TWIR 12/8-12/12

This Week I Read (TWIR)

I’m starting up a new series to collect different articles, projects, and stories that I find every week so that I have a way to keep track of cool shit on the internet. I’ve found that I often want to re-read or re-share something I’ve read before, and this way I’ll have an easier way of aggregating those links.

Without further ado, here’s the first TWIR! A common theme I’ve seen recently is a yearning for the early internet—quirky, niche hobby blogs, experimentation with new technologies, and media produced without the influence of AI or algorithmic curation.

Physical Media

Domain-Driven Refactoring Chapters 1 & 2

I’m reading through this for a book club at work. So far it’s highlighted a software engineering problem we’ve known about for a while, but consistently and willfully ignore: many difficult problems we face arise from accidental complexity as a result of cultural and organizational misconfigurations. There are almost always too many degrees of separation between the people using the software and the people writing it. As agentic coding improves in quality, I anticipate that we’ll be forced into addressing this and we’ll see roles like “sociotechnical architect” become common.

The Great Hunt

I’m a huge fantasy nerd, and my recent deep-dive has been Robert Jordan’s Wheel of Time. The Great Hunt is book two, which I’ve read before, and I’m enjoying it more this time around. I started reading the series because it was famously completed by Brandon Sanderson, who I’ve been reading and enjoying heavily. After The Great Hunt, I’ll probably read Wind and Truth before pivoting back to The Dragon Reborn so that I’m finally caught up in The Stormlight Archive.

Digital Media

Size of Life

This was a fun interpretation of those grade-school science book diagrams showing relative sizes of different living things, from nanometer-scale parts of cells up to some of the largest living organisms known to man. I thought it was a fun throwback to some things from my childhood that made me fall in love with science.

If You’re Going to Vibe Code, Why Not Do It in C?

Aside from the discussion about choosing a language that works well for agentic code generation, Ramsay puts into words a lot of feelings I’ve had lately about AI agent-driven software engineering. In particular, I’ve been working through a similar conflict between the efficacy of vibe coding and the personal satisfaction I get from solving hard problems myself. The sheer volume of high quality output I can generate using agentic coding tools makes me feel robbed of a skill I’ve worked really hard to cultivate. There’s definitely a mindset shift happening that I’m sure I’ll write more about in the future.

Japan issues mega-quake advisory after M7.5 tremor

I originally studied geology in school, and have always been fascinated by natural disasters as a result of geologic events like earthquakes and volcanic eruptions.

Turtletoy

This one goes back to the experimentation of the early internet that I mentioned earlier. Generative art isn’t something I’ve played around with much, but it feels like it hits a unique cross-section of art and tech that is compelling to me. I’ve been periodically checking back in on this throughout the week, and there are some pretty interesting pieces that users have created.

The f*** off contact page

First off, the design of Chan’s website is awesome. Seriously, go click around and experience the retro-OS-like front-end. The article and their website together reinforce the “early web” nostalgia hit that folks seem to be seeking out these days. It’s also just a good discussion about how we seem to have lost the plot on designing software for end users instead of shareholders. The writing style is also a big motivator for me to just be myself instead of over-professionalising and sanitizing everything.

From Text to Token: How Tokenization Pipelines Work

This was a neat, thoughtful look at something that I’ve been thinking about as my AI agent use has increased. I thought it explained the tokenization process and why some software might prioritize certain aspects of text over others depending on the use-case (code search vs. URL parsing vs. fuzzy matching).

The highest quality codebase

They took those hilarious early-stage generative AI memes about taking a prompt to an extreme by looping over it several times and applied it to a relatively simple web app to create the “perfect” codebase. Jokes aside, I think it reveals an emerging problem with vibe coding in that I’ve seen in my day-to-day. If you over-index on generically improving things without specific, targeted guidance on what needs to be improved, much of the generated implementation and test code can be sub-optimal.