I'm reading the article, but one of the things that annoys me about this whole discussion (and wider discourse in the modern age in general) is spending time refuting marketing/bullshit claims....
I'm reading the article, but one of the things that annoys me about this whole discussion (and wider discourse in the modern age in general) is spending time refuting marketing/bullshit claims. "Anthropic claims AGI will solve everything and that's bullshit, so everything they say is wrong" is, unfortunately, not true, and I feel like this entire article frames it's argument like that out of the gate. No doubt the marketing IS bullshit, but that doesn't mean there's 0 value (although that may also be true).
Explaining the direction beforehand would have been more traditional in an infrastructure project like this, but meanwhile the delay conveniently allowed the story to be carried by sexy headlines like The Register’s Anthropic’s Bun Rust rewrite merged at speed of AI. Much invest. Very wow.
Annnd i'm already kinda doubting the authors ability to present unbiased facts and let us draw our own conclusion. The story from Bun's dev, which yes was hidden in HN comment threads, is that they didn't expect it to go this smoothly, so it was more of a "well fuck it lets see what happens" followed by "oh shit its already done? uhh didn't plan for that".
That may not be true, but you can see those comments on HN, first where they said it was unlikely to be merged, and later the exact explanation above (I don't have it word for word and unfortunately don't have time to do a HN search like last time but it might be in one of the two linked threads from that comment (and there are good tools to search HN comments out there if you just search it, not that anyone really should want or need to).
Still, as I read Andrew’s piece I found myself cheering out loud. I may have briefly jumped around the room. Some called his take a “meltdown”, all I can say is he’s gained a new fan today.
Yeah okay i'm just not going to vibe with this author am I. Even if you are, I wish we could save the editorial for after the facts.
Who to believe
Okay it's mild, but I hate headings like this halfway through an article that has clearly already demonstrated bias. That said I do agree mostly with their summation here.
Running at “crunch time” all the time is bad for health and bad for productivity. That is a robustly established fact about knowledge work
I...hate this whole discussion when it comes up. There are MASSIVE success stories from this kind of nonsense, and they tend to get ignored. There is 0 doubt it's unhealthy and 99.99% of the time outright stupid, but it has produced results (not always great), but it reminds me a lot about musical/sports training and how that should all be done (see the dramatized thesis of Whiplash).
To be clear, I agree, fuck that. Red flag, I'm out, good luck, maybe you're the next Microsoft, I can live with missing that boat.
The rewrite rationale is fluff/All pros no cons
I basically agree with these segments, and have said as much about providing actual data points, but I have more on that later.
I expected the next sentence to discuss Bun’s style guide, why it wasn’t working, perhaps how it evolved over time… nope. They seem to just pay lip-service the primary way the community addresses their problem, shrug their shoulders and move on. Did I miss something?
Bun chose Rust’s borrow-checker because AI isn’t enough.
Okay so, "the big point" that I think is somewhat missed in this discussion.
Lets start two weird examples:
Automated Cars:
An interesting criticism of automated cars is that they're designed to work on roads and infrastructure for humans. If you could just spin up automated level 5, no human needed, vehicles tomorrow and design with that in mind you wouldn't need stop signs, lanes, lights, or basically anything but smooth surfaces. These are tools for humans, not automated systems.
An indie darling: VVVVVV is a game that did very well. Lots of good reviews, probably made a lot of money. So you as a potential game developer find out the dev has open sourced the code and look under the hood.
To explain to the non-coder inclined, this would be like being an engineer today, and everyone LOVES this new car that came out. Good metrics across the board, does all the right things, so out of curiosity you look at one in a shop to see what they did. You notice they're using a lot of screws...like a LOT of screws...like holy shit wait they don't have a single weld in this vehicle ITS JUST ALL BRACKETS AND SCREWS OH MY GOD.
The fuck is my point:
My belabored point is that I feel like this is roughly where AI is kinda heading. Coding languages are tools to people put ideas/solutions into production. VVVVV was a person who didn't know all the tools they had, but they still made it work, and current automated solutions like AI are running on "roads" built for humans when they could probably do quite a bit more with something more tooled for them. That does not mean AI is a solve all, or even long term profitable.
