Rok Garbas

Nix

You can't make AI predictable. But you can make everything around it predictable. That changes more than you'd think.

AI agents learn by trial and error. Nix's atomicity, rollbacks, and project isolation make those errors cheap and reversible - turning mistakes into a feature, not a risk.

Technologies explored in 2016: Haskell struggles, Rust progress, Elm adoption at Mozilla, Python functional thinking, and daily Nix work. Year-end reflections.

After a year of implementing Nix-based DevOps at RhodeCode, I'm back on the job market. A retrospective on building reliable enterprise software installation with Nix.