and launched its own service for

Scale moving into expert hiring showed that high skill human labor had become part of the core AI stack, not a side service. Once labs needed doctors, lawyers, scientists, and senior coders to judge hard model outputs, the winner was no longer just the company with the biggest labeling workforce. It was the company that could source, vet, route, and manage scarce specialists fast, then plug them directly into evaluation and post training workflows.

• Scale had already built the operating system for this kind of work. Its test and evaluation framework used generalist red teamers first, then escalated edge cases to domain experts in fields like biology, medicine, economics, cybersecurity, and nuclear physics, which made expert staffing a natural extension of its existing workflow engine.
  3 scale 4 outlier
• In practice, Expert Match sat closer to managed staffing than classic marketplace recruiting. Outlier roles show the mechanics, identity verification, skill assessment, task routing, hourly pay, and project based work where experts create prompts, rank outputs, and write feedback that becomes training data for frontier labs.
  2 scale 4 outlier
• That put Mercor into a tougher lane than generic freelance hiring. Mercor was growing by building a dedicated marketplace for pre vetted experts and reached $50M ARR by the end of 2024, but Scale entered from the other side, starting with customer demand, labeling operations, and nearly $950M annualized revenue by August 2024.
  1 sacra 5 sacra 6 sacra

The next step is deeper vertical integration, where expert networks, evaluation software, and training data production merge into one product. As models push further into law, medicine, finance, and coding, platforms that can both find the right expert and capture that expert's judgment as reusable data will take a growing share of AI spend.