An AI-built TypeScript checker raises a harder question than speed
language server
A tool that gives code editors checking and help.
compatibility
How closely a replacement behaves like the original.
geometric mean
A combined average used for speed comparisons.
What happened
On GitHub, a website for sharing code, developer pingdotgg published ts-rust. The project is also called tsc-rs. It is an experimental Rust port of the TypeScript 7 compiler, checker, and language server. TypeScript is a programming language developed by Microsoft. Rust is another programming language.
The project says it ports Microsoft’s newer TypeScript compiler, written in Go, another programming language. It keeps the same command line, language server, and API. That aims to make it easier to try as a replacement.
The unusual part is how the port was made. Large language models wrote most of the code. The project’s author says they have not read one line of it. That statement makes testing and review central to the story.
How the port was built
The README describes several attempts. Earlier runs with OpenAI models, from the company behind ChatGPT, wrote more than 1.3 million Rust lines. They used more than $400,000 in token-priced API work over several months. Those attempts stopped at about 84% compatibility.
Then another model, Opus 5.5, started from scratch. The author says it produced a working first version in ten hours. The later run cost about $24,047 in API spending over two weeks. These numbers come from the project’s own account. They show scale and speed, but they do not prove quality.
What the project reports
The port is pinned to a September 29, 2026 TypeScript 7.1.0-dev revision. The README says all 181,711 ported Go tests pass. It also says the language server and API match results from oracle test sets.
On 60 open-source projects, type checking took about half as long as the Go version. In another comparison across six applications, ts-rust was 1.61 times faster than TypeScript 7 on a geometric mean. Bun, another checker, was faster than both in that comparison.
Those are useful measurements. They are still project-reported benchmarks. They do not replace independent testing on different machines and projects.
Why this matters
A compiler port is not only a code-generation task. It must preserve behavior that other projects already depend on. A tool can pass many tests and still fail in unusual project layouts or future updates.
That is why the author’s admission matters. If the main creator has not read the code, who understands its design? Who fixes a bug when a test misses it? AI can produce a huge amount of code. Humans still need to decide what the tests mean and who owns the result.
What we can confirm
The project includes installation instructions and early-release warnings. It supports Linux x64 and macOS arm64 builds. Windows and Linux arm64 builds are not available yet.
The README lists known problems. Some monorepos can expose the same files through different paths. Some project builds can read old or missing output. Memory can grow during long editing sessions. The tool also follows one specific upstream revision.
What remains unknown
The README does not establish independent review of the whole codebase. It also does not show how the project will follow future TypeScript changes. Long-term maintenance, broader compatibility, and the true cost of fixing missed cases remain open questions.
Hacker News, a technology discussion site, gave the post 94 points and 186 comments. That measures community attention. It does not prove the project is correct.
The next evidence should come from more real projects, human code review, and sustained maintenance. The interesting question is not whether AI can write a compiler port. It is whether people can understand and responsibly keep it working.
Sources: the ts-rust README and the Hacker News post.
An AI-made TypeScript tool looks fast, but people must check it
📰 Full story: An AI-built TypeScript checker raises a harder question than speed
A new tool tries to check TypeScript programs faster with Rust.
ts-rust
A new tool that checks TypeScript code.
large language model
A computer program that can write code.
arm64
A processor type used by some computers.
💡 The gist
- AI helped build a new TypeScript checking tool.
- The tool reports strong tests and faster checks.
- Hacker News showed 94 points and 186 comments. Attention does not prove correctness.
What is ts-rust?
TypeScript is a programming language made by Microsoft. Rust is another programming language. ts-rust is an experimental tool written in Rust.
It checks TypeScript code for problems. It also helps editors understand code. The tool tries to behave like Microsoft’s newer TypeScript compiler.
A compiler handles a program. A type checker looks for wrong uses. A language server helps an editor give useful information. ts-rust tries to provide all three parts together.
The project lives on GitHub, a website for sharing code. Its README says large language models wrote much of the project. A large language model is a computer program that can write code.
The author says they have not read one line of the code. That makes this project unusual. Many programs need people to understand their design. They also need people to fix problems later.
