There's this thing where it seems to me like people are stuck in limbo between their old lives and this spirit world of "agentic coding". Looking at an interview with Nathan Sobo talking about Delta (of course I'm curious) and the first time in the interview that he seems to get excited is when he's talking about the technology behind it with conflict-free replication while to me it sounds like the premise of this pivot is that the code doesn't even matter anymore and it's all just conversations with agents now. Why should anyone care about data structures in this world? Isn't it pointless? To the person who started a business to build a code editor because he cared about code and wrote it for a living, a conflict-free data structure to sync state would be interesting and something to learn and be passionate about, but to the person with fifteen chat windows talking to agents on a codebase that is just a fleeting artifact, all of that is surely no longer relevant at all. Yet that is the topic of the conversation.
In case you didn't know, there are people being paid to have opinions on AI. Like Hank Green, apparently! I too have opinions on AI. Sadly, not the opinions that one gets paid to have.
”It’s just a tool.” Oh yeah. Just a tool. That you rent. From a nazi. A tool. Which consumes the atmosphere. To produce an answer. Which is wrong. Just a tool
I presume that Modular getting bought by Qualcomm and Mojo going 1.0 is the last we'll ever see of that. Well played to Chris Lattner for bamboozling Qualcomm into making him a billionaire, too bad the livable planet has to be destroyed for y'all to get your private jets
An excellent explainer for why LLM resource consumption is worth talking about https://lists.debian.org/debian-vote/2026/08/msg00101.html
It's hard to say this without sounding like a complete asshole, but, like, this was immediately obvious to me without having used the tools myself. I now have been exposed to the tools (Claude Code specifically) via work, but my initial impression has only been reinforced.
"The problem is that code produced by an AI agent looks reasonable, but can contain ‘hard-to-spot bugs’ that end up causing major problems. As a result, you should carefully review your agent’s output, but this is difficult. As the engineer told me, it’s “famously hard” to understand code you didn’t write yourself, so this extra step becomes “easy to just blow it off (especially when we are all trying to ‘10x’ our velocity).” Soon, systems start to break."
I don't really understand that phase that the developer in the story seems to have gone through. I don't understand how to experience
“If I had to guess,” he concluded, “I’d say a task that would have taken me a week now takes me 2 days.”
A part of me kind of wish that I could feel that way. It must be a good feeling! Everything suddenly got easier!
A DIATRIBE AGAINST BIOLOGICAL DETERMINISM
I am of the strong belief that software patents are evil, companies that file software patents are evil and people who hold software patents should be ashamed and ridiculed. If you're a developer and one of your brags is holding a software patent, you suck and I'm not impressed. This mini-rant brought on by watching a talk where the speaker introduced themselves by bragging about their software patents. Pathetic.
Just went and got a photo taken to renew my driving license. “Are you happy with this photo?” Well, no. I preferred the photo where I was ten years younger
Carson Gross on AI in education:
Very interesting to me personally, I had a major crisis following my last teaching gig, the consequences of which were 1) gram and 2) me not taking on more teaching gigs, at least for now. Gross seems to have been able to find a path to keep going, but one aspect that he didn’t mention at all and which concerns me is the addictive nature of these tools.
It sounds reasonable to say oh, we should focus our efforts on the students who want to learn. But from what I am seeing these tools are so tempting and so insidious that even those who want to learn end up using them very heavily very quickly. In an environment where there’s cocaine dispensers everywhere, people using and recommending cocaine everywhere and almost every company is exclusively hiring cocaine addicts it seems naive to say “the ones who want to learn won’t touch the cocaine”.
I also wonder where Gross himself is in his journey here. His first VM was written without AI, his second sounds to have been largely vibe-coded. The question is, does he have the self-control to write a third VM and not use AI? At every decision point, there’s an easy answer just sitting there at the back of his mind. I can just take a little bit, just once, just this one time…
Someone expressed frustration at people talking about Steve Yegge and his (mis)adventures. Personally I talk about Steve Yegge mostly because I find him very funny. Steve Yegge is just a funny character, he's like a mid-50s rich dude with a ponytail repeatedly crashing his convertible Ferrari into a fence while trying to impress some kids by burning rubber and doing a donut. Except he doesn't have a Ferrari. He has an imaginary friend he calls "The Seneschal".
I’ve discovered a neat life hack, I call it the “Steve Yegge bullshit loop”. Instead of getting paid six figure sums to talk about how I use the bullshit machine, predicated on spending six figure sums on the bullshit machine, I short circuit the bullshit loop by skipping all the bullshit. Saves me a lot of time and anxiety
As I get older I’m feeling more and more the need to live according to my convictions, which unfortunately for me my convictions makes very difficult
Finally got around to watching “The Unreasonable Effectiveness of Constructive Data Modeling” by Alexis King and hell yeah what a great infodump on using type systems effectively. It also makes me want to make a variant of the talk which is basically “go watch that talk” plus Gleam, as in look at this language which nails this perfectly.
I think Zig is pretty good in this regard as well fwiw, and Rust of course can be used well if used well (not a given…).
So I guess Mitchell Hashimoto decided that his political views weren't clear enough and decided to clarify by launching a startup funded by Guillermo Rauch and Tobias Lutke.
Watching live drawing. Captivating!
Gram 3.2.0 tagged and released. Hopefully it works, the release was made while travelling.
3.1.0 didn’t compile for Windows, hence the version jump (but users of the deb/rpm repositories may have gotten the 3.1.0 release yesterday).
You know the typhoon has hit when you're getting rained on in the hotel lobby
First they came for the em-dashes
And I did not speak out
Because I was not using SmartyPants.pl