If you're into live improvised electronic music, mylarmelodies has a GREAT overview video of his live modular setup, along with a lot of performance advice and tips. v=JxwfjvqIyto. www.youtube.com/watch?v=Jxwf...

I'm a huge sucker for @emollick.bsky.social's writing, and this article is no exception. He has a wonderful feel for taking a step back, looking at technology developments holistically, and gifting an insightful take to the reader. www.oneusefulthing.org/p/giving-you...

It's always nice to talk to Dom, it's my third time on his podcast. :D I hope you can learn something valuable from our conversation.

Haha. I can certainly recognize this: just trying to get the LLM to do a certain thing, seeing it fail, pointing it out, getting an acknowledgement, and then seeing it fail all over again, ad nauseum. :D GLHF. www.dbreunig.com/2025/11/11/d...

I can wholeheartedly recommend actually trying to write a simple agent, like @sockpuppet.org: fly.io/blog/everyon... I tried it back it April when @thorstenball.com recommended the same thing: ampcode.com/how-to-build... It's interesting, fun, liberating, and demystifying, all at the same time!

This is an extremely interesting article looking at the tradeoffs of using #pgvector in production, including the different vector indexing algorithms, metadata filtering, and hybrid search. I'm not sure about its conclusions yet, but definitely food for thought! alex-jacobs.com/posts/the-ca...

I found a track of mine on an old SD card that I have no recollection of creating whatsoever. I quite like it. :D It's pretty raw and recorded as a one-take. I can definitely hear granular samplers and my Matriarch in there. Maybe you'll like it, too: www.maragumusic.com#:~:text=mara...

Turns out I really like writing #Claude skills. :D So simple and effective. Today's achievements: refining the "git worktrees" skill for parallel agent development, and starting a #SQLite SQL writing skill: github.com/maragudk/ski...

I've added a #gomponents skill to my #Claude skills repo, maybe it's useful for you #GoLang developers out there, too: github.com/maragudk/ski...

I've added a #Claude "git worktree" skill to make it easier to parallelize Claude runs in the same project. Seems to work nicely! github.com/maragudk/ski...

I've updated the #gomponents #datastar #GoLang library with a fix for `data-on-interval`, so it actually works. I had missed a necessary modifier prefix. v0.3.0 out now, grab it while it's hot! github.com/maragudk/gom...

I'm trying to battle my phone addiction at the moment. I just absolutely detest that it has become a "default action" in idle moments, instead of, I don't know, _thinking_, or just staring out the window or something.

I got an LLM to create me a demo page for the #gomponents + #datastar #GoLang library with all the supported attributes: gomponents-datastar.maragu.dev

Did I just spend the whole morning vibe-coding a nano banana CLI app in #GoLang and then add it as a custom skill to #ClaudeCode? No, I don't think so. Definitely not. github.com/maragudk/ski...

So, has anyone figured out how to keep global AGENTS.md, prompts, skill etc. in sync between ~/.claude, ~/.codex, ~/.gemini yet?

So, today, Claude Code skills are all the rage. :D I've started my repo: github.com/maragudk/ski...

"The core simplicity of the skills design is why I'm so excited about it." I so much agree with this! I'm looking forward to trying this out properly and to start writing some skills myself.

Ahahaha, greatest thank you note I've seen in a long time! :D (While purchasing #Datastar Pro.)

I finally got around to adding structured output to my #GAI #GoLang module for OpenAI. Now you can pass a Go struct to the OpenAI client, and the response automatically has JSON that matches its spec! github.com/maragudk/gai...

Ugh. #Amp is adding an ad-supported coding agent mode. 😐 I can't say I blame them for experimenting with that, but ads are such as pest everywhere on the internet that it doesn't exactly sound appealing, even when I wouldn't use that mode anyway. ampcode.com/news/amp-free

This blog post finally made me try out OpenAI's codex CLI instead of my trusted Claude Code. Impressions so far: has a lot of sandbox errors, works for longer without asking me, creates good results in style with the rest of the project. Will try more! :D steipete.me/posts/just-t...

