Anthropic recruits 1,000 engineers to sharpen Claude Code

Yesterday 15:06

Anthropic is carrying out a large-scale effort to improve the performance of its AI coding assistant Claude Code by relying on a network of around 1,000 contract software engineers. The initiative, managed through data company Snorkel AI, focuses on testing, evaluating, and refining code generated by the AI system. Engineers reportedly receive payments of up to $280 per task, with assignments typically requiring about an hour to complete.

Known internally as Project Marlin, the program relies on blind A/B testing methods. Participants compare code generated by two different versions of an AI model and determine which performs better based on criteria such as efficiency, readability, and maintainability. Because evaluators do not know which model version produced the output, the process is designed to reduce bias and provide more reliable feedback.

The assignments are based on practical software development challenges drawn from public repositories on GitHub. Engineers create realistic pull request scenarios involving feature additions, bug fixes, software restructuring, and security improvements. They then prompt Claude Code to generate solutions before assessing the results. The tasks mirror the daily responsibilities of professional developers and provide detailed feedback that can be used to strengthen future model performance.

The project reflects a broader shift across the artificial intelligence sector. Training and improving advanced AI systems increasingly requires highly skilled technical workers rather than traditional data labeling teams. Founded by researchers from Stanford University, Snorkel AI supports AI development projects for major technology companies including Google and Meta. Skilled engineers working on these programs can reportedly earn more than $3,000 per week, while firms such as Scale AI and Mercor have expanded similar operations worldwide.

Neither Anthropic nor Snorkel AI publicly commented on the reported project. However, the investment comes at a time when Claude Code is strengthening its position in the AI coding market. According to industry feedback from startup founders and venture capital investors, the tool has become a preferred coding assistant across many emerging technology companies.

Anthropic executives have also pointed to growing internal adoption. Company leaders stated that artificial intelligence now produces more than 90% of the code written within the organization. The company further reported that Claude Code generated $2.5 billion in revenue during its first year and captured a significant share of the AI coding market.

Project Marlin demonstrates that even the most advanced AI systems continue to depend on extensive human expertise. Behind the rapid progress of coding assistants stands a large workforce of engineers whose evaluations and judgments help shape tools that could transform software development in the years ahead.