The best description of this section of my site is something like “Twitter of my own” and Nadia’s words:
I often scribble half-baked ideas, reactions to things I’ve read, or something useful I’ve heard. Sometimes they turn into longer blog posts or projects, but most of them sit in my notes app, unused.
…
(Please note: you are crawling my brain. These are rough notes, which means ideas are experimental and conviction level is highly variable!)
I use my Stream for capturing quotes from things I’m reading, and Twitter/Bluesky for reaction-posting. This page is another entry in the genre, but one that I own outright. One where I have even less concern for quality, since nobody is subscribed to it.
October 2026
15th October — 1:18 AM
If you’d told me 3 years ago that agents would be superhuman at programming and not half bad at visual design, and that I’d have more website-making work than I’ve ever had before… well, I’m not sure I’d have believed you.
But such are the times.
8th October — 8:22 PM
I should never have got into this keyboard shortcut design business. Hours of my life lost to picking the optimal keys that nobody will bother using.
September 2026
29th September — 4:59 PM
IT IS DONE! This site has achieved perfection.
It took a day or two to come up with the new layout for the writing page (now called Evergreen), and to add the custom headings per post type (they look great!), and to bring perfect consistency in styling across the site. Every single nit, bug, or small annoyance I’ve felt about the site has been sanded down. I even successfully resolved my dissatisfaction around how my internal vs. external links were styled.
27th September — 6:24 PM
People are welcome to disagree but I think Opus 5.5 is actually the beginning of the “end of programming”. Not the end of software, of course, because (for those with eyes to see) software is SO much more than code and moving bytes around.
26th September — 3:08 PM
This website is on the brink of becoming a perfect one. It only lacks a few things:
- A design for the Writing page that sorts everything by topic and contextualises the important posts.
- Distinct styling for essays, notes, worknotes and riffs. My initial idea here is to give them all different title styles (very large font size for essays, bold for notes, uppercased mono font for worknotes, and smaller font size for riffs), and maybe highlight the nav bar item corresponding to their section?
25th September — 12:12 PM
I don’t enjoy reading Hemingway, but I am more and more drawn to the idea of writing as simply and briefly as possible. Perhaps my distaste for Hemingway has to do with plain prose in fiction, where I prefer a pleasant voice and flowing sentences? Anyway, I resolve to ruthlessly edit, readers’ preference for dramatic oration and repetition be damned.
23rd September — 3:57 PM
When I tell people “editing is how you get better at writing”, I need to make it more clear that I don’t mean at the level of sentences, or sanding down the more annoying tics in their voice.
Those do affect the overall quality of the piece, but the important thing is to be “editing” your approach to communicating the Idea itself. Trying out different narratives and frames, deleting the intro, skipping or adding examples in the right places, etc.
10th September — 1:06 AM
Looking for a phrase that means “take responsibility”, in the sense that software should have minimum-viable affordances for all the data types it handles, and not depend on a third party utility.
Concrete example: if your app deals with images, it should have basic cropping functionality (most messaging apps do this).
9th September — 12:30 PM
I’m going to refrain from actually having a public opinion on childhood education until I’ve had experience with tutoring schoolchildren, or have children of my own.
But I do think you can just get by with learning to read/write. Words, music, math notation and code. It’s all text. It’s also the only thing you’ll really remember how to do forever, compared to learning factual details or process knowledge that you may or may not need in the near future.
9th September — 3:03 AM
Since I don’t have much else I can use LLMs for, I’ve just been blowing through the list of features and enhancements I’ve wanted to add to this site for a while. I’m no Gwern, but I might soon be able to pass as Gwern-lite.
3rd September — 5:47 PM
New assumptions for software in the next few years:
- Client devices will stop improving, they might even regress on average
- Memory usage on any device comes at more of a premium
- Person-sized software, including tools, will have a larger user base
August 2026
27th August — 1:46 PM
I am probably world-class at coming up with ideas for novel collections of things.
