Anthropic taps 1,000 engineers to fine-tune Claude Code

Anthropic enlisted roughly 1,000 freelance engineers through data vendor Snorkel AI to A/B test and refine Claude Code outputs, according to Business Insider.aol
The project, code-named “Marlin,” has contractors compare code from different model versions using real-world GitHub scenarios without knowing which version they’re grading.aol
Claude Code reportedly generated $2.5 billion in first-year revenue and now dominates the AI coding market, per Anthropic’s head of design.youtube

Anthropic Taps Snorkel AI and 1,000 Engineers to Fine-Tune Claude Code

Anthropic is running a large-scale, behind-the-scenes operation to sharpen its AI coding tool, enlisting roughly 1,000 contract software engineers through data vendor Snorkel AI to evaluate and improve Claude Code’s output, according to a Business Insider report published May 31.aol

Project Marlin

The effort, known internally at Snorkel AI as “Marlin,” pays freelance engineers up to $280 per task to create coding prompts and review AI-generated code, with each task typically taking about an hour. The work centers on A/B testing: contractors compare code produced by two different model versions, selecting which output is preferable based on efficiency, clarity, and maintainability. Contractors are not told which model version they are evaluating, ensuring objectivity in their assessments.aol

Tasks draw from real-world development scenarios using GitHub repositories. Engineers create pull request scenarios involving feature additions or bug fixes, prompt Claude Code to generate solutions, and then grade the results. Examples include restructuring a system’s metadata processing or implementing security measures for a machine learning platform — work that mirrors what professional developers encounter daily.hyper

The Human Infrastructure Behind AI

The project illustrates a broader industry trend in which the demand for AI training labor has shifted from simple data labeling toward tasks requiring advanced technical expertise. Snorkel AI, founded in 2019 by Stanford researchers, supports AI model development for major technology companies including Google and Meta. Skilled engineers working on such projects can earn more than $3,000 per week, and similar platforms such as Scale AI and Mercor operate globally.hyper

Both Snorkel AI and Anthropic declined to comment on the project.hyper

Claude Code’s Growing Dominance

The investment comes as Claude Code consolidates its position in the AI coding market. A Business Insider survey of more than two dozen startup founders and venture capitalists found a growing consensus that Claude Code has become the default AI coding tool inside startups. Anthropic’s CFO Krishna Rao said in May that AI now writes more than 90% of the company’s own code. Claude Code’s creator has said the product generated $2.5 billion in revenue in its first year and captured 51% of the coding market.businessinsider

The Marlin project underscores how even the most advanced AI systems still depend on large pools of human expertise to reach professional-grade performance — an invisible workforce shaping the tools that may eventually reshape software engineering itself.