The Christmas Compiler Miracle

Today’s issue: Solving massive supply chain attacks with Magic: The Gathering, gearing up for Waku 1.0, and pitching my local masonic lodge on the blockchain brotherhood.

Welcome to #451.

---

The Christmas Compiler Miracle

My children awoke early on Christmas morning and raced downstairs, eager to open the many presents they were sure had been left under the tree for them by Santa Claus and their loving parents.

But there were no presents. There were no packages. There was nothing under the tree at all, except a single laptop opened to an unassuming README.md file.

“Papa, where are all the toys?” the children asked with alarm.

“Fear not children,” I calmly replied, “for unto us this day is given jax-js, a machine learning library for the web that runs on Wasm and WebGPU.”

It’s basically the JAX Python library in pure JavaScript. You get NumPy-style arrays, plus the signature “composable transformations” like grad(), vmap(), and jit(). The twist is that it runs entirely in the browser, so you can do heavy numerical computing without needing a Python backend or CUDA installs.

But isn’t JavaScript JIT famously bad at tight numerical loops like this? Yes, which is why jax-js mostly avoids it. Instead, it compiles your array program into kernels that run at near-native speed.

Here’s what it looks like under the hood:

• You write JAX-style array code in JavaScript, which jax-js traces into a small math program instead of immediately executing it.

• That program gets lowered into optimized kernels, then compiled to WebAssembly (CPU) or WebGPU (GPU).

• When you use jit(), jax-js fuses a long chain of operations into a single kernel, so you don’t waste memory on a bunch of intermediate arrays.

This means you can use JavaScript to train small models, run inference, compute gradients, and build interactive ML demos that are fully client-side — and they can all run anywhere a browser runs.

Bottom Line: So sure, my kids were a little upset at first that they didn’t get a Switch 2 or a puppy this year. But they did learn that JavaScript is the true reason for the season, and that doing weird Python math in the browser might just be the greatest gift of all 🎄🥰.

---

Now available for QA teams: AI platform for deep coverage from QA Wolf

QA Wolf’s new AI assistant helps any team build and maintain automated test coverage for 80%+ of their product workflows—from simple logins to complex transactional flows—just by chatting with the AI.

• Test complex workflows reliably: Prompts generate Playwright and Appium code, instead of flaky plain-English steps.
• Run regressions in minutes: Full suites execute 100% in parallel.
• Own your tests: Open-source code, no vendor lock-in.

Join the waitlist here.

---

1. The WebKit team just released CSS Grid Lanes, which it calls “the future of masonry layouts on the web.” I also recently spoke about “the future of masonry” at my local lodge meeting, but they cut me off halfway through my pitch on how Masonic NFTs could use the blockchain to verify who is a true brother.

2. Go beyond IDEs with Tonkotsu’s desktop app. It’s disciplined, multi-agent engineering built for professionals. Build a thorough technical plan, then delegate coding to a team of agents. Pretty slick. [sponsored]

3. Nick Miller wrote an in-depth breakdown on static allocation with Zig, based on his work on his key/value store. And I must say, it’s nice to see him doing so well after getting his heart broken by Jess after she moved out of the loft.

4. Waku released a v1.0 alpha of its minimal React framework with full RSC support.

5. The Concise TypeScript book is a free and open-source guide to TypeScript that was recently updated and proves that the word “concise” can mean different things to different people.

6. Kieran Gill wrote about how nobody really knows how to scale LLMs to larger codebases – but he has a few theories on what we should try next.

7. CodeRabbit’s OSS Program gives open-source projects free access to all their Pro-tier code review features. Used by projects like Bun, Vue, Nix, and 100k+ others to ship better code. [sponsored]

8. Una Kravets wrote about directional CSS with the scrolled state query.

9. Peter Steinberger wrote about shipping at inference-speed, and how far vibe-coding has come this year.

10. A group of 16-year-olds pwned Twitter, Vercel, Cursor, and Discord with a supply-chain attack by exposing some critical vulnerabilities on Mintlify, an AI documentation platform. But they promised not to expose any customer data as long as an adult bought them some watermelon vapes and Magic: The Gathering cards. Merry Christmas, my little gremlins.