What are you doing this week? Feel free to share!
Keep in mind it’s OK to do nothing at all, too.
Three toy projects, with very different rules around AI. Because I need a control group to run an experiment, I suppose. Also because writing fancy stuff by hand is still most excellent fun.
Project 1: A lovingly hand-built language targetting WASM. GitHub claims this is a Toy WASM Lisp, but there's no actual Lisp there. Right now, I am engaging in an enormous yak-shave by implementing a low-level WASM "system" language with (some day) integrated support for both linear memory and the new WASM GC types. Recently implemented:
// EXPECT: f() == 1
export func f() -> bool {
1u8 < 132u8 && 20 > -20
}
Currently in progress:
func nothing() {}
// EXPECT: f() == 2
export func f() -> i32 {
nothing();
local x = 2;
x
}
For this project, my AI use has been limited to (1) small Zeta2 inline completions in Zed, typically a few lines at most, (2) using browser-based AI as a research agent to deal with truly awful tree-sitter and WASM docs, and (3) occasionally allowing a open-weight AI agent to debug some weird tree-sitter issue. Of which there are many. The result: My low-level knowledge of the WASM ecosystem has expanded by leaps and bounds. As always, hand-crafting code is quite enjoyable and rewarding, and oh my goodness is this more fun than reading thousands of lines of PRs. (It helps that this is far from my first compiler project, so I can do all the compiler bits straight out of my own head.)
"Project" 2: Having Fable write a toy Prolog implementation in WASM WAT assembly. To be clear, I didn't write any of this. Here's my tiny initial design. Here's Fable's plan, after asking me a few clarification questions (this plan is not simple—there's a lot of esoteric bits here). Here is Fable's entire working miniKanren-like Prolog in 1,000 lines of WebAssembly, done as a 61-minute 1-shot for $16.75 in tokens. It's actually a really well-done bit of code, like something you might find in Lisp in Small Pieces. Here's a web-hosted Prolog interface you can play with.
Result: I got a cool toy Prolog, but I learned nothing I hadn't already learned about WebAssembly from project 1, or about Prolog that I hadn't learned from The Reasoned Schemer. A human can only learn so much about WebAssembly GC or Prolog implementation tricks by watching an AI work for 61 minutes, after all. Actually learning anything is now strictly optional, and oh my isn't that terrifying.
"Project" 3: Having Fable reimplement my favorite bits of the classic PalmPilot application HandyShopper. Again, I didn't write any of this. You can find my original prompt, Fable's plan, screenshots, and a downloadable APK on GitHub. This was a 1 hour 45 minute 1-shot for $44.40.
Result: I now have a shopping list feature (per-store aisle sorting!) that I've been sadly missing for decades, and it's great. But I also learned almost nothing about modern Android development. If I actually want to learn anything at all from this, I'll need to sit down a read a whole bunch of code. And that will still teach me far less than building an app myself.
Thoughts. Draw you own morals from these 3 projects, 1 by me, and 2 by Fable with a bit of my input. To put it bluntly, I fear we are about to see a huge split between projects where maintainers actually understand what's going on, and projects that no human has ever looked at. Similarly, we'll see a split between actual programmers, and people who just ask the models to do all the work. See also my thoughts on lemon markets for software.
I have feelings. Part of my reaction is mourning for work that I love, certainly. Another part, weirdly, is a bit of happiness that ordinary people can probably now actually build a little custom thing that works for them, when they couldn't before. (I am a multi-decade believer in the importance of software freedom. And not just freedom for people who could probably get hired by Google, but ordinary people, because the alternative was always the enshittified world that we actually got.) Overall, ugh, I fear that we are swimming in deep waters and there are sharks lurking. Or to use another metaphor, much of the establishment is skating to where the puck was last year. An almost nobody, me included, is really skating to where it's going.
I've also been skimming a whole lot of well-written anti-AI books these last few weeks. But most of those authors are still living in 2025 and nothing there "speaks to my condition", as religious writers might put it. I'm not really interested in arguing about water use (ffs, just used closed cooling loops) or community AI councils or whatever. My central worry, really, is that we have a tiny handful of companies hoping to spend trillions of dollars with the dream of fully automating human-level cognition. And really, pick your dystopia: WALL-E, where nobody thinks any more? SkyNet, where some fool builds something smarter than us? Some grim corporate cyberpunk future where companies can buy tireless near-human-level cognition for $0.20/hour, rendering actual humans mostly surplus to requirements? I want an activist theory that can at least think about those risks, too. Or about whatever partial version of those risks might actually come true.
And even if the AI companies fail, they will burn a substantial portion of humanity's actual wealth trying. We need to have a real conversation about whether anyone actually wants what we'd probably get if they somehow succeeded with their pipe dream. And if we don't actually want the pipe dream, should be spend trillions of dollars chasing it?
Started the week with a tilted production server. OS was unresponsive, and BMC was stuck. One could fix the other, but when both are hung, good times are ahead.
I’m finishing a small browser puzzle game called WordCell. https://chunq.itch.io/wordcell
It’s a compact word solitaire puzzle: move stacked letter tiles by alphabetical order, uncover the letters needed for the target word, and use special tiles to break tricky board states.
