Google open-sourced AX on September 18, 2026, unveiled during a livestreamed I/O developer session: a high-throughput, declarative orchestrator designed to run billions of autonomous agent workloads inside a single cluster, built on top of Google's Agent Substrate for sandboxed execution. It's Apache 2.0-licensed and lives at github.com/google/ax -- a major AI lab putting a production-grade agent-orchestration runtime into the open, rather than keeping it as internal infrastructure.
The numbers moved fast: the repo gained +2,305 stars in a single day (today), on top of roughly 1,968 stars and 623 commits reported earlier in the week, and Google says more than 150,000 developers registered on the project's GitHub page within 48 hours of launch.
Reception hasn't been uniformly enthusiastic -- coverage specifically notes Hacker News pushback on the "billions of agent workloads" framing in Google's own description, with skepticism about whether that number reflects real current usage anywhere versus aspirational design capacity for a system that's five days old. Worth treating the star count as developer curiosity, not validation, until production workloads actually show up on it.
The timing is notable alongside yesterday's entry on trycua/cua's "computer-use 2.0" trending run: two different companies, one week apart, both betting that the infrastructure around running AI agents -- not the agents' own reasoning -- is what's actually still unsolved and worth standardizing next.
Google open-sourcing a production-grade agent orchestrator drew real developer attention (2,305 stars in a day, 150k+ registrations in 48 hours) but real validation is unproven workloads, not stars -- and the 'billions of agents' framing is already drawing skepticism worth tracking alongside the adoption numbers.