23rd August — 9:12 PM
I am approaching Gwern-levels of design tweaking and subtle improvements on my site. I just added text-wrap: nowrap to all <code> elements so they don’t break mid-snippet. But I think at this point it’s somewhat deserved; my site actually has enough content on it now.
21st August — 1:15 PM
I would like to create a combination of web analytics and progress updates across my different internet projects. Mostly as a design challenge, but also because my personal site is for live activity, and history needs its own place.
21st August — 1:13 PM
I need to make this file easier to write to, or depreciate it as a failed experiment.
July 2026
14th July — 6:20 AM
A funny series of realisations:
- I can add an API to tldraft that fetches page contents, so I can build a static site at docs.joodaloop.com to host my work notes.
- Wait, I can just write my worknotes in Markdown and host them the same way. tldraft should be just for collaborative things.
- Why docs.joodaloop.com? I can just have them live on my site, all I need is a place to add them
- Workbench is the perfect place for work notes, duh!
9th July — 4:45 AM
Three things: Twitter timeline, library items, live ephemeral chat.
6th July — 9:44 PM
Every time I go without the internet for a few hours, I am given a taste of the calm focus my mind is capable of when it knows no distractions are forthcoming. I cannot afford to live like this, but it changes the experience of being enough that I am always tempted to.
3rd July — 5:42 PM
More and more, I see how the best software people I know have a dogged persistence rather than any particular genius for math/logic. Granted, my criteria for best is personal, and isn’t “CS researcher”-shaped as much as it is “make things”.
June 2026
26th June — 11:23 PM
An unfortunate consequence of the glorification of System 2 thinking is that it comes at the expense of System 1 instinctual action.
In practice, S2 is at least as flawed as S1, if not more so. Because it relies on our limited reasoning capabilities instead of wisdom that resides in the body, borne of repeated practice against reality.
Obvious implications here for “general intelligence” (just think hard) vs. tools trained on the task itself so that it becomes second nature.
21st June — 3:58 PM
I think I’m going to make my next (long-delayed) shareholder letter a simple “list of things I’ve thought about in the past months”, and maybe make it a special birthday issue.
7th June — 8:48 PM
I’ve been skeptical of pattern languages, because the ones I’ve seen have been very wishy-washy about matching solutions to problems. And in domains that are subjective, they are often too prescriptive.
But I think I can make a pattern language for local-first software/distributed systems that is actually good, mostly because it’s exactly the kind of domain that deserves one.
4th June — 11:08 AM
It’s very silly how much I do not rely on the internet as a way to make friends and gain allies. Actions that don’t seem to make sense relative to how much I believe in friendship-over-HTTP otherwise:
- Avoiding all group chats
- Not asking mutuals with experience for help (“I shouldn’t bother them”)
- Not emailing new people to make friends (“I can’t ask them to maintain another comms thread”)
- Under-involving people in projects that are low-pressure asks (draft review, intros)
3rd June — 3:16 PM
I wish people who otherwise take their websites seriously didn’t delegate things to an LLM (especially design!). Even “skilled prompting” won’t help much, because:
- the stupid little LLM tells will overshadow everything
- the LLM will steer since it controls all the iterations
April 2026
26th April — 4:17 PM
We could make UIs feel more flexible, and much nicer for high frequency interactions, if we made sure to implement an equivalent for every capability that text editing offers. Some that come to mind immediately:
- Custom re-ordering, how we just move text around
- Freedom of copy and paste, with ability to easily modify after pasting
- Great ergonomics for “editing actions” like drag-to-select and insert-between, etc
- Annotations, allowing for universal commentary
25th April — 3:56 AM
The easiest way to have a poor relationship with practice is by not doing it earnestly, i.e. first deciding on what you want, and then finding the kind of practice that will get you there.
19th April — 7:12 PM
Found this video of a Mk.gee live performance from 2019. It’s so…bad? Like, if that were my band, my songs, and my singing, I’d be ashamed of it. The progress from there to Two Star and the Dream Police is crazy.
18th April — 6:17 PM
luv 2 come back from vacation and spend hours getting my stuff in order again
16th April — 4:32 PM
Being accused of being older than my age never gets old, eh? Might be worth figuring out which traits at which age lead to it, e.g. having a low discount rate is more impressive when young, and less so at, say, 40 years old.