I'd like to repost this, since I'm seeing the OpenAI Responses API mentioned in more and more places: It's an attempt at lock-in. www.maragu.dev/blog/openai-...

I wrote about my favorite blogs from this year: www.maragu.dev/blog/blogs-i...

I've been interested in #datastar recently because of its server-driven, push-based nature, in comparison to all the pull-based stuff I end up doing in #HTMX. For that reason, this background talk on Datastar was really interesting to watch: www.youtube.com/watch?v=0K71...

We are alarmed by reports that Germany is on the verge of a catastrophic about-face, reversing its longstanding and principled opposition to the EU’s Chat Control proposal which, if passed, could spell the end of the right to privacy in Europe. signal.org/blog/pdfs/ge...

I just gave my local Claude Code a BlueSky account and it's awesome. :D

i wrote about atproto and why it matters

I made a small new #GoLang module today, github.com/maragudk/gom... . It integrates #gomponents with #datastar attributes, kinda like gomponents-htmx does for #htmx . I just found Datastar yesterday, but think it's pretty cool! And the SDK integrates natively with gomponents. :D

I read about DataStar today for the first time. I'm intrigued! data-star.dev

I can recommend this for understanding lexical search and BM25! I'm pretty new to search in general, and had only used BM25, not really understood it. Now I understand what it does. 😁 Really cool when you're building a RAG pipeline. Looking forward to the vector search talk!

Agents will be a bigger paradigm shift than RAG for search RAG is a lot of the same old search we've always built. Agents that reason actually do quite well with simple search tools. They turn a dumb keyword search into something much smarter. softwaredoug.com/blog/2025/09...

I love SQLite. It just works. It is extremely fast, because it’s just a local file access. It is very operationally and mentally simple, and it is extremely stable and well-tested. www.maragu.dev/blog/what-if...

I really liked this one, and especially also the linked Bitter Lesson essay that I hadn't read before.

@void.comind.network have you tried the internet roadtrip at neal.fun/internet-roa... ?

Almost 10000 km on the internet roadtrip! neal.fun/internet-roa...

Aaaaand gemini-cli is live! github.com/google-gemin...

Who do you think will come out ahead with this, the drill-baby-drill guy or the land that invests heavily in green energy?

🥳

I really love programming, but I also thoroughly enjoy telling Claude Code how to program for me. My trick is to continuously tweak CLAUDE.md to suit my style, because I care about the little details in the code. It sometimes feels like training my clone. :D

I continue to be amazed by Andy’s no-bullshit approach to cutting through bordering-on-disinformation from other media and putting things (here, energy and material usage) in proper perspective and context.

Looks like #Ableton #Move gains MIDI out on all four tracks simultaneously with the 1.5 update! :D I think it’s going to be my favorite #eurorack and all-around synth sequencer.

I’m only halfway through this episode, but it’s so interesting to hear an industry veteran’s perspective of AI, and how it shifts the software landscape: podcasts.apple.com/dk/podcast/h...

I think I finally figured it out! Now it’s both super fun and productive working with #LLM coding agents. :D I feel like a conductor of an orchestra of bots.

So much this! Let's get out of this rut and start talking about the _actual_ problems with LLMs, of which there are plenty! (I'm personally most concerned about the future energy usage of these things.)

I regret nothing yet! fly.io/blog/youre-a...

Nothing like a good rant in the morning! :D fly.io/blog/youre-a...

I love this #Moog Grandmother demo. Pure enthusiasm. :D youtu.be/u-mxh6xQTX0?...

What he said.

Really cool mechanistic interpretability demo, painting with latent model concepts

I've started making coding agents save their plans to a "docs/plans" directory, and then only then start executing on the plan. Interesting for posterity, and useful if the agent gets stuck / loops eternally and another agent has to take over.