has anyone with conference room naming authority been bold enough to call their building’s largest audience hall Security Theatre
Speaking obliquely, I’m currently working on a project to demonstrate that I can write infrastructure code. I hadn’t seen devicemapper’s thin provisioning support before - this seems pretty cool!
domain owners should get a TLD email address; i own micahrl.com and so i should be able to receive email at micahrl@com.
Testing some domains, all of these render as punycode in the browser, I wonder what micro.blog does with them… now in Markdown format
- <⍼.testpuny.micahrl.com>
- <ꙮ.testpuny.micahrl.com>
- <🌙.testpuny.micahrl.com>
- <θ.testpuny.micahrl.com>
- <ə.testpuny.micahrl.com>
- <ɔoɯ.ɯᴉɔɐɥɹl.testpuny.micahrl.com>
Testing some domains, all of these render as punycode in the browser, I wonder what micro.blog does with them
⍼.testpuny.micahrl.com ꙮ.testpuny.micahrl.com 🌙.testpuny.micahrl.com θ.testpuny.micahrl.com ə.testpuny.micahrl.com ɔoɯ.ɯᴉɔɐɥɹl.testpuny.micahrl.com
see if you buy a fun domain you have to at least do something with it, like this is the bare minimum imo iso8601.date
DNS would be so much more fun if Unicode didn’t show up as punycode so often
it is outrageous to me that the domain chex.quest is registered and just parked. theres only one thing that domain could possibly be for
Idea: HTML for shortcodes
Friday, October 10, 2025
I have a lot of experience with Hugo and I like the concept of static site generators in general. I am not really a fan of the shortcode syntax, though. I’d love to see a version of shortcodes that uses HTML instead.
E.g.:
# Example post
Here's my post. It was built at <shortcode-buildtime format="iso-8601"></shortcode-buildtime>.
<shortcode-quotefig
cite="https://ask.metafilter.com/55153/Whats-the-middle-ground-between-FU-and-Welcome#830421"
caption="<a href='https://ask.metafilter.com/55153/Whats-the-middle-ground-between-FU-and-Welcome#tangerine'>tangerine on MetaFilter</a>"
>
This is a classic case of Ask Culture meets Guess Culture....
</shortcode-quotefig>
The shortcode engine I’m thinking of might convert that to this HTML
<h1>Example post</h1>
<p>Here's my post. It was built at <time datetime="2025-10-10T00:05:04-12:00">2025-10-10T00:05:04-12:00</time>.</p>
<figure class="quotefig">
<blockquote cite="https://ask.metafilter.com/55153/Whats-the-middle-ground-between-FU-and-Welcome#830421">
<p>This is a classic case of Ask Culture meets Guess Culture....</p>
</blockquote>
<figcaption>
<a href='https://ask.metafilter.com/55153/Whats-the-middle-ground-between-FU-and-Welcome#tangerine'>tangerine on MetaFilter</a>
</figcaption>
</figure>
HTML already has all the things shortcodes have to have: a name, optional attributes, and an optional body. They also automatically work great in any editor, and mix well in Markdown. And if you’re using an SSG, you’ll already know at least a little HTML.
It would be kind of like web components implemented as static preprocessor macros. Or, to be spicy, you could call it “Unreactive Server Components”.
Startup idea: tackle builds across the stack
Tuesday, September 30, 2025
We need a single tool that does all of:
- local builds (replace Make/Gradle/etc)
- CI builds (replace GitHub Actions)
- LLM sandbox builds (replace Claude Code terminal frontend)
- Cloud LLM builds (replace GitHub Workspaces)
These are weird bedfellows, but build systems touch all of these.
- Local and CI builds both have to encode artifact dependencies, and doing this for both local and CI builds usually requires a lot of fragile code generation.
- (Plus, iterating on CI build changes is awfully painful, considering the execution time; reducing duplication means that most CI build system changes can be tested locally in a much tighter loop).