However when you consider people like the Bun dev, who yes, ignore obvious and potentially easy solutions like "just follow the style guide", "just tell the AI to do it in Rust" IS a viable answer, and a potentially lucrative datapoint. It sounds like, yes, AI writes worse Zig code, and that code is harder to review when it breaks. But perhaps because of its popularity or perhaps because of features like the borrow checker, AI might write better Rust code. And that code might be easier for a human to review in the cases it's needed (i find the article to be pretty pedantic about that point. "Oh you said it's AI only" yeah yeah we know humans still have to check but we're not looking for 87.73% or whatever. Again marketing vs reality issues)
IF THAT'S TRUE which nothing we have at this point says it is, that's a strong argument for what the Bun dev did. They don't explain it well (in part because it'd take more self reflection than has been shown), but the real argument here is "hey if you're going to be an AI shop, use languages AI likes ,then it works even better".
There's a very legit argument around the rest of this, but it's worth pointing out that most of that doesn't matter if the Bun dev is right (and, ironically, Bun probably doesn't matter in the long run if the Bun dev is right...). Why care about style guides when there's an option to throw it at an AI and let it crunch through? The answer will be "how much does that cost vs talented devs". However I don't think it's a bad thing that "worse" coders like the Bun dev can output better products if AI helps, it's only a question of if it can do it cost effectively.
This seems to be a major point that is conveniently ignored fairly often in these discussions (I will give a disclaimer here that I have mixed results with agentic tooling, my experience has been...
… perhaps because of its popularity or perhaps because of features like the borrow checker, AI might write better Rust code.
This seems to be a major point that is conveniently ignored fairly often in these discussions (I will give a disclaimer here that I have mixed results with agentic tooling, my experience has been that it’s overall pretty awful at writing anything over a certain level of complexity). For years, Rust advocates have exclaimed things like, “If it compiles, then I know it probably runs!” I myself have said things like this about Rust. Agentic tooling is all about building harnesses for a clever statistical tool that makes a really good guess about how to complete some text. Is it really any surprise that making the lowest level of this tooling more rigorous helps this autocompletion loop write working code?
I'm not sure I understand all of the hate Andrew (Zig creator) got from his article post. I saw a lot of opinions saying he was unprofessional, but it makes a lot of sense where he's coming from....
I'm not sure I understand all of the hate Andrew (Zig creator) got from his article post. I saw a lot of opinions saying he was unprofessional, but it makes a lot of sense where he's coming from. If I spent 10 years working on something, and there's essentially a campaign against that thing, I'd probably react the same way.
Hello! I don't know if anyone else called him unprofessional, but I did so because (a) he accepted money from someone, and (b) had a private meeting to discuss particulars of their project. Then...
Hello! I don't know if anyone else called him unprofessional, but I did so because (a) he accepted money from someone, and (b) had a private meeting to discuss particulars of their project. Then the moment that person did something he objected to, he used the contents of that private meeting against them.
I'm not a devout Zig-ist, but irrespective of every other point brought up, that's enough to get you blacklisted in my books.
Also I think this post is pretty pro-Andrew, and anti-Jarred? Pull quote:
Still, as I read Andrew’s piece I found myself cheering out loud. I may have briefly jumped around the room. Some called his take a “meltdown”, all I can say is he’s gained a new fan today.
Yes, you're right. I was just using this as an opportunity to bring it up. I suppose my bigger point is I just don't understand why this got as big as it did. Both sides can be justified at the...
Also I think this post is pretty pro-Andrew, and anti-Jarred?
Yes, you're right. I was just using this as an opportunity to bring it up. I suppose my bigger point is I just don't understand why this got as big as it did. Both sides can be justified at the same time.
I'm reading the article, but one of the things that annoys me about this whole discussion (and wider discourse in the modern age in general) is spending time refuting marketing/bullshit claims. "Anthropic claims AGI will solve everything and that's bullshit, so everything they say is wrong" is, unfortunately, not true, and I feel like this entire article frames it's argument like that out of the gate. No doubt the marketing IS bullshit, but that doesn't mean there's 0 value (although that may also be true).
Annnd i'm already kinda doubting the authors ability to present unbiased facts and let us draw our own conclusion. The story from Bun's dev, which yes was hidden in HN comment threads, is that they didn't expect it to go this smoothly, so it was more of a "well fuck it lets see what happens" followed by "oh shit its already done? uhh didn't plan for that".
That may not be true, but you can see those comments on HN, first where they said it was unlikely to be merged, and later the exact explanation above (I don't have it word for word and unfortunately don't have time to do a HN search like last time but it might be in one of the two linked threads from that comment (and there are good tools to search HN comments out there if you just search it, not that anyone really should want or need to).
Yeah okay i'm just not going to vibe with this author am I. Even if you are, I wish we could save the editorial for after the facts.
Okay it's mild, but I hate headings like this halfway through an article that has clearly already demonstrated bias. That said I do agree mostly with their summation here.
