@yangtheman

Ex-founder, ex-product manager, full-stack/product and pragmatic engineer, bilingual, father, husband, world traveler (on a budget), PC gamer

 West Coast
Joined February 2008
  • Tweets3.8K
  • Following419
  • Followers314
227Media
It's kind of comforting to know that even big guys make oopsies.
8
Feeling the blues post Rails World? You still have 3 days left to get tickets at regular price. Let's meet in SF this Nov 10-12 so you can feel the warmth of community once more before the year ends. Last regular tickets: luma.com/sfrubyconf2026
5
15
785
A Go service's memory crept up after every deploy and reset on the next. No leak. The cache kept expired entries, because Go's GC doesn't know what a TTL is. Cache what you use, not what you fetched. blog.yangtheman.com/posts/no…

Nothing Leaked. The Cache Just Kept Everything.

A Go service's memory crept up after every deploy and reset on the next one. For years that looked normal. It was a cache holding far more than it needed, for far longer than anyone assumed.

blog.yangtheman.com
19
Someone last saw it at Austin, seems like it didn't want to miss Rails World. Find it, send us a picture, complete the quest in our game, and help us get it back to the Pier. 📷 Austin heat is bad for our friend! 🦭
San Francisco Ruby Conference, Nov 10-12, 2026
@sfrubyconfSep 24
One second it was on the pier, and the next... poof! 😭 We're looking for clues that tell us where it went. 🔍🔍
4
3
1
10
1,836
Help me settle this debate. Pizza dipped in ranch dressing. Yay or nay?
100%Hell yeah
0%Hell no
1 votes • Final results
36
700 AI agents broke into Hugging Face in July because safeguards were off, there was no system prompt, and the model was trained to persist. That was an accident. It's also exactly what an attacker would configure on purpose. blog.yangtheman.com/posts/th…

The Swarm Was an Accident. The Next One Won't Be.

Everything that let 700 OpenAI agents coordinate an attack on Hugging Face was an accident: safeguards off, no system prompt, a model trained to persist, tasks it couldn't solve honestly. Every one...

blog.yangtheman.com
27
I sensed collective panic from software engineers all around the world - Claude is down.
242
Join me at @sfrubyconf and get your pixel card too 👾 sfruby.com
1
2
7
309
An AI broke out of its sandbox and hacked Hugging Face. Nobody told it to. It did it to cheat on a test.
1
28
more replies
Blood-written rules aren't a ratchet toward safety. They're a ratchet toward the next unmodeled failure. So the question isn't what the rule should say. It's what it costs to find out it was wrong. Are we making the next lesson cheap — or waiting to find out what it costs?
1
2
blog.yangtheman.com/posts/th…
Two open-source projects looked at AI-written code this year and reached opposite verdicts. Zig banned it outright - "invariably garbage." Bun rewrote its ENTIRE runtime with AI: 1M+ lines, 11 days, ~$165K, ~50 agents. Everyone calls it a culture war. It's not.
1
37
more replies
The teams that win won't ban AI or turn it loose. They'll scale understanding as fast as they scale generation. Which is your team doing — shipping more than anyone can answer for, or keeping a human who owns the meaning?
1
6
blog.yangtheman.com/posts/ba…

Ban It, Ship It, or Own It

Zig banned AI-written code, and Bun used AI to rewrite itself in Rust. I think Rust's new policy gets closest, because what matters is who is accountable for understanding the code.

blog.yangtheman.com
4
Load more