Restart the USA in a new context window

Welcome to all the new folks on Bluesky! It's nice here. You can create custom feeds that can be topical, algorithmic, etc. This is the one you should use: bsky.app/profile/did:... Please share this for reach

If you can substitute "hungry ghost trapped in a jar" for "AI" in a sentence it's probably a valid use case for LLMs. Take "I have a bunch of hungry ghosts in jars, they mainly write SQL queries for me". Sure. Reasonable use case. "My girlfriend is a hungry ghost I trapped in a jar"? No. Deranged.

Today's toy: atlogin, an OIDC provider for the AT network, compatible with Tailscale custom OIDC. Try this: login.tailscale.com and login as user "<your-bsky-handle-without-the-@>@at.apenwarr.ca". You'll join my test tailnet (ACL isolated from other users). Source: github.com/apenwarr/atlogin

Announcing @modalfoundation.bsky.social, a nonprofit making open social tech viable at scale. Modal will connect experts, strategists, builders and communities to create a human-centered internet. The foundation will oversee @freeourfeeds.com, @eurosky.social and more to come. modalfoundation.org

Lexicon Garden just got AI-friendly! llms.txt + MCP features make it easy for AI agents to explore and invoke ATProtocol methods. https://ngerakines.leaflet.pub/3mca3ybkvbc2t #ATProtocol

I have to assume this was inspired by Aesop Rock. www.youtube.com/watch?v=Xf0p...

"If it could happen to her, it could happen to anybody." Protesters gathered in downtown Wichita today to speak out after federal officers shot and killed a woman in Minneapolis and two people in Portland. Story | https://www.kwch.com/2026/01/11/dozens-gather-protest-ice-trump-downtown-wichita/ 

This x1000. LLMs are incredibly exciting but the idea of best-in-class software dev capabilities being centralized and monopolized is terrifying.

Still adjusting to being able to say "I made this" when really I asked an LLM to make it for me. I know much of what goes into it is the decision making around the what and how of the thing being built (features/non-features, tech stack, architecture), but still! All things in time I suppose.

This is your regular reminder that authoritarianism is just domestic abuse writ large.

oh this is cool

Building a very simple editor for @standard.site data. Then planning to update seth.computer to render posts from there instead of local markdown files.

Testing out @margin.at and very much enjoying annotating the web again as a former hypothesis user. Responsive dev fixing things quickly. Excited to continue to see this shape up!

Statement by Federal Reserve Chair Jerome H. Powell. This is fucking crazy!!!!

"What took generations to build is being dismantled in months, and with it goes not just expertise but what remains of the shared American faith in expertise itself."

Furthermore: always prefer simple, portable, and open solutions.

Added an "LLM shaming" section to RFD 576 rfd.shared.oxide.computer/rfd/0576#_ll...

Embracing my finitude has never been my strong suit. Now AI has undone any progress I’ve made on that front.

> To allow ourselves to even contemplate a prohibition of the development of certain kinds of computer programs is to delude ourselves into thinking that the problem is a technical problem rather than a human one. 👏👏👏

> To allow ourselves to even contemplate a prohibition of the development of certain kinds of computer programs is to delude ourselves into thinking that the problem is a technical problem rather than a human one. 👏👏👏

Sprites.dev by @fly.io is a very cool new thing: it solves two of my pet problems at once, developer sandbox environments for coding agents and a JSON API for executing untrusted code I wrote more here: simonwillison.net/2026/Jan/9/s...

i think a core beef that i have with a lot of this "automation is bad" sentiment lately is that there is no inherent moral good in laboring the story of humanity is one of invention, where we improve our conditions by building things that help us do more things more easily

I had a pile of docs to write at work, explored how to integrate LLMs to the process. - Iteratively refined an outline, adding detail (no wordsmithing!) - Wrote voice/delivery guidelines (concise, link to existing docs) - Iterated on claude output page by page The result was solid, and fast.

As I keep saying, the speed with which AT Protocol tooling allows people to move is impressive.

is he... y'know... a friend of claude

It's great that I can use an LLM to generate more code and fully fledged apps so much more quickly than I could before. However, I still prefer small, does-one-thing-well, simple programs and services that I don't really _need_ an LLM to write for me. Simple will always be better than complex.

Here's a fun @standard.site prototype: Atom feeds for any publication! e.g., @octet-stream.net's: geomys-atsite.exe.xyz/profile/octe... Two very cool things: 1. this was ~550 lines of code github.com/FiloSottile/... 2. the feed URL stays the same even if the handle changes or moves PDS!

The official Bluesky app is designed for iOS and is comically small on macOS and is not resizable. So, I made a macOS app that wraps the web app. modemlooper.net/skyline/

The ecosystem aligning on shared lexicons is important to fully realizing this vision too. Happy to see that fairly well happening so far. @standard.site, for example. Also @popfeed.social using the same lexicons as @bookhive.buzz Love to see it

I vibe with the general sentiment, but I’d rather say it’s a post-everything-app protocol. The new paradigm, the Atmosphere, is a bunch of domain-specific apps built on an everything-protocol.

Since diving further into ATProto and its ecosystem, then interoperability piece increasingly becomes the most exciting aspect. I suppose it’s kind of the whole thing but still, it opens up so many opportunities! Lots of ideas I have to build on top of existing lexicons.

That one post from the Google engineer who was like “Claude code did our whole year long project in an afternoon”… yes that is impressive. But I bet a substantial part of that year was spent determining precisely how to describe the problem that they wanted to solve.

Late present showed up today, honestly might be my favorite album of all time. Very excited.

this is your mission. roll up to this daycare in full tactical gear to harass some kids and teachers. when you get spooked by your own shadow like a panicked horse and start shooting your gun, our pedophile president will call the person you shot a terrorist. good luck soldier

Looks like @selfhosted.social is returning 502s. I pointed a friend of mine at this and they migrated to it and now can't get to their account. Anyone else seeing this? Insight into what's going on?

Be aware that there are accounts attempting to steal session keys. This particular method won’t work using the Bluesky app for a couple of reasons, but third party apps may be vulnerable to similar attempts. We will attempt to automatically label these as Platform Manipulation.

Shout out to one of my favorite tools: Kill the Newsletter. Turns email newsletters into RSS feeds. Dead simple, does one thing and does it well. kill-the-newsletter.com

Drop your RSS feeds in the replies so people (including me) can find them 👇

New year, new Oxide and Friends predictions episode! @bcantrill.bsky.social and I were joined by @simonwdc.bsky.social and @steveklabnik.com to predict 1, 3, and 6 years into the future. Winners from previous years? Apple in and out of the VR/AR business? Looking prescient! Agents not a thing? Sure!

New on my blog: "Why Study CS? Thoughts on LLM-assisted software engineering" kmicinski.com/claude-code-...

Identity is a powerful thing. This first image isn't wrong though. Things like system design, architecture, security, etc, those skills have always been important. My advice: take the time freed up from having to memorize language and library syntax and focus it into sharpening higher value skills.

i love the beginning of the year because everyone starts blogging. and if you (yes you) were thinking about starting, this is your sign

has anyone tried --safely-skip-permissions

cowsay --dangerously-skip-permissions

In the future, you won't even need to cook food. You'll just ask the AI and it'll describe the food to you

Universally applicable.