When David Heinemeier Hansson took the stage at Rails World 2026 to talk about going "pencils down" on hand-typing code — and about how much Rust he and 37signals are now shipping — a chunk of the internet read it as a concession speech for Ruby on Rails.
Ben Halpern, co-founder of DEV and of the Forem software that runs it, argues that reaction has the story backwards, and he has production numbers to back the claim up. Over the past month, Forem has been replacing internal Ruby modules with Rust — inside the same Rails monolith.
The mechanism is an autonomous maintenance process Forem calls `rust-loop`. Running continuously alongside eight other maintenance loops, it ports CPU-bound and allocation-heavy Ruby hot paths into an in-process native extension, `gems/forem_native`, built with the `magnus` and `rb-sys` crates. The extension is split into a pure-Rust `forem_core` crate with no Ruby dependencies and thin per-function `magnus` wrappers that borrow Ruby string buffers in place, avoiding serialisation and memory copies.
The reported results: markdown pre-processing, frontmatter normalisation and reading-time tokenisation run about 54 times faster with zero Ruby heap allocations; home-feed ranking and dynamic `pgvector` SQL expression assembly are around 20 times faster with 97 percent fewer Ruby objects allocated. HTML sanitisation, inline SVG icon rendering, @-mention parsing, URL normalisation, Public Suffix subdomain routing, constant-time CSRF and JWT verification, and secret scrubbing on agent transcripts have also moved. Three pure-Ruby dependencies — gemoji, emoji_regex and wcag_color_contrast — were deleted from the Gemfile outright.
Crucially, none of this asks contributors to install a Rust toolchain: roughly 300 KB precompiled platform binaries load in under a millisecond for `bin/dev`, and the production container compiles a stripped shared library in a multi-stage builder that leaves no Rust toolchain in the final image. Each ported function is pinned by RSpec differential specs that compare 1,000-plus randomised inputs against the legacy Ruby implementation.
Halpern's framing is that Rails was never only about Ruby syntax. Its real competency, he argues, is taste and integration — pulling migrations, background jobs, Kamal deploys, native extensions and agent-safe database introspection into defaults that feel obvious in hindsight. Where a dynamic framework once hit a wall and forced teams to split out a Go or Rust microservice, the same work can now happen in-process.
The proof point he offers is a language feature rather than a language choice. Rails 8.2 is landing `bin/rails query` — a read-only-by-contract database command that emits structured JSON with `schema`, `models` and `explain` built in — and it has been backported to Rails 7.2, 8.0 and 8.1 via the `rails-query-command` gem. The problem it solves is specific to the current moment: coding agents reading static source files guess wrong about production bottlenecks, and nobody is going to hand an autonomous agent an unrestricted `rails console` against live data. Forem had already built its own read-only Postgres gateway to get the same capability; the point is that the ecosystem now gets one shared convention instead.
The caveat is that this is one company's experience, described by its own co-founder, on a codebase its team knows intimately. But the argument lines up with a broader shift: as AI agents make rewriting implementation code cheaper, the durable investment moves to the schemas, routes, boundaries and contracts that stay put — and frameworks that make those cheap to keep clean hold their value regardless of which language fills in the hot loops.




