tsc-rs: AI ported the TypeScript compiler to Rust, unread
tsc-rs is an AI-written Rust port of the TypeScript 7 compiler that passes 181,711 tests and runs 1.61x faster. Its author never read the code. Here's the lesson.
tsc-rs is an experimental Rust port of Microsoft's TypeScript 7 compiler, and an LLM wrote all of it. The README says it passes all 181,711 ported Go tests and type-checks about 1.61x faster than tsc 7 on real apps. It also says this, in the author's own words: "I've never read a line of this code." That one line is the most useful thing in the repo for anyone who ships AI-written software.
What actually happened
The project lives under the pingdotgg GitHub org, the team behind T3 Code. Per the README:
- What it is. A port of the Go-based TypeScript 7 compiler, type checker, and language server. Same
tscCLI and API. MIT licensed. On npm astsc-rs. - Tests. All 181,711 ported Go tests pass. TanStack Query core and Hono give diagnostics identical to Go. Across 120 open-source repos, output differs only in listed known issues.
- Speed. About half of Go's type-check time across 60 projects. On six real-world apps, 1.61x faster than tsc 7 and 11.4x faster than tsc 6. Bun's checker is still faster.
- Cost. OpenAI models came first and topped out near 84% compatibility after more than $400,000 in tokens. Claude Opus 5.5 started from scratch and finished in two weeks for about $24,047 in tokens.
- Gaps. Linux x64 and macOS arm64 only. Monorepo and
tsc -bedge cases. Editor features like quick fixes, hover, and completions are not ported.
Why it matters for your business
The tests did the work, not the model. This port worked because Microsoft had already written 181,711 tests and a reference implementation to diff against. The agent had an oracle. Most business codebases don't. Point the same agent at an untested legacy app and you get code that looks right with nothing to prove it.
"Unread" is fine for an experiment, not for your checkout. The author is honest that it's an early release. Your payment flow, your tax logic, and your customer data need a person who has read the code path and can explain it.
Model choice changed the bill by about 17x. Same task, same goal: over $400K versus about $24K. If you run coding agents, benchmark two models on a real slice of your work before you commit a budget.
Migrations are the sweet spot. Port, translate, upgrade: jobs with a working original and a test suite. That's where agents earn their keep today.
Key takeaways
- tsc-rs is an LLM-written Rust port of the TypeScript 7 compiler, MIT licensed, on npm as tsc-rs
- Passes all 181,711 ported Go tests; about 1.61x faster than tsc 7 on six real-world apps
- Claude Opus 5.5 run cost about $24K; earlier OpenAI attempts cost over $400K and stalled near 84%
- The author says they never read the code. The test suite is what made it trustworthy
- Write the tests first, then let agents do the migration
Want AI to modernize your code without the guesswork? We write the test harness first, then run coding agents against it, so every change is checked before it ships. See how we run AI-assisted migrations, or look at what we've shipped.
Sources: pingdotgg/ts-rust on GitHub.
- #typescript
- #rust
- #ai-coding-agents
- #claude-opus
- #testing
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
OutSystems Agent Experience GA: coding agents with guardrails
OutSystems Agent Experience is GA, opening its low-code platform to Claude Code, Cursor, Codex and Kiro. Speed was never the problem. Rework was.
Read itGitHub's AI secret detection model bills credits on every push
GitHub's new AI secret detection model finds passwords with no token format. AI push protection now uses AI Credits. Set a hard budget cap before you opt in.
Read it