Nicholas Carlini
Research Scientist at Anthropic
Anthropic researcher who led the 16-agent C compiler experiment — a proof point for parallel AI agent coordination producing 100,000 lines of production code.
About Nicholas Carlini
Nicholas Carlini is a researcher on Anthropic’s Safeguards team, known for work at the intersection of AI safety and capability evaluation. In February 2026, he led an experiment that became a landmark demonstration of multi-agent software development: 16 Claude agents building a C compiler from scratch.
The C Compiler Experiment
Carlini tasked 16 parallel Claude instances with writing a Rust-based C compiler capable of compiling the Linux kernel. Over nearly 2,000 Claude Code sessions and $20,000 in API costs, the agent team produced 100,000 lines of code that passes 99% of GCC’s torture test suite.
Key innovations in the approach:
- GCC as oracle — used the established compiler as an automated verification standard
- Git-based locking — prevented agents from duplicating work on the same files
- No internet access — clean-room implementation depending only on Rust’s standard library
- Parallel decomposition — each agent handled distinct compilation targets independently
Notable Positions
On Agent Verification
“Claude will work autonomously to solve whatever problem I give it. So it’s important that the task verifier is nearly perfect, otherwise Claude will solve the wrong problem.”
This insight — that the quality of the verification mechanism matters more than the quality of supervision — has become a foundational principle for multi-agent system design.
Video Mentions
Multi-agent C compiler experiment
Claude will work autonomously to solve whatever problem I give it. So it's important that the task verifier is nearly perfect, otherwise Claude will solve the wrong problem.