years
• 42:33 – Lessons on customer obsession from Uber Eats
• 51:07 – The bar for new businesses and founder alpha
• 60:33 – Thinking about gross margins
• 65:02 – Survival as a precursor to winning
• 69:31 – Principles of team building and hiring
• 72:11 – How Jason uses AI as a daily tutor
• 75:39 – Lightning round: books, products, mottos

Closing Summary

Jason Droege’s first public interview as Scale AI CEO is a masterclass on how enterprise AI, data labeling, and expert-driven models really work in practice. The conversation is rich with practical advice—on adapting as a founder, surviving hype cycles, building high-functioning teams, and evolving products with customer urgency and gritty, unconventional research. Listeners are left with a frank sense of how slow, operational, and expert-driven genuine AI progress is beneath the hype, and how history’s biggest tech leaps are always “dug up” one painful, well-negotiated trench at a time.

Find Jason:

• X (Twitter): @jdroege
• Scale Careers: scale.com/careers

Recommended for: Founders, product leaders, enterprise tech builders, and anyone serious about the real (not just theoretical) future of AI.