Hey David Gerard, long time follower second time caller. (Here's the first time.)

So I recently saw a post of yours—attached below—about a post from a lead developer of the "Julia language" trying out Claude Code (which I will call "Clode" for the rest of the post for no particular reason except it saves me keystrokes and is funny) and finding it "not very good".

David Gerard @[email protected] Aug 29, 2025, 12:11 PM Link to post

the extremely bottlenecked project lead for the Julia language tries out Claude Code. you'll be shocked to hear it's not very good.

https://discourse.julialang.org/t/the-use-of-claude-code-in-sciml-repos/131009/8

"Claude can only solve simple problems that a first year undergrad can do, it can’t do anything more, it’s pretty bad."

discourse.julialang.org Jul 24, 2025

The use of Claude Code in SciML repos

So it’s pretty public that for about a month now I’ve had 32 processes setup on one of the 64 core 128gb RAM servers to just ssh in, tmux to a window, and tell it to slam on some things non-stop. And it has been really successful!.. with the right definition of success. Let me explain. I think the first will answer the others. Basically, Claude is really not smart at all. There is no extensive algorithm implementation that has come from AI. I know some GSoCers and SciML Small Grants applicants...

I came into the post imagining from your description, perhaps, an overworked lead developer for the Julia programming language. Lead developers of open source projects are famously overworked and under-appreciated. In my mind, this lead developer, swamped with issues and feeling the crunch, opts to try using Clode to see if it lives up to the hype and if it could make his life easier. This developer finds it to be less than useless, and then writes a scathing critique of it on the Julia language forums. In my opinion, this is how the post is framed. However, it is not that.

The forum post is actually from Chris Rackauckas, a lead developer of SciML, which is an organization that's all about scientific machine learning.¹ While he seems to be part of the Github organization for the Julia language, he has only made 9 commits to the language itself, 6 of which were for the README. Since this Rakauckas is a lead developer of an organization that focuses on machine learning, it's unsurprising that he would be at least a bit AI curious.² In the forum post, Rackauckas states that, while he finds Clode useless for anything advanced, he still felt as though there was a niche that could save him time.

So from this one principle arose:

This claude thing is pretty dumb, but I had a ton of issues open that require a brainless solution.

Rackauckas continues on to say that Clode allows him to no longer have to spend 12 hours each week working on trivial Github issues. He solves this now by throwing "32 instances of claude at the problem for 8 hours."

I have had to spend like 4am-10am every morning Sunday through Saturday for the last 10 years on this stuff before the day gets started just to keep up on the "simple stuff" for the hundreds of repos I maintain. And this neverending chunk of "meh" stuff is exactly what it seems fit to do. So now I just let the 32 bots run wild on it and get straight to the real work, and it's a gamechanger.

I was planning on ending this post by saying that, despite everything, I didn't consider this post a win for AI boosters. I initially considered it to still have enough caveats to maybe not be an entirely positive spin for the groups of people who are personally or financially invested in the success of AI, but that was before I found Rackauckas' blog. As of time of writing, his most recent blog post is titled A Guide to Gen AI / LLM Vibecoding for Expert Programmers, written about a month after the forum post. In the blog post, he promotes the idea that you can treat AI as a cheap army of junior developers,³ he claims that expert programmers are the only people who should be vibe coding, that he pays for a $200/month Clode Max subscription, and that in one month he used enough tokens for $5,200 of compute. On top of all this, I was able to find out that he has a major role at two different AI companies. We are no longer talking about just some lead developer on an open source project. This guy is someone whose behavior can be explained through the lens of financial motivation. Boosting AI and keeping the bubble going means he will continue to get paid. Which perhaps matters since I still have no idea how he's paying for the 64-core 128gb RAM servers⁴ he is running all his instances of Clode off of. At this point, the only possible dunk on AI remaining is the potential own-goal of only saying that AI could save him 12 hours a week of trivial work, given how much financial stake he has in the success of AI.

Despite everything, there yet remains some hope. Rackauckas has submitted his own post to the Orange Site,⁵ and the reception seems to be... mixed. Negative comments include "this appears to me to be a surprisingly low-effort article" from polotics, "this may be the most 2025 blog post ever written" from ppqqrr, and "I've got a TONNE of low-hanging fruit that I can't give to an intern, but could easily sick a tool as capable as an intern on. This was not that." from jpollock While the comments section of an Orange Site post doesn't make for the most interesting reading, it still would've made for a more factual basis for a post than what you've written.

Thanks for reading, Evie Finch

(Pretend I have a really cool email signature here. Like the Paul Allen's business card of signatures. Trust me, it's really cool.)

here is a list of every muffle or mute i use on tapestry, my rss client of choice. a lot of what i'm hiding is promotions of some sort, usually for members-only content, which i think makes sense. but what has really started crossing the line is stuff talking about ai. i follow some normal-ish publications about web dev and stuff, and the amount of ai hype i've seen on the is driving me kinda crazy, so i am now exercising my power to be the god of my own feed and silence it all. it's muffled though because obviously this hides any ai criticism from people i might care about so it's just muffling it, requiring me to click through to see it. tragic.

i’ve been reading a bit about ai stuff recently and one thing that has really bothered me is that a lot of people using the framing provided by corporations by calling the problems "hallucinations". they’re not hallucinations. a computer can't hallucinate. these "hallucinations" are caused by the computer being unable to fundamentally understand the world because they can't understand. period. it just bothers me because you're criticizing ai, but you're criticizing it using their language and on their terms and that sucks. don’t play their game and don’t play it by their rules.