Anthropic has pushed the boundaries of AI coding with its Claude Opus 4.6 model. Nicholas Carlini, a researcher at Anthropic, reported that by using 16 instances of Claude, the team created a C compiler from scratch. This project spanned over two weeks and involved nearly 2,000 coding sessions, costing around $20,000 in API fees. The end result was a Rust-based compiler that can build a bootable Linux 6.9 kernel for x86, ARM, and RISC-V architectures, consisting of about 100,000 lines of code. Each Claude instance operated in its own Docker container, working on a shared codebase without a central coordinating agent. They managed tasks by creating lock files and pushing code updates independently. This setup allowed the AI to identify problems and resolve merge conflicts autonomously. The compiler has been made available on GitHub and demonstrates impressive capabilities, achieving a 99 percent pass rate on the GCC torture test suite and successfully compiling and running the classic game Doom. The success of this project highlights the strengths of AI in specific coding tasks like creating a C compiler, where the specifications are clear and well-established. Unlike most software projects, which often lack comprehensive testing frameworks, this task has extensive resources and reference compilers that guided the AI's development process. The challenges of real-world software development typically involve defining what needs to be tested, rather than merely writing functioning code.