I...hate this whole discussion when it comes up. There are MASSIVE success stories from this kind of nonsense, and they tend to get ignored. There is 0 doubt it's unhealthy and 99.99% of the time outright stupid, but it has produced results (not always great), but it reminds me a lot about musical/sports training and how that should all be done (see the dramatized thesis of Whiplash).
To be clear, I agree, fuck that. Red flag, I'm out, good luck, maybe you're the next Microsoft, I can live with missing that boat.
I basically agree with these segments, and have said as much about providing actual data points, but I have more on that later.
Okay so, "the big point" that I think is somewhat missed in this discussion.
Lets start two weird examples:
Automated Cars:
An interesting criticism of automated cars is that they're designed to work on roads and infrastructure for humans. If you could just spin up automated level 5, no human needed, vehicles tomorrow and design with that in mind you wouldn't need stop signs, lanes, lights, or basically anything but smooth surfaces. These are tools for humans, not automated systems.
An indie darling:
VVVVVV is a game that did very well. Lots of good reviews, probably made a lot of money. So you as a potential game developer find out the dev has open sourced the code and look under the hood.
To explain to the non-coder inclined, this would be like being an engineer today, and everyone LOVES this new car that came out. Good metrics across the board, does all the right things, so out of curiosity you look at one in a shop to see what they did. You notice they're using a lot of screws...like a LOT of screws...like holy shit wait they don't have a single weld in this vehicle ITS JUST ALL BRACKETS AND SCREWS OH MY GOD.
The fuck is my point:
My belabored point is that I feel like this is roughly where AI is kinda heading. Coding languages are tools to people put ideas/solutions into production. VVVVV was a person who didn't know all the tools they had, but they still made it work, and current automated solutions like AI are running on "roads" built for humans when they could probably do quite a bit more with something more tooled for them. That does not mean AI is a solve all, or even long term profitable.
However when you consider people like the Bun dev, who yes, ignore obvious and potentially easy solutions like "just follow the style guide", "just tell the AI to do it in Rust" IS a viable answer, and a potentially lucrative datapoint. It sounds like, yes, AI writes worse Zig code, and that code is harder to review when it breaks. But perhaps because of its popularity or perhaps because of features like the borrow checker, AI might write better Rust code. And that code might be easier for a human to review in the cases it's needed (i find the article to be pretty pedantic about that point. "Oh you said it's AI only" yeah yeah we know humans still have to check but we're not looking for 87.73% or whatever. Again marketing vs reality issues)
IF THAT'S TRUE which nothing we have at this point says it is, that's a strong argument for what the Bun dev did. They don't explain it well (in part because it'd take more self reflection than has been shown), but the real argument here is "hey if you're going to be an AI shop, use languages AI likes ,then it works even better".
There's a very legit argument around the rest of this, but it's worth pointing out that most of that doesn't matter if the Bun dev is right (and, ironically, Bun probably doesn't matter in the long run if the Bun dev is right...). Why care about style guides when there's an option to throw it at an AI and let it crunch through? The answer will be "how much does that cost vs talented devs". However I don't think it's a bad thing that "worse" coders like the Bun dev can output better products if AI helps, it's only a question of if it can do it cost effectively.
This seems to be a major point that is conveniently ignored fairly often in these discussions (I will give a disclaimer here that I have mixed results with agentic tooling, my experience has been that it’s overall pretty awful at writing anything over a certain level of complexity). For years, Rust advocates have exclaimed things like, “If it compiles, then I know it probably runs!” I myself have said things like this about Rust. Agentic tooling is all about building harnesses for a clever statistical tool that makes a really good guess about how to complete some text. Is it really any surprise that making the lowest level of this tooling more rigorous helps this autocompletion loop write working code?
I'm not sure I understand all of the hate Andrew (Zig creator) got from his article post. I saw a lot of opinions saying he was unprofessional, but it makes a lot of sense where he's coming from. If I spent 10 years working on something, and there's essentially a campaign against that thing, I'd probably react the same way.
Hello! I don't know if anyone else called him unprofessional, but I did so because (a) he accepted money from someone, and (b) had a private meeting to discuss particulars of their project. Then the moment that person did something he objected to, he used the contents of that private meeting against them.
I'm not a devout Zig-ist, but irrespective of every other point brought up, that's enough to get you blacklisted in my books.
Also I think this post is pretty pro-Andrew, and anti-Jarred? Pull quote:
Yes, you're right. I was just using this as an opportunity to bring it up. I suppose my bigger point is I just don't understand why this got as big as it did. Both sides can be justified at the same time.