Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Nicholas Carlini ran 16 Claude agents in an endless loop, each in its own Docker container, pushing to a shared git repo. Agents claimed tasks by writing lock files, merged each other's work, and had no orchestrator. Carlini's main work was building tests, CI, and a GCC-based oracle so agents could fix different kernel bugs in parallel.
About 2,000 sessions produced a 100,000-line Rust compiler that builds Linux 6.9 on x86, ARM and RISC-V and passes 99% of most test suites. It still calls GCC for 16-bit x86, its assembler and linker are buggy, and its output is inefficient.
It cost just under $20,000 in API fees over two weeks, and the quality of the test harness decided whether agents stayed on track. Run it in a container.
I want to build a small [tool, e.g. a Markdown-to-HTML converter in Python] from scratch with a spec-first approach. First write a short spec and a test suite in this folder, then split the work into independent pieces and use sub-agents to implement them in parallel. Stop when all tests pass, and give me a summary of what each sub-agent built and any tests you had to skip.
Score every demo request against your ICP, update HubSpot and ping the right rep in Slack within a minute of the form submit.
Every new Zendesk ticket gets a category, priority and sentiment tag plus a draft reply in an internal note before an agent opens it.
Copilot and Claude review each PR for bugs, missing tests and risky changes before a human looks, so reviewers can focus on design.