Generalist Market Pressure on Coding Specialists
Datacurve
The market is splitting into cheap, scalable human labor and expensive, high signal software expertise, and that split is exactly where pricing pressure shows up. Scale AI, Surge AI, and Mercor can reuse large recruiter networks, QA systems, and workflow software to sell coding tasks at lower prices, but their core machinery is built to route and score labor at volume, not to generate hard, repo level engineering work like Datacurve’s test backed code tasks and agent traces.
-
Scale already sells bundled labeling, self serve tooling, data management, synthetic data, and workflow automation, and reached $1.5B ARR by end of 2024. That lets it discount coding work as one line item inside a much larger relationship, especially with customers that already buy broader model training infrastructure.
-
Surge and Mercor show two different generalist attacks on specialists. Surge uses a 50,000 contractor network and managed RLHF workflows to sell premium expert tasks at scale, while Mercor uses AI screening across 300,000 professionals to source domain experts fast, even if much of its revenue is still gross pass through spend.
-
The category has moved away from Mechanical Turk style annotation toward credentialed experts. That helps specialists, but it also invites adjacent marketplaces into coding data. As reasoning models demand narrower expert input, more recruiting and expert network platforms can repurpose their supply into post training work.
Going forward, specialists win by owning the hardest workflows, not by matching generalists on price. The durable segment is high consequence coding data where customers need reproducible tests, clean provenance, and engineers who can actually solve the task, which pushes the market toward fewer commodity labeling vendors and more purpose built engineering data suppliers.