As Anthropic's post points out, even as agents get better and build more of their own infrastructure, reliable shared tooling like this is still very valuable. Rederiving the same setup on every run is slow, expensive, and hard to trust for real science.

Our goal was to get