Anthropic reportedly put 16 AI agents to work on one of the hardest challenges in computer science: building a C compiler from scratch. The result? • ~100,000 lines of code • Compiles the Linux kernel • Passed 99% of GCC’s torture tests • Runs complex software like FFmpeg, Redis, PostgreSQL, QEMU — even Doom • Took about 2 weeks and roughly $20K For context, GCC took decades and thousands of engineers to reach its level. But here’s the detail people are missing: No human wrote the code — yet the human role didn’t disappear. The researcher constantly redesigned tests, built CI pipelines when agents conflicted, and stepped in whenever all 16 agents got stuck on the same bug. The job didn’t vanish. It evolved. Humans are no longer just writing software… they’re designing the systems that let AI write it. Hard to argue AI is “hitting a wall” when experiments like this exist. Are we entering the era of AI-built infrastructure? 👀💬 anthropic | ai agents | coding ai | future of so...