This week I’m tuning the 10-puzzle campaign so it lands around 30 minutes without becoming too slow or too easy.
It’s upgrade week, so Kubernetes and AWX updates for me. Also, it’s my wedding anniversary this Wednesday, so looking forward to that!
Reading Thomas Hobbes' Leviathan.
I'm also putting "fluid simulation" on my hobby project list, but I need to understand more to start.
I just picked up a copy of Leviathan, due to going down a brief rabbit hole about social contract. Seems dense, so I'm expecting to be reading this one off and on over the next year or so.
Hopefully hearing back about a job application I sent last week that I felt good about.
It's been too hot to go cycling but I'll try to get back in the saddle this week. Also need to clean the house, do some pushups, do some laundry, organize my office, and do some budgeting. Will also try out 007: First Light
Besides that:
Been chugging along with kaizen on Konform Browser (a Gecko-based web browser which I just attempted sharing here on lobsters for first time) stable releases in-between preparing the next major version.
Being based on FF ESR means this is an unusually busy and exciting time of the year since we have a rough year of feature updates in one go to rebase on. Will see if I get some time this week to adjust for and adopt the redesigned Settings layout or if we keep it classic at least initially. Feedback and impressions are much welcome if anyone tries out the browser (already aware of the readme being in need of a significant rewrite, thanks ;))!
I'm working for a few more weeks before I have two weeks vacation. My current contract is a one-man embedded project, so I don't have to be vacation stand-in for anybody. I usually like the peace and quiet of working in July, it's even better when you don't get more responsibilities.
As for personal projects, my programming language has been on the backburner for a few months while I have done other stuff (like I've been playing around with F# and SDL3). I moved my gaming PC from Windows to Fedora, it works pretty well[0], but it certainly led me to think less of the kind of people who hang around Linux forums to berate anybody who dares have a problem with their setup[1]. And I've been thinking about "What if Terralang but with Scheme instead of Lua?"
[0]: Which means Slay the Spire 2 works. I've beaten A10 on all characters now.
[1]: Case in point: I had a problem with getting my fancy nvidia card working. I search for a solution and end up on Fedora's forum, where someone has had a similar problem. First out is some busybody who is scandalized that anybody DARES using an Nvidia card. Useless. Then after some more or less relevant guesses, there is a promising lead in that one has to wait for an unspecified time after one has installed the nvidia packages for a background compilation to be finished, which occurs without any feedback. Which is very helpful as it's the correct solution, but of course this is coupled with berating the user for not understanding that before they reboot they should wait for this background task which is never shown or referred to in the UI. Finally, somebody mercifully delivered the incantation to rerun that compilation separately.
I'll be spending most of it at work, but I do have a to-do list for what little free time I can steal back from my job:
Figure out why my Ragnar (Raspberry Pi Zero W, PiSugar and Waveshare ePaper v4) will not swap over to AP mode when I'm surrounded by unfamiliar networks.
Decide which eurorack module I want to build next. I set out at the beginning of the year to put together a useable skiff of only DIY kits and learn about modular synth after getting addicted to playing with an East Beast. So far I have an MKI x ES VCA and VCO, a Sequencer and LFO from Guru Gara, and Synthrotech's Power Up completed. Suggestions welcome.
Draft the next post for my blog, which has sat neglected for too long.
entering the last two weeks before i start my new job. this week's goal is to finish painting the hallway / staircase and maybe get a few sessions at the gym now that it isn't "serious threat to health" levels of hot
I'm working on a simple simulation application in Interlisp and LOOPS (Lisp Object-Oriented Programming System), the object extension of Interlisp.
Vacationing in Norway with my 14yo boy, visiting my parents, my sister and her family, and hoping to meet up with some friends and go hiking.
Doing the Usagi game jam! Come check it out!
On vacation, so mostly chilling, and by chilling I mean - making a box combinator implementation in Zig to catch up with the new 17.x changes :D
For work: Today we just rebranded ThetaEdge to Thetix. Announcement is here: https://thetix.ai/blog/thetaedge-is-now-thetix
I also released a major new feature allowing users to trade stocks and options within the app. This saves on both time and improves safety (no risk of fat fingering). User reviews trades in a cart, can modify them, and execute when they are ready.
For the rest of the week I am concentrating on bug fixing and performance improvements. Plus testing different models and running our Eval suite to see if I can further improve the AI system we built.
For personal: Today I just released version 0.8.1 of Abject which has some new protocol optimizations making the UI much more responsible, especially when you are running the server on another machine. Also includes support for Abjects running in a wasm sandbox allowing anyone to build them in other languages. I ported KnowlegeBase to C++ as part of the system abject for both a performance bump and to test it out. The release also includes audio output support so you can build abjects that play sound (very nice for games for example).
For the rest of the week I'll be thinking of a way to further migrate components to native as necessary and continue adding Abjects at the system level. I can not overstate how useful I am finding this project for day to day life.
Abject is free and open source: https://abject.world
Writing: I am also going to work on some writing this week. I have a cool blog post in progress that I will publish on my personal blog showing my earliest AI work from 20 years ago and how it ties to where we need to go. I also have an idea for a blog post about how maybe the future of software looks like the past a bit in Japan. My blog is https://blog.mempko.com