Fetching from the wire…
01
02
03
04
05
06
07
08
Top 5 · 2026-05-10 · source-backed
Jarred Sumner dropped a bomb on Hacker News this week: Bun's experimental Rust rewrite passes 99.8% of pre-existing tests on Linux x64 glibc. 960,000 lines of Zig, rewritten to Rust, in approximately six days. With AI assistance.
Let that timeline sit for a second. Six days. Nearly a million lines. 99.8% test compatibility.
The 658-point, 631-comment HN thread is exactly what you'd expect. Half the comments are "this changes everything" and the other half are "this changes nothing because the code will get thrown out." Both camps are right in ways they don't realize. Sumner himself said there's a "very high chance all this code gets thrown out completely." Bun hasn't committed to shipping the Rust version. The rewrite preserves the same architecture but gains Rust's ownership model, lifetime enforcement, and destructors. It's an experiment, not a product announcement.
But here's what I think matters more than whether Bun ships Rust: this is the most concrete data point we've ever had for AI-assisted large-scale rewrites. Not a toy project. Not a weekend hack. A production runtime with a real test suite that catches real regressions. And the test suite is doing the work that matters. 99.8% compatibility means the AI didn't just translate syntax. It translated behavior.
I've been skeptical of the "just rewrite it with AI" crowd because most rewrite stories are anecdotes without verification. This one has a test suite with thousands of cases, and it passes almost all of them. That's a different kind of evidence.
The actionable takeaway for builders: if you have a legacy codebase with good test coverage, AI-assisted rewriting is no longer theoretical. The test suite is your safety net. Without it, you're just generating plausible-looking code. With it, you're generating verified code. The bottleneck isn't the rewrite. It's whether you wrote enough tests before you started.
One thing I keep thinking about: Sumner is an exceptional engineer who understands both Zig and Rust deeply. He knows what correct output looks like. AI assistance in the hands of someone who can evaluate the output is a different tool than AI assistance in the hands of someone who can't. Human taste as the bottleneck. Again.
Each link below shares sources, entities, or timing with this story.
Jarred Sumner uses Claude / Shared entities / Shared topic / What happened next
Linked by a graph relationship (Jarred Sumner uses Claude); both cover Jarred Sumner, Linux, Rust, Sumner; overlapping topics (code, compatibility, days, different, rust).
Linked by a graph relationship (Jarred Sumner uses Claude); both cover Jarred Sumner, Linux, Rust; overlapping topics (rewrite, rust, suite, test).
Codex CLI uses Rust / Shared entities / Same source domain / What happened next
Linked by a graph relationship (Codex CLI uses Rust); both cover Hacker News, Rust; reported by the same outlet (news.ycombinator.com).
Codex CLI uses Rust / Shared entities / Shared topic / Earlier coverage
Linked by a graph relationship (Codex CLI uses Rust); both cover Hacker News, Linux; overlapping topics (code, comment).
Jarred Sumner uses Claude / Shared entity: Hacker News / Shared topic / What happened next
Linked by a graph relationship (Jarred Sumner uses Claude); both cover Hacker News; overlapping topics (code, different).
Jarred Sumner uses Claude / Shared entities / Same source domain / What happened next
Linked by a graph relationship (Jarred Sumner uses Claude); both cover Hacker News, Half; reported by the same outlet (news.ycombinator.com).
Jarred Sumner uses Claude / Shared entity: Hacker News / Same source domain / Shared topic / What happened next
Linked by a graph relationship (Jarred Sumner uses Claude); both cover Hacker News; reported by the same outlet (news.ycombinator.com).
Jarred Sumner uses Claude / Shared entities / Earlier coverage / Tension
Linked by a graph relationship (Jarred Sumner uses Claude); both cover Hacker News, Rust; earlier Hacker News coverage from 2026-04-16.