Mercury runs 2 million lines of Haskell in production serving 300,000 businesses and $248B in 2025 transaction volume, with engineers mostly learning Haskell on the job.
Key Takeaways
Purity in Haskell is a boundary discipline, not a language property – runST is the canonical example of containing impurity behind a typed exit.
Encoding operational incantations (“flush audit log inside transaction”) in types prevents them from eroding as teams grow and senior engineers leave.
Mercury’s reliability philosophy centers on adaptive capacity – graceful degradation, operator legibility, blast-radius analysis – not just bug prevention.
Patrick McKenzie’s “half of your coworkers have under a year of experience” dynamic is a named design constraint at Mercury; type-encoded invariants survive team churn where wikis don’t.
Smaller ecosystem risk is real: Haskell library maintainers lack clear ownership, creating hesitation around breaking changes even when better designs are known.
Hacker News Comment Review
Commenters broadly agreed that encoding invariants in types is the core value, with debate over whether Rust and TypeScript deliver comparable power – OCaml’s GADTs and row polymorphism were cited as stronger than TypeScript’s nominal-type workarounds.
Productivity comparisons divided the thread: one engineer reported being 2x faster in Rust after 3 years of Haskell; another called switching a backend from TypeScript to Rust a major mistake due to higher-order function friction.
Several commenters attributed Mercury’s success as much to good engineering culture and being well-run generally as to Haskell specifically, and noted that hiring generalists with no prior Haskell helped Mercury impose its own style from day one.
Notable Comments
@1024bits: Hiring Haskell novices let Mercury instill culture and style from scratch – pre-vibe-coding, those hires wouldn’t have hacked away without instruction.
