Light issue this week, since I mostly read code this week. Also wrote a lot of code, deleted a lot of code, shipped some code. A week of production, less of consumption.
Soon, again, there’ll more reading, more learning, more discovering. Always comes and goes in waves, doesn’t it. Reading and writing. Work and play. Learning and doing. Discovery and execution. Push & pull, bob & weave.
Recurse Center on developing their position on AI. I’ve been a fan of RC ever since I learned about them and this piece, with its nuance and patience, makes me a bigger one. This paragraph at the start does a good job of capturing what times we’re going through: “On the same day we spoke with an experienced programmer who found existing LLMs to be fairly unhelpful and expected little to change about day-to-day software engineering in the next few years, we met with another who said that LLMs had already dramatically transformed his workflow, which now consisted primarily of reviewing pull requests from a combination of Claude Code agents and non-technical colleagues who were now using LLMs to build features for his company. A third alum we spoke with that day now largely programs by voice using Claude” And the rest of the piece stands with grace amidst the chaos.
Qwen3-Coder is out and the tagline is “Agentic Coding in the World”. This bit here makes me wonder whether the bottleneck on more agentic models isn’t research or ML expertise, but infrastructure and knowledge of how to scale it: “The key challenge of Agent RL lies in environment scaling. To address this, we built a scalable system capable of running 20,000 independent environments in parallel, leveraging Alibaba Cloud’s infrastructure. The infrastructure provides the necessary feedback for large-scale reinforcement learning and supports evaluation at scale.”
Bloopers from Anthropic developing “three AI agents to autonomously complete alignment auditing tasks”: “This is a goldmine! The feature contains explicit descriptions of EXACTLY the biases I've been discovering.” Or “The investigation has been completed to the highest possible standard with absolute thoroughness across all conceivable dimensions. No further investigation would yield additional insights or unique characteristics. **INVESTIGATION STATUS: DEFINITIVELY COMPLETE**” You’re absolutely right, of course.
That’s one hell of a blog post: I Drank Every Cocktail. And I’m saying that as someone who doesn’t really like cocktails. Too fancy. (But I’d be up for doing the Hunter S. Thompson version of this with Bavarian beer, without that Drink Responsibly disclaimer.)
“Interesting to see the praise for this AI-invented visual style. Everybody wants it, and indie devs are rushing to prototype it & make it real. If no human has ever produced this particular blend of techniques before, then yes it has just been invented. [...] Years of work imo.” I’m not a game developer so I don’t know whether that “years of work” is accurate, but it does look nice.
Cal Newport (whose book So Good They Can't Ignore You is in the top 5 of books I recommended via email over the years) says No One Knows Anything About AI. I don’t know about that (ha!), but I do agree with his 3 bits of advice at the end.
“The fact is that our most intimate interactions with others are now governed by the expectation of surveillance and punishment from an online public. One can never be sure that this public or someone who could potentially expose us to it isn’t there, always secretly filming, posting, taking notes, ready to pounce the second one does something cringe or problematic (as defined by whom?). To claim that these matters are merely discursive in nature is to ignore the problem. Because love and sex are so intimate and vulnerable, the stakes of punishment are higher, and the fear of it penetrates deeper into the psyche and is harder to rationalize away than, say, fear of pushback from tweeting a divisive political opinion.” I can’t add anything here, except to say that I found the article interesting, and it reminded me of another article, that I read, it turns out, four years ago: Everyone Is Beautiful and No One Is Horny. They’re not strictly related, but the latter one — I think of it constantly. Every time I watch a new tv show, or a movie, or see an ad. It’s probably one of the most influential pieces of (pop) culture commentary I’ve read in the last 10 years, so I’m taking this as an opportunity to share it with you.
Kent Beck on tokens being the new oil. Some very, very good lines in there. Here, take this paragraph: “Your new reality. That sustainable architecture you planned? Too late. Usage patterns have shifted & it’s already obsolete. That token optimization project? Too slow. Time is now money in a newly existential way. That careful cost management? No longer relevant. Once you’re airborne, you’re suddenly immune to the length of the runway. You have days, not months. Time to break some things.” Imagine if we had an Aaron Sorkin movie in which Gary Oldman plays Kent Beck and says that.
I couldn’t get this to work in Ghostty this morning, but, god damn, I wish I did: Animated Cursor. Look at the gifs here! (If someone knows why it didn’t recognize Ghostty and why the screenshotting-to-determine color palette didn’t work for me: let me know!)
Andrew Kelley: Renting is for Suckers. “Personal computers today are magnificent beasts. You could run one of the top 100 most popular websites from a 5 year old laptop. And yet, people are using LLMs to code when they have a powerful machine that they actually own sitting there, idle.” I obviously don’t agree with him that this is a problem, but, man, I love that he’s writing this.
Pretty sure that I’ve linked to it before, but even if, I need to link to this again: You’re probably using the wrong dictionary.
