Why libraries, when anyone can generate code?
Because systems don't fail in the middle of a function — they fail at the seams. Between your logic and the cloud SDK. Between the schema Terraform provisioned and the schema the code writes. Between what passed on your laptop and what runs on a serialised worker with real IAM. Generated code multiplies the seams; it doesn't harden them.
A good library is where hardening accumulates: a stable contract, a conformance suite every implementation must pass, honest documentation of what doesn't work, and a signed, versioned artifact. That's not typing — that's judgment. And it's precisely what AI agents need to build against: we know, because we build our libraries with agents, and the only reason it works is the contracts and gates they're held to.
- Lines of code
- Boilerplate & scaffolding
- First drafts of anything
- Plausible-looking implementations
- Contracts that hold across clouds & years
- Tests that catch what emulators can't
- Production scars, written down honestly
- Signed artifacts a stranger can trust
Culvert
Data pipelines defined once against a language-neutral contract set — sixteen interfaces, two languages — and run on any cloud by swapping adapters, not rewriting logic. GCP today, AWS alongside it, Azure on the roadmap. Built in the open, deployed and tested on real infrastructure before every release.
“Nothing here is announced before it has run on real infrastructure. When code is cheap, the only scarce signal left is proof.”