I've been doing A LOT of work to prep for richer TUIs, input events in particular, so not many (visual) updates, but I finally got a really nice key event system in place. Here's a list component

Acquisition announcements What's said: "Everything is going to get so much better" What happens: Everything gets significantly worse, or disappears altogether, but some people make a lot of money and go off to repeat this cycle.

sw.kovidgoyal.net/kitty/keyboa... has been so valuable in helping me interpret keys when stdin is in raw mode. I'm sure I'm not going to implement every detail perfectly across every terminal, but hoping this guide will help me get pretty close!

Me: "In JavaScript, how can I create a writeable stream from an async generator?" LLM: "Here's how you can create a readable stream from an async generator" Useless. This is probably 95% of the responses I get. Utterly useless. Why do I keep trying?

Pro JS tip: if you have `Symbol.toPrimitive` implemented in your class, it will be called with a single argument ("string") when the instance is interpolated in a template literal, even if you have `toString`.

Fun learning for today, for macOS: add the line ``` auth sufficient pam_tid.so ``` to `/etc/pam.d/sudo_local` and you can use touch ID for `sudo` commands in the terminal. Apparently should live through system updates too, unlike a lot of these things.

I need to move myself off of nix for program / dotfiles management. It's nice when setting up a new machine, but otherwise a huge PITA. Also, Node 25 has been out for a month and (AFAICT) is not in nix-unstable. No thank you — and no, I'm not making my own derivation. I got better things to do, yo.

Marketing for so many AI companies is the fiercest dumpster fire of absolute trash I've ever seen (looking at you in particular, Noon AI). Can we just stop with this already. You all should be ashamed of your existence. Your moms aren't proud of you. I despise you.

Me, to Claude: give me some words that end in ABC Claude: *lists out 14 words, more than half don't end in ABC* 🤦‍♂️

I started out updating my personal site. I wanted to extract the static site generator I built for it, to share with everyone. I decided I needed to polish the CLI a bit first, so it needed a nice Terminal UI (TUI). Now I'm building a fancy TUI library. This is what my unemployment looks like 😅

Hope you all enjoy the slipperiest and the sudsiest of soapboxes ᕕ( ᐛ )ᕗ

Cloudflare's "Verifying you are human" page is such a painful experience. It's such a sad state of affairs when we need to build crummy tech to protect against our other crummy tech. It wastes our energy and potential on something that _should_ be unnecessary. It degrades the lived experience. Sad.

Debugging layout algorithms and rendering for a TUI = plus signs, vertical bars, and hyphens everywhere

My thesis: AI isn't ushering in the Imagination Age. It's actually reinforcing our current age — the IKEA age — where we crank out low-quality crafts that satisfy the needs of a lower class (ostensibly) vs building long-lasting stuff with strong foundations and positive compounding effects for all.

My cell service provider sent out an email saying you should enable 2FA to keep your account secure, but does not have 2FA anywhere in account settings 🤦‍♂️

To add to this point: when I write code, I'm not just writing code, but also constantly: * Debugging * Reevaluating my architecture / goals * Registering edge cases / thinking about tests * Building a mental model of the system * Considering the customer impact of every line * much more; no chars…

Seeing a lot of people — often from engineering directors, VPs, and CTOs… — arguing that it's great that LLMs free up devs from writing code so they can focus on just high level. I got a bone to pick with you all.

Instead of "worse before it gets better" why don't we just try "better... and then even better!" 🤔

Super happy with v1 (gist.github.com/thegedge/491...). The basic layout engine + renderer is <300 lines of (unoptimized) code (compare: react-reconciler = 1.68MB unpacked). I'm going to keep exploring to see how easy it is to write more complex components. Thoughts on what we have so far in 🧵

This is coming together so effortlessly and beautifully, it's almost scary. I guess Halloween _was_ only a couple of days ago, so spooky vibes are still lingering.

I've been exploring slightly different ways to specify a UI. I'm building a TUI and don't want to be another Ink. I'm avoiding build steps (well, I guess you don't _have_ to write JSX) and heavy dependencies (e.g., react, react-reconciler).

TFW you write out something correct and Cursor's autocomplete wants to rewrite it to something wrong.

I've been working on a virtual buffer that has decent Unicode support, and wow, such complicated. I'm going to try to jot down learnings and whip up a post for this! Least fun learning: you can't identify how a string will be presented in a terminal on its own. You need access to the font.

Although I've been fairly active on here recently, I was thinking how microblogging is...not great? So much nuance is lost in such short-form writing, and we — myself very much included in that "we" — often lean towards punchy posts that often aren't constructive.

I've been curious about why EVERY product seems to want to jam "AI chat" into everything. This bandwagon approach seems anti-design / anti-product. Similarly for other patterns that don't benefit users. Designer friends: how you feelin? PM friends: why do so many products have anti-user features?

