The repository pingdotgg/ts-rust was created on October 7, 2026, described as "an experimental Rust port of the TypeScript 7 compiler (tsc)," pinned to a late-September upstream TypeScript build.
What the README claims
- Correctness. "All 181,711 ported Go tests pass," and "TanStack Query core and Hono check with diagnostics identical to Go's."
- Speed. "Compared with
tsc7,tsc-rsis 1.61× faster" on the author's real-world app benchmarks. - Cost. The earlier attempt used "over $400,000 in API priced tokens with GPT-5.6 Sol and GPT 6 Astra," wrote "over 1.3m lines of Rust over multiple months of /goal loops and never got past like 84% compat." Then: "throw Opus 5.5 at this. It had a working v0 in 10 hours," for "~$24,047 of API spend over 2 weeks."
- Human role. "I've never read a line of this code." Rules for the agents live in files such as AGENTS.md; a section headed "The Slop Line" marks where LLM-written text begins.
- Known gaps. Memory "grows slowly during long edit sessions (about 20 MiB per 1,000 edits)," some monorepo and build-mode issues, and only Linux x64 and macOS arm64 binaries.
Browne summed it up on X: "I burned ~$400k of Codex tokens and got nowhere. Burned ~$20k of Opus and got there in 2 weeks."
The debate
The Hacker News thread drew more than 200 comments. A commenter posting as DanRosenwasser, who identified as a "TypeScript team member," called it "impressive work," adding that "3 of these ports have popped up in the last week." The sharpest critique was about tests as the specification: "Give them years of human labor worth of tests and they will fix compile errors until it works (as well as the tests can ensure). Now you have a codebase no one has read." Others asked whether there are gaps where the original authors thought tests unnecessary, noted the real-world gains amount to seconds, and questioned Browne's neutrality toward Anthropic.
Why it's a trend. Three ports in a week says large mechanical translations with a big existing test suite are now within reach of coding agents. The two-model cost comparison is one developer's experience on one project, with different models, harnesses and starting points -- not a controlled benchmark.
I think the cost headline is the least interesting part, and the most likely to mislead: the README says Opus 5.5 "started from scratch" on the code, but it still had a known target, written agent rules in the repository, and the lessons of months of failed runs. The durable lesson is about where verification lives. ts-rust works because TypeScript arrived with 181,711 tests written by people over years; the agents converted that human investment into a second implementation. That's a real and repeatable pattern for ports, migrations and rewrites. But "never read a line" is fine for an experiment and a problem for anything people depend on, because the parts tests don't cover are exactly where an unread codebase hides its bugs. The responsible version of this trend keeps the agents and adds the missing step: differential testing against the original on real projects, plus human review of the riskiest paths before anyone ships it.
ts-rust, an agent-written Rust port of the TypeScript 7 compiler, passes all 181,711 ported tests and its author has never read its code; his $400k-Codex versus $24k-Opus story is one project's experience, and the real lesson is that agents can turn a large human-written test suite into a new implementation -- while the untested parts stay unreviewed.