How much work did it take?
The README describes a long and expensive process. Earlier model runs wrote more than 1.3 million Rust lines. They still stopped near 84% compatibility. Another model made a working first version in ten hours. That run cost about $24,047 over two weeks. These numbers come from the project itself. They show effort, not guaranteed quality.
What does the project report?
The README says 181,711 tests passed. It also reports checks on 60 open-source projects. Those checks took about half the time of the Go-language version.
Another test compared six applications. The README says ts-rust was 1.61 times faster than TypeScript 7. Bun, another checker, was faster in that comparison.
These results come from the project team. They are not independent proof. Tests show what people tested. They cannot show every problem that nobody tested.
Why does this matter?
A checker must work like the older tool. Small differences can confuse large projects. A fast tool is not helpful if it gives different answers.
The project is also an early release. The README lists problems with some large code collections. Memory can grow during long editing sessions. Windows and Linux arm64 builds are not ready.
People should watch future releases. They should also watch human reviews and tests on more projects. The key question is simple. Can people understand this AI-made code well enough to maintain it?
Hacker News attention helps people notice the project. It does not tell us whether the tool works correctly.
Sources: the ts-rust README and the Hacker News post.
💬 Is the Rust TypeScript port fast and trustworthy?
The project shows what AI can build, but its speed, price, and maintainability are still disputed.
- Commenters call it an experimental project separate from the official compiler. A TypeScript team member said the work was impressive and worth investigating.
- In the author's self-report, AI produced over 1.3 million Rust lines but reached only about 84% compatibility. Another model made a working version from scratch in 10 hours and went further in about one-tenth of the time. The reported $400,000 is API-price accounting, so commenters disagree about the actual money spent.
- One commenter self-reported a 1–3 second compile-time advantage over tsc7 on a real project, but there is no independent benchmark in the discussion. Recreating Go’s strings and standard library may also add overhead.
- The earlier choice of Go was understood as a choice for easier porting, not simply maximum speed. Rust is therefore not automatically a huge win.
- The main concern is trust: the author reportedly did not read the generated code. Existing tests help, while critics say tests alone cannot guarantee maintainability; others suggest comparing with the original and adding tests through agents.
- A quick check reportedly found no obvious unsafe code, but it was not an audit. For now, the project is better viewed as an experiment and a source of ideas than as a proven replacement.
initial digest at 186 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.
A computer helper made by AI
📰 Full story: An AI-built TypeScript checker raises a harder question than speed
AI helped make a new checker for computer instructions.
ts-rust
A tool that looks for mistakes in TypeScript.
Rust
Another way to write computer instructions.
Hacker News
A website where people discuss technology.
TypeScript is a way to write computer instructions.
Rust is another way to write them.
ts-rust is a new checking tool.
It looks for mistakes in TypeScript code.
AI helped write much of the tool.
The maker says they have not read the code.
So people still need to read and test it.
The project says it can check code quickly.
It also says some problems remain.
Hacker News is a place where technology fans talk.
The post got 94 points and 186 comments.
That means many people noticed it.
It does not mean the tool is surely correct.
Sources: the project page and Hacker News.
💬 A story about TypeScript made with AI
AI made a very large program. People still need to check whether it is fast and safe to use.
- This is an experimental version, not the official compiler. A TypeScript team member said it was interesting.
- The author's self-report says the AI used over $400,000 in API-priced work and wrote over 1.3 million Rust lines, but reached about 84% compatibility. Another AI made a working version in 10 hours and moved ahead in about one-tenth of the time. People disagree about the real cash cost.
- One self-reported test was 1–3 seconds faster than tsc7. But copying Go’s parts may also make the program slower. This result was not independently checked.
- People wonder whether tests are enough when humans do not read the code. No obvious unsafe code was reported in a quick check, but it was not a full inspection.
initial digest at 186 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.
💬 An LLM-built Rust TypeScript port: speed, cost, and trust
HN commenters see the project as a valuable experiment, but debate the meaning of its performance claims, its accounting, and whether unread generated code can be maintained safely.
initial digest at 186 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.