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, CEO of Scale AI, discusses the company's shift from generalist data labeling to expert-driven training data and evals for AI labs and enterprises. The conversation also covers the Meta investment, how models are moving from knowing things to doing things, and lessons from his career at Scour, Uber Eats, and running new businesses.

01Key learnings

• Model improvement now relies heavily on experts (80% of Scale's expert network holds a bachelor's or higher) defining what good looks like, so evals and expert judgment are core work.
• Enterprise AI pilots that reach 60-70% accuracy feel close, but robust automation of important processes often takes 6-12 months of work on accuracy, compliance, and change management.
• Validate a new business by checking whether it can sustain high gross margins and whether competitors can match the economics in two or three years; this filters out weak ideas quickly.
• Build a product for the customer's urgent daily priorities, not just their annual pain points, or adoption will stall even if the value is real.
• Figure out what the customer's incentives actually are, including ego and career growth, and tailor your pitch and proof to them rather than taking stated requests literally.
• When hiring, screen for curious problem solvers who collaborate well and lead, and balance team chemistry and complementary strengths over matching exact prior scale experience.
• Start a venture from genuine insight and a multi-year commitment, not from a vague customer problem; expect to pivot and persist through long stretches of difficulty.
• Use AI as a personal tutor to learn fast-moving technical concepts and to surface key points from dense internal documents.

02Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.