4th April — 6:17 PM
Immense scarcity of tacit-ish learning material in fields that are physical, where it’s not just about acquiring new abstractions. The closest you can come is by correcting easy-to-notice mistakes and using a variety of analogies, hoping they land for the learner.
3rd April — 10:03 PM
Asked a friend “do you actually want to make apps?”, with the implication being apps were over/a solved problem. And the second I asked it I realised I still wanted to make them myself. Cool little self-realisation right there.
March 2026
23rd March — 5:03 AM
Animations on phones are fine because your finger is usually blocking the thing being tapped, so delayed feedback is good. On web you can clearly see, and also don’t want to wait. Re-clicking is more expensive than tapping (?)
21st March — 5:54 PM
How much of what I do everyday is “work”? I’m not confident enough to say “it all adds up”, knowing how careless I am about time spent.
18th March — 12:16 AM
I might be wrong, but it feels like the approach that I’ve arrived at for SSSync might actually be the most generic sync solution one could have. It improves on Replicache by making use of named and versioned mutations, and a more efficient (but easier to get wrong?) server contract than the CVR tracking they require.
If done right, it will play well with systems like ATProto too, since the relay and PDS architecture is very similar to my snapshot and delta sync approach.
14th March — 12:28 AM
The internet is made up of links (not anymore, but bear with me here), but they are author-decided. At most, the few places that still have comment sections will allow the passing reader to make hypertext connections of their own. This is how Pagerank worked for a long while, I imagine. But now, why not a set of linked connections drawn on a layer above the sites themselves, by visitors who draw links between things? Like are.na, but for the web as a whole.
13th March — 1:15 AM
Thinking about app development frameworks again. Despite how terrible Apple have proven in recent years, Swift still looks like the best bet given:
- No need to manage memory
- Apple has a new Head of Design in Steve Lemay
- Windows is going to continue to be horrible
- MacBooks are more accessible thanks to the Neo
11th March — 12:12 AM
I am madder than I should be at the state of design in internet software, given that I’ve not made any serious contributions of my own yet. Shouldn’t have let myself lose momentum on Scatterpad work, even if the problems I’m stuck on are genuinely difficult ones.
4th March — 11:12 PM
Feels like a good time to define a bunch of file formats around things that all the personal software seems to wrap. Stuff like time tracking, metrics, lists of media, notes, etc. Apps don’t have to follow the schema exactly, but they should at least have a clear mapping from their schema to the standard.
But, to be honest, this is probably not a big deal, in the “just ask the LLM to look at the schema and write the migration script lmao” sense.
3rd March — 11:12 PM
I should probably start selling a service like “why you don’t actually use your personal website” more explicitly with my Webcraft sites. Mostly because I’m annoyed at abandoned websites, and particularly annoyed when they’re websites I’ve made.
I’ve seen enough failure modes to have fixes for the most common ones, and am especially sensitive to things that are likely to be Ugh Field generators.
3rd March — 9:20 PM
I’m bad at reading DD-MM-YYYY style dates because my brain doesn’t convert numbers into months fast enough. And almost every way to write the actual month name alongside a date just looks funny to me:
- “14 March” reads as “one three march”
- “14th March” is almost nice enough but semantically wrong
- “March the 14th” is just too much
So DD • Month is my chosen compromise, it’s sufficiently quirky while remaining functional, with good whitespace to boot.
3rd March — 7:32 PM
I just set up this page, and am already feeling the useful effects of its existence. In particular, each time I start typing here, I’m nudged into one of either:
- tweeting out the thought
- drafting a Riff instead
- framing things in a way that’s slightly more polished than I would if writing to just myself (which usually takes the form of a lazy fragment)
The reason this is good is because putting in just a little extra effort seems to improve both my thinking and the verbal expression of said thinking. And this place is low effort enough that I actually start writing more often. If it graduates out of here to another place, all the better. If it doesn’t, it’s still somewhere and the stakes are low.