Competitors Bundling Environments and Labor
Fleet
This competitive set shows that RL environments are quickly becoming a bundle sale, not a standalone category. Surge AI, Mercor, and Turing already sell labor, evaluation, and workflow infrastructure into the same frontier labs, so they can attach environments to existing budgets and buyer relationships. That means Fleet is not just competing on environment quality. It is competing against vendors that can package experts, benchmarks, and post training operations into one procurement motion.
-
Surge is the clearest example of a labeling vendor moving upstack. It has turned human annotation into public research assets like EnterpriseBench and CoreCraft, with 2,500 plus entities, 23 tools, and stated partnerships with OpenAI, Anthropic, Meta, and Google. That makes its environments a natural extension of work many labs already buy from it.
-
Mercor and Turing approach the same opening from different starting points. Mercor begins with a marketplace of vetted doctors, lawyers, bankers, and other specialists, then layers on benchmarks like APEX and APEX-Agents plus enterprise workflow tooling. Turing begins with a software talent platform, then packages Docker based UI and MCP environments with verifier scoring for frontier labs.
-
The broader pattern is that data vendors are trying to become training infrastructure vendors. Handshake has pushed from contractor supply into evaluations and RL environments, and Scale says nearly half of new training projects now involve RL environments. The market is moving from paying for labeled answers toward paying for full simulated tasks, scoring, and improvement loops.
The next phase favors companies that control both the human layer and the environment layer. As labs buy larger end to end post training programs, vendors that can source experts, generate tasks, score outputs, and feed the results back into training will capture more spend. Fleet is heading into a market where the winning product will look less like a dataset and more like a repeatable factory for agent improvement.