Sometimes I feel inadequate seeing what these models can do, but then I think they wouldn't be doing any of it if I didn't have the expertise and drive in the first place. So I might be still relevant yet (for now).
A key to productivity with agents is that you have to raise your ambition - give it bigger chunks, take bigger swings.
This is the reason why _experts_ do better with agents. They have a laundry list of "I would do X if only I had time", beginners have no such sense
1
9
Still in the top 5 (of human) committers, but getting hard to keep up. Still have 1,500 compatibility findings I'm working through, so watch out, ha.
.@yukihiro_matz appears to have achieved hard takeoff. He's merged 200 PRs in the last 24 hours to spinel and 1000 this week
2
32
Ruby is slow because we optimized for developer happiness. Now that AI is the developer, what do we optimize for? Is
speed all that matters?
Been running a beta version of Campfire rewritten in Rust today! The code itself is still ugly as sin, and 6x as verbose, but the conversion was free*, and I never had to look at the Rust directly, so this is awesome 🤩 github.com/basecamp/once-cam…
13
Spinel, the Ruby AOT compiler, is written in C by LLMs. It generates C, running Ruby code very fast and safe.
Ruby is very expressive and token-efficient, a good target for humans ideating with LLMs. Could this be the architecture we're generally moving toward? 🤔
Replying to @mitsuhiko
C is the way IMHO. LLMs write incredibly better C than any other language. You pay Rust abstraction costs even when you don't touch it yourself, AI allows to change design and evolve continuously so the smallest friction to change design ideas, the better.
2
26
Interesting that others felt the same. I was constantly asking Claude to "just give me the headline." Detail is nice, but I'm only human!
You can now set Claude Code's output style to Concise.
Claude leads with the result, keeps responses short, and still gives full detail when you ask.
Turn it on in /config → Output style, or set "outputStyle": "Concise" in settings.json.
20
Still going! 2,087 discrepancies between Spinel and CRuby found so far over nearly 39 hours. It's got to be slowing down now soon, right? ...right?
I started a Spinel compatibility sweep (vs documented CRuby subset) with GPT-5.6 Luna in Codex. It's been running for over 15 hours (wow!) and found more than 800 discrepancies, and still going! But I promise I won't spam @yukihiro_matz, ha. Fable 5 will help with a PR strategy.
I started a Spinel compatibility sweep (vs documented CRuby subset) with GPT-5.6 Luna in Codex. It's been running for over 15 hours (wow!) and found more than 800 discrepancies, and still going! But I promise I won't spam @yukihiro_matz, ha. Fable 5 will help with a PR strategy.
Don't sleep on small models! GPT-5.6 Luna is really impressive for the size, and cheap! Perfect for those high-volume, long-running workloads like this. Going continuously for 15+ hours and only used about ~7% of my ChatGPT Pro 5x plan weekly usage. openai.com/index/advancing-t…
17
💥 Ruby 2D compiled to a native binary by @yukihiro_matz's Spinel. No interpreter, no VM. Looks smooth! And this demo app is just plain Ruby, code similar to: ruby2d.com/examples#bouncing…
There's not full parity with CRuby yet, but quick progress is being made! Matz, and his agents, are closing issues about as fast as I can file them. Impressive! github.com/matz/spinel/issue…
Very excited about this project and all the possibilities ahead. 🚀
These probably won't last long (make sure to get one *with* battery). a.co/d/0gvjsSMQ
7
Oh, this is too cool. Got to think of some demos to port from ruby2d.com/examples
teapawt.net/johannestscharn/status…
This replaces scrolling for me now 😅
Had to take one of those little devices for a spin after seeing them everywhere recently.
Opus 5 wrote the renderer and sim in C to get the most out of the ESP32 S3. Will do something that leverages mic + speaker next…
21


