- It was pretty wild to launch Claude Code outside of a container, but I probably would have done the same to be first to market. What’s crazy to me is that GitHub has just followed suite. No! Put the LLM in a container and stop asking me if you can execute
makeorfind! (oh and by the way, you can depend on a suite of tools instead of first tryingrgand recovering from a missing command withfind -exec grep, etc) But then, wait, now my LLM container has to have all my dependencies in it. And there you go: the LLM tool has to be integrated with the build system. - In exactly the same way, the cloud LLM tool needs access to all those same dependencies, and it has to know how to build the code anyway, and it might even be hosted by the same vendor that hosts your CI (i.e. GitHub).
My spicy take is that the build system is the hard part, all the LLM stuff is just dressing. Here’s my proof: no one has made a good build system yet, but all the other pieces that exist (Actions, Workspaces, Claude Code, containers) are fine. I’m not sure why this is: every time I set up builds I can’t believe I have to do it twice. The only project I’m aware of that even tries to do both local and CI builds well is Bazel, but they’re really up front that it’s slow for small local stuff, and this tool would need to embody the principle of making easy things easy and hard things possible. That first clause means it has to be fast for local builds, and also easy to install (no separate JVM install step). Bazel has whiz bang features like pulling build artifacts over the network for local builds, which is nice if you’re a company that is already big enough to fund your own build system(s), but not a feature I would start with.
Microsoft has all the business pieces to make this pencil: CI builds, LLM CLI tool, and LLM cloud tool. But could it get adoption by tying it all to their proprietary services like this, or would it have to be an open ecosystem?
Would also be an interesting product space for a startup, but how to make money on it? No idea, I’m just the technical cofounder, see, that’s what I rely on all you idea guys for. Hop to it, please, the downtown office I’m eyeing for our fully remote company won’t pay rent on itself!
Working on a polyglot (Java/Go/JavaScript) at work, it really seems like build and CI systems are two halves of the same thing, which should be built together. I built my dependency graph once for make, and now I have to do it again for Gitlab CI… and keep them in sync forever.
Decided that being able to easily insert work before my working copy is worth using a new tool. I keep reaching for a commit command and feeling disoriented. Going to take some getting used to.
I suspect there may be architectural software decisions that have cost human kind more time and money than Douglas Crockford’s intentional removal of comments from JSON, but I don’t believe there are very many
the idea that the government should step in to prevent 12 year olds from asking the cloud for images of darth vader c walking or whatever is just silly www.wired.com/story/dis…
Another LLM feature I manifested
Claude finally made it to AI 4, glad to see them catching up to ChatGPT which is probably going to release AI 5 soon.............
is it considered a faux pas to shitpost on micro.blog? anyway i think about the slackware faq a lot.My kingdom for a password manager that reads my text messages and emails. Jesus fucking christ I am tired of SMS codes and magic links.
The biggest sign that OpenAI believes in the future of machine intelligence is the abject laziness they put into the consumer facing aspects of their own product. Take this as a compliment or an insult as you will.
I wish there was good third party highlight/bookmark support for LLM answers. A few times recently I’ve gotten responses that included some particularly useful framing or formulation. Wish I could save the whole conversation with the important stuff highlighted. I need a Pinboard for ChatGPT.
Final unfiltered thoughts on Claude Code
Saturday, March 29, 2025
The last three things on my mind from this week of building LDAPEnforcer from scratch.
- Claude Code only has the bluntest access to tools like the type system. When you ask it to make a conceptually simple change that I might start by deleting a variable definition and then looking for red underlines in my editor, it just finds one part of the code at random and changes it, and looks for references elsewhere. If it had a direct interface to type systems, linters, etc, that’d make it a lot faster.
- Watching it do dumbshit things like make up arguments to
cator read file contents withxxdwas cute at first and then very annoying (especially if you’re paying attention to/cost…). - It desperately needs an always-safe set of tools it can run. If I’ve already given you access to the codebase, you shouldn’t ask for permission to run grep, but as I mentioned previously you really should know that piping grep’s output to something else has implications beyond grep). But beyond that, is having the LLM directly call shell commands really the best way to do this?