Scale AI CEO Jason Droege, in an interview with Axios, stressed that artificial intelligence remains too unreliable for many mission-critical uses across business, military, and government sectors. "The cost of mistakes in these environments can be high," Droege said during the conversation from the company's San Francisco headquarters. The remarks come as Scale celebrated its 10th anniversary this week.
Droege, who took over from founder Alexandr Wang last June — when Wang became Meta's first chief AI officer and Meta acquired a 49% stake in Scale — is working to redefine the firm's identity. He aims to position it not just as a data annotation provider but as an AI infrastructure and deployment company. This strategic pivot was outlined in an internal memo titled "The Reliability Race," reported here for the first time.
In that memo, shared with more than 1,300 employees, Droege wrote that "reliability at this level depends on human intelligence." He emphasized that Scale's forward-deployed engineers ensure reliability for specific customer workflows and use cases. Droege told Axios he synthesized this message after extensive conversations with customers and prospects.
The CEO wants to cut through the noise around AI, claiming people "are talking about hundreds of topics around AI constantly." His focus on reliability draws on lessons from building earlier products. For Scale, the challenge lies in convincing skeptical enterprise and government clients that AI can be trusted for high-stakes operations.
A counterargument holds that no amount of human-in-the-loop engineering can guarantee 100% reliability in complex, unpredictable AI systems, making Droege's promise potentially overhyped. Critics note that even frontier models suffer from hallucinations and brittleness that human oversight cannot fully eliminate.