nearly every time I open a GitHub link I get this.
I AM logged in
I run my AI sessions in the cloud through claude/codex, with github connections, which I imagined would have a different system for rate limiting.
And I don't think they are making that many requests anyway, but who knows..
I just saw an impressive UI effect in a tweet announcing "Free prompt".
I clicked the link.
It asked me to register. I did.
The prompt was literally the actual code.
Random bad prediction.
One day the use of unreleased AI models to outcompete products in other areas will be considered an unfair advantage and people will try to regulate it like the latency equalization thing from exchanges.
Before I started working as a developer, I remember thinking, "This product took 5 people 3 months to finish? That's a 1-dev, 1-week project."
Then I started doing it for a living and realized how different it is. On solo projects, the bottleneck is how fast you can output what's on your mind. On your job, it's how fast people can agree on the next step. So it's easier to understand why companies have barely seen AI move the needle on productivity.
So yeah, I'd say AI delivered the promised 100x, much more than that. But it can't do much about things that aren't its responsibility.
Silk went from first commit to self-hosted in 89 days. For comparison, Rust took 5 years, Zig took 7.
I don’t mean that as a measure of quality. It says about as much about quality as lines of code count.
But it’s a wild reflection of the times we’re living in
Silk went from first commit to self-hosted in 89 days. For comparison, Rust took 5 years, Zig took 7.
I don’t mean that as a measure of quality. It says about as much about quality as lines of code count.
But it’s a wild reflection of the times we’re living in
Working at @vercel seems like a blessing and a curse for side-project addicts. It looks like you get to work on the coolest things you can think of, but no matter how impressive your work is, people will brush it off as useless AI crap. (At least on twitter)
Important announcement: I’ve decided to take on investors in Silk. You can now buy shares in my programming language with Claude Code subscriptions.
We expect this funding round to give us enough runway to make it to lunch.
Update on silky_smooth, the graphics library I am working on for developing design tools.
writing P3 currently runs in 7 whole seconds in debug but ~500ms in release build.
It's time to optimize.
I want to talk a bit more about this, because I think it's either super cool or super lame if you believe JavaScript will die (I don't).
Silk UI is a GPU-accelerated UI framework for building desktop apps, like GPUI. But it differs from GPUI in a few important ways.
You can choose to control the UI with JavaScript
You write your app in React, Vue, or whatever your thing is.
To do that, Silk UI embeds a JavaScript runtime that is also written in Silk: SilkJS (I know, super creative with names). SilkJS started as a port of QuickJS, but it has grown a lot since then. It's now both an engine and a runtime, and it passes almost all of the 4k+ Node tests. The ones it misses are cosmetic, Node-specific details with no impact at runtime.
It uses Vite, so you get all the hot reloading niceties you're used to.
You may be thinking "that's kinda like React Native", and yes...it's a similar model.
How is that different from Electron or Tauri?
Both wrap a browser or webview so you can use web technologies. But a browser has to worry about a lot of things your tiny app will never need, and it's all or nothing: even production apps ship devtools. A browser also needs layers of sandboxing to keep web pages safe. Those are irrelevant for a native app, yet every OS interaction still has to cross that boundary.
Here's the same React app (the one in the video) on all three:
Shipped size
Silk UI: 38 MB
Electron: 287 MB
Tauri: 5 MB (tiny, since it uses your OS webview if you're on Windows or Mac)
First frame (warm / cold)
Silk UI: 156 / 246 ms
Electron: 262 / 488 ms
Tauri: 875 / 1047 ms
Memory
Silk UI: 162 MB
Electron: 291 MB
Tauri: 377 MB
(Measured on a remote Linux machine without a GPU, so all three are on software rendering.)
You pay for what you use
Silk UI was built piece by piece: the JS engine, the font shaping, the rendering, and yes, the language and compiler too. That lets me make sure only the pieces you actually need end up in the final build. No devtools in production, no date and locale tables if you don't use them, you can even remove WebAssembly if you don't need it.
Bonus:
Silk cross compile between all major targets it supports
Can I try it?
Not yet. Well, it's open source, I'm sure your agents can handle it. That said the DX is still very rough. But I wanted to share it anyway