Am I micromanaging my AI agents? I checked my own sessions: only 4 of my 19 messages corrected something. The real problem was an agent that followed my instructions too well. luislavena.info/writing/roug...

I thought reference counting (ARC) would be slow compared to a garbage collector. Then I benchmarked my Ruby-like language. Most of the time went to malloc and free (61 to 85%), not to ARC (0 to 23%). Glad I measured before I optimized the wrong thing. August notes: luislavena.info/writing/roug...

A Ruby-like programming language with no garbage collector. I designed it, an agent wrote the compiler. Most of the bugs were memory bugs. No surprise there. July notes, with the hours and the token bill: luislavena.info/writing/roug...

I built an agent skill whose main job is to talk me out of building things. Yup, no more "great idea, here's the architecture." It demands evidence first, and only writes requirements if the idea survives. New video on YouTube: youtu.be/ZnNok4pVX4Y

Why I let AI reject my own ideas

YouTube video by Luis Lavena

youtu.be

🚨 Fresh Flatcar Alpha, Beta, and Stable releases are now out with Linux 6.12.87 and a large set of kernel security fixes. The recently disclosed Linux “Dirty Frag” privilege escalation flaws are addressed in these releases.

Flatcar is unaffected by Copy Fail (CVE-2026-31431)! The kernel option causing it was never enabled, because Flatcar is minimal by design and only ships what containers need Sometimes the best mitigation is to reduce the attack surface by not shipping vulnerable components 🔒

AI agents running on your machine inherit every credential you have. AWS, Kubernetes, SSH keys, all of it. I run mine inside 'Dexter', a locked-down Debian VM. No sudo, no synced credentials, no push access. Just the tools the agent needs. Short walkthrough: www.youtube.com/watch?v=F_mv...

Locking down AI agents

YouTube video by Luis Lavena

youtube.com

Hey if you still have an Angelfire or Tripod site, or if you like one, now is the time to back it up, like, urgently

Having a linguistics background I think there's a huge misconception when it comes to prompting LLM's, the vast majority thinks that coding is "hard" compared with speaking a natural language. Well it's actually the contrary from a utilitarian POV less verbs at your disposal => easier to be specific

Kelsey Hightower@kelseyhightower.com · 9mo ago

Generative AI is a slop generation machine by default. You have to put in a lot of work to get something of quality from it.

rv has a big release, with windows support and gem CLI tools! the fastest way to generate a rails app is now `brew install rv; rvx rails new`. check it out: https://spinel.coop/blog/rv-0.5-cli-tools-windows/

NOPE.

Ruby @rubylang.bsky.social (RubyInstaller) made to Microsoft Store! That is a lot of work! Congratulations to the team! 🎉🥳👏 rubyinstaller.org/2026/01/27/r...

Me: Here are some errors, fix them. Claude: This test case involves multiline plain scalars, which is a complex issue. Let me look at other failing tests to find easier wins first. 🤖🐓

hey folks, i’m a rails / ruby dev with a bit over 10 years of experience professionally and i was just hit with my second layoff. looking for a new role would love to chat! #rails #ruby.

2025 is ending. A lot of things happened, some bad, some good, but there is only one thing worth more than words can describe. I'm thankful for what 2025 brought to our lives and eager to look at 2026 with a smile. ❤️

Alongside the Ruby 4.0 release last week, I published charm-ruby over Christmas, bringing @charm.land's excellent terminal UI libraries to Ruby. A full TUI framework, reusable components, styled output, forms, markdown rendering, spinners, charts, and a lot more! Demos below ⬇️

Wrapping up before xmas dinner, pushing what Claude Code Opus worked out: github.com/luislavena/j... Still needs work to be 100% compliant, but looks promising!

GitHub - luislavena/just_html.cr: Port of JustHTML to pure Crystal (done by 🤖 Claude Code Opus)

Port of JustHTML to pure Crystal (done by 🤖 Claude Code Opus) - luislavena/just_html.cr

github.com

Luis Lavena@luislavena.info · 10mo ago

Following @simonwillison.net test of porting JustHTML to another language, decided to task Opus 4.5 and port it to #crystallang, been running for 2 hours, ran out of context space 1 time, but looks promising.

Following @simonwillison.net test of porting JustHTML to another language, decided to task Opus 4.5 and port it to #crystallang, been running for 2 hours, ran out of context space 1 time, but looks promising.

GitHub: up to a 39% reduction in the net price of GitHub-hosted runners!🥳 Also GitHub: a new $0.002 per-minute GitHub Actions cloud platform charge that will apply to self-hosted runner usage 🤡 Guess its time to complete the move to #codeberg

Decided to fire up Claude Code with Opus 4.5 and try to re-implement a toy language in Crystal targeting WASM 3.0, gave it a few sources of inspiration and comes with really interesting feedback.

Announcing the Gem Fellowship, a grant program for improvements to Ruby-related open source projects. gem.coop/fellowship/ Want to improve your favorite gem? Submit a proposal, starting next month.

Been there, done that, OSS burnout: opensourcepledge.com/blog/burnout... I still recall getting emails yelling at me about their "lost profit" due a bug on my *check notes* free code. Or getting mocked by others supposed _fellow_ open-source devs, who knows why.

Burnout in Open Source: A Structural Problem We Can Fix Together | Open Source Pledge

Burnout is affecting the entire Open Source ecosystem. Here's what we could do to make things better.

opensourcepledge.com

If we are having a conversation and during it you say to me "let me ask [insert your AI chat agent name here]" I will immediately consider the conversation over. You cannot outsource your thinking to a blob of floating point numbers (it is ok not to know the answer).

"Thank you for sharing those resources. You're absolutely right - I was overcomplicating the solution." 🙃

🤖🎈💥

Brian Merchant@bcmerchant.bsky.social · last yr.

There's been endless talk about an AI bubble, but less about exactly how, why, and how much it's a bubble. So I turned to the framework put forward by scholars Brent Goldfarb and David A. Kirsch, authors of "Bubbles and Crashes," for assessing tech bubbles. Spoiler: On a scale of 1 to 8, AI is an 8

AWS us-east-1 is turning issues, so I guess half the internet websites will be down or having issues? 🫣 health.aws.amazon.com/health/status Well done folks! Yay to the distributed nature of the internet! 😅

When AI wants to test everything under the sun, you can just make it go back to the basic by asking "check how we are doing it right now"... KISS, baby. 😉

Dear fellow open-source devs: its OK to use an AI to help you out (heck, I use it too!) and I would understand you don't have the time to review every reported issue, but I would rather prefer radio silence than an automated "AI analysis" to the issue I just reported. 😔

One day the industry will recognize the drawbacks of AI agents and nondeterministic automation, and rediscover the UNIX philosophy of chaining together small purpose built tools in a low cost and predictable way, otherwise known as shell scripts.