Sixteen Claude AI brokers working collectively created a brand new C compiler

Amid a push towards AI brokers, with each Anthropic and OpenAI delivery multi-agent instruments this week, Anthropic is greater than prepared to point out off a few of its extra daring AI coding experiments. However as ordinary with claims of AI-related achievement, you’ll discover some key caveats forward.

On Thursday, Anthropic researcher Nicholas Carlini printed a weblog submit describing how he set 16 situations of the corporate’s Claude Opus 4.6 AI mannequin unfastened on a shared codebase with minimal supervision, tasking them with constructing a C compiler from scratch.

Over two weeks and practically 2,000 Claude Code periods costing about $20,000 in API charges, the AI mannequin brokers reportedly produced a 100,000-line Rust-based compiler able to constructing a bootable Linux 6.9 kernel on x86, ARM, and RISC-V architectures.

Carlini, a analysis scientist on Anthropic’s Safeguards staff who beforehand spent seven years at Google Mind and DeepMind, used a brand new characteristic launched with Claude Opus 4.6 known as “agent groups.” In follow, every Claude occasion ran inside its personal Docker container, cloning a shared Git repository, claiming duties by writing lock recordsdata, then pushing accomplished code again upstream. No orchestration agent directed site visitors. Every occasion independently recognized no matter downside appeared most evident to work on subsequent and began fixing it. When merge conflicts arose, the AI mannequin situations resolved them on their very own.

The ensuing compiler, which Anthropic has launched on GitHub, can compile a spread of main open supply tasks, together with PostgreSQL, SQLite, Redis, FFmpeg, and QEMU. It achieved a 99 % cross charge on the GCC torture check suite and, in what Carlini known as “the developer’s final litmus check,” compiled and ran Doom.

It’s price noting {that a} C compiler is a near-ideal job for semi-autonomous AI mannequin coding: The specification is many years outdated and well-defined, complete check suites exist already, and there’s a known-good reference compiler to examine towards. Most real-world software program tasks have none of those benefits. The arduous a part of most improvement isn’t writing code that passes checks; it’s determining what the checks must be within the first place.