It me. I definitely struggle with the "don't get bored" bit. I've been fortunate to get to run on a lot of teams in my career though, in a "go where I'm needed" capacity. Startup idea: job board for companies to post small contracts, some kind of automated vetting/payment process, etc.

I've been building out a custom TUI library for m8t. What a great learning experience, if you want to learn how a terminal handles having data written to it (especially all the ANSI escape sequences)

Pro tip: If you have a unit test like `describe("function", ...)`, do `describe(function.name, ...)` instead to create a strong coupling between your unit test name and the unit under test. Bonus: symbol renames (e.g., F2 in VSCode) will automatically update the test name.

I remember a math prof I had in my undergrad, in the context of teaching us about "or" and "exclusive or", confidently stating that human language "or" is never "exclusive or" lolololol This same prof marked one of my *correct* answers wrong because I jumped a step in getting to the answer 🙃

TFW you're working on something to parse ANSI escape sequences and a failing test emits those escape sequences without the "reset" bit, so all of your terminal output is now a hyperlink

I love Novo Amor, and this new collab with Yvette Young (always love being introduced to new artists through collabs!) has such a beautiful texture to my ears. Chill vibes. So staccato. 8-bit math rock-esque. Very fun. (linked to the YT video, premiering in 6 hours, but you can listen on Spotify)

Yet another day of trying to use Cursor to help me out, but wasting a bunch of my time. Hitting `Undo All` and writing the code + tests myself since, unlike this so-called "AI", I actually have _intelligence_ 🙄

In other (bad) news: the US is judge, jury, and executioner now. Imagine you were alleged for some crime by someone and they toss a grenade at you. Colombian government described it as "like applying the death penalty in a territory that is not yours" www.bbc.com/news/article...

Nobody: Elon Musk: "I don't feel comfortable wielding that robot army if I don't have at least a strong influence."

Me asking Cursor's LLM chat to help me find undocumented functions 🤦‍♂️ I only post a small sample, but my experience with Cursor's chat is ~10% success, ~20% of "ugh, okay, I'll clean up the mess", and the rest is bad. I continue to be shocked at Cursor's valuation. We've set the quality bar SO LOW.

I saw an article recently where Wikipedia is saying its website traffic from humans is down ~8%. It made me think, if true: LLMs need tons of training data, but if LLMs eliminate too much human traffic from the websites it sources, those websites may no longer be able to afford to exist, and then… 🤔

When I left my job in July, I set out to do a huge feat of human endurance: run a marathon. I've done some halfs before, but running a full one still felt like a huge leap. This weekend I completed it 🎉 I was on the tail end of recovery, but almost achieved my original goal of <4 hours. So close!

I switched to vitest and you just immediately know when something was designed well. Intuitive. Great out of the box. Node team should have taken the time to learn from existing test frameworks/runners. I may try to dig up some GitHub issues or something. Curious how they approached designing it.

I've used a fair number of test frameworks/runners/whatever you wanna call it in my career, and Node's test runner is feels so much worse than anything else I've used so far. It's actively working against the way I want to use it. Curious if anyone else is unhappy with it?

I rarely get sick, but when I do it's less than a week to race day (or some other important event that can't be rescheduled).

I have SO MANY draft posts in a note. Maybe one day I'll fire them up on a schedule. Will probably lose all of my followers. Would be a good time.

literally no one: cursor's autocomplete, if I were to "vibe":

Saw this today when I (doom)scrolled through LinkedIn. This shouldn't be what we celebrate. If an employee at your company has to miss their once in a lifetime celebration, in my opinion, you've failed to build a company worth working for. But, hey, at least the shareholders are happy, right?

To be fair, I played a bit of a role in this complaint. If you nest Markdown paragraphs inside of `<p>` in your MDX file, ya gonna get invalid HTML (fixed it by switching to a `<div>`). I wish they'd help me out there, y'know? I like tools that help out babby webdev.

MDX is amazing at producing invalid HTML. I love it so much.

Every time I read through comments, I'm always surprised at how most argumentative comments are shockingly bad at making those arguments. Was this always the case and social networking highlights it, has social networking itself degraded it, or has language education been degrading over time? 🤔

Seems like a huge oversight that `Array.isArray` doesn't narrow a readonly array type 🤔

Another fun one today! Once I tidied up some CSS — font sizes and switching away from zoom to transform, in particular — I was able to quickly (just a couple of minutes!) create an overview mode for my presentation framework 🎉 I've been wanting this to quickly assess any (visually) broken slides.

Spotify nailed the kind of Thursday morning vibe I'm looking for. I _know_ I can channel grandmother energy this morning.

You ever work on something, and after you're done step back and take a good look at it and think "hot damn, you did good". It's a good feeling :)