Agents write code.
Your team needs to know it is correct.

Generated code arrives faster than any human review process can absorb. The bottleneck is not skill; it is architecture. Reviewing each diff harder doesn't scale. Shifting the contract one layer down does.

We work on the verification layer underneath an agent pipeline: where configuration is typed, where the build fails before a derivation runs, where the same description drives validation, documentation, and dependency analysis. We start with a structural review. Two or three sessions, no code changes. The point is to find where a production failure became visible too late, and trace it back to the missing check.

Where the work fits: Nix module systems whose validators have started to feel ad-hoc; deployment pipelines where one team's change silently breaks another team's invariants; agent-coordination layers where an audit trail is needed but the existing event log can't carry one. The work is structural. The language is whatever fits — Nix where the problem is Nix, your stack everywhere else.

We're currently running the complete technical infrastructure for the Lie-Størmer Center, Norway's national center for mathematics, out of Tromsø. The same stack runs this site.

Lie-Størmer-senteretUiT The Arctic University of NorwayAMTI

We built and maintain nix-effects, an open Nix toolkit for typed descriptions and generic walkers (github.com/kleisli-io/nix-effects).

If something here matches, the fastest next step is an email describing what broke and when.