Sixteen Claude AI agents working together created a new C compiler (arstechnica.com)

In a groundbreaking experiment, Anthropic has successfully deployed 16 instances of its Claude Opus 4.6 AI model to autonomously develop a C compiler, demonstrating the potential of AI agents in software development. Under the guidance of researcher Nicholas Carlini, these AI agents collaborated over two weeks and approximately 2,000 coding sessions, resulting in a 100,000-line Rust-based compiler capable of building a bootable Linux 6.9 kernel on various architectures, including x86, ARM, and RISC-V. Remarkably, the project incurred around $20,000 in API costs. This achievement is significant for the AI/ML community as it showcases the effectiveness of multi-agent collaboration in complex coding tasks, leveraging a new feature called “agent teams.” Each Claude instance operated independently within Docker containers, managing their tasks and resolving coding conflicts without centralized oversight. The resulting compiler not only compiled popular open-source projects like PostgreSQL and FFmpeg but also excelled in tests, achieving a 99 percent pass rate on the GCC torture test suite and even compiling the iconic game Doom. While the project highlights the capabilities of AI in coding, it also emphasizes the challenges of real-world software development, where problem definition and testing are often more complex than merely writing code.