They chose it early on, it works, and it makes obscene amounts of revenue. End of story. That doesn't mean it was the "greatest" choice, or has a perfect technical architecture.
Rewrites are never easy, even the bun rewrite. But a non-UI developer tool with a rigid API surface contract (and associated tests) will always be easier to trust after a rewrite than a partially tested UI tool with ambiguous functionality.
Your counter argument would be valid for a 2000 or a 2020 business decision about some tech stack.
But the whole point of their product is that it supposedly nullifies such "business" concerns around the use of technology, by making it cheap and fast to build whatever you like automatically.
That they wont, or worse, couldn't, speaks against that.
There’s two arguments in competition:
1. LLMs make it cheap (in the time sense) and easy to build
2. Rewrites and/or writing something in a native app or program is harder and more time consuming
I think I am willing to take it as an axiom that a native version of CC would be superior from a user perspective. Performance, etc.
I just don’t see how one can say that building things reliably good is easy now when the company providing these tools can’t even do it well.
Why would it be harder and/or more time consuming rather than the opposite? It's like having full specs and a full test-suite to match against the result you want.
I don't know how you can't see it, to me it's blindingly obvious: risk aversion.
Let's say the risk for a problem is 1%; hell, let's put it at 0.1% even. For a company at this scale even that amount of risk is too much.
I trust Opus/Fable to drive prod database migrations and backfills. I don't trust it with our financial ledger. I trust it with part of the infra. I don't trust it with backups. Etc.
You and others are arguing against a premise that nobody defended, namely "Claude can rewrite everything, for free and with zero mistakes". A bit of a straw man, don't you think?
And that's not even touching the fact that writing a GUI app is difficult for LLMs due to difficulties in it getting feedback and a "feel" whether it delivered what was asked (though I know people are working on it).
> I think I am willing to take it as an axiom that a native version of CC would be superior from a user perspective. Performance, etc.
How does that follow, and from where? I never once noticed any visual jank/lag in the TUI; not in iTerm2, not in Kitty, not in Alacritty, not in WezTerm and not in Ghostty. And even if we exclude those two, me and many other devs are quite fine with a TUI and don't miss a GUI program for everyday coding.
Not saying that our preference is superior -- but it'd be strange to blatantly claim: "for dev purposes, GUI > TUI/CLI".
0 - https://finance.yahoo.com/news/anthropic-ceo-predicts-ai-mod...
Now whether they can do it -- that's a separate, and a much more interesting, discussion.
This is a spectrum, it's not just 0% vs 100%. Even Fable frakked up a few things really badly in my professional work (though to its credit after a very detailed 2h chat it self-corrected and fixed all the blunders).
I would also challenge "fast" -- Fable (and I assume Mythos) are wicked fast and efficient and even they can't compensate for f.ex. slow recompilations or test suite reruns or security scans, linters etc.
Reminder that the Bun's Zig-to-Rust rewrite took 11 days with dozens of agents working 24/7 and the author put the cost they'd pay (if they had to pay) at about $168k.
Billions of dollars of annual revenue go through Claude Code, and the people who work on it must be a lot of millions in headcount.
The time matters a fair bit, it's probably time someone can't spend breaking things in the name of new features, but if rewriting the stack had even a tiny impact on retention or driving higher usage, it would pay back that $168k.
You might get more maintainable code for $168k by hiring humans, but if its something bespoke it isn't going to get you far (or far quickly).
Vibing is straight out gambling right now, the thing is, I don't know who the house is any more.
In my work I have Fable/Opus add bespoke linters and tie them to verify/precommit tasks (including in CI) so nothing of the reworks and 15+ initiatives we have in flight ever regresses.
Some discipline will remove that worry of yours. It's not nice to immediately assume that the rewrite's orchestrator (Jarred) did not use discipline. He even outlined a few false starts.
The major open harnesses are in the ~800k lines of code size ( https://simonwillison.net/2026/Jul/15/grok-build/ ) and CC is the oldest and quite-likely largest judging by features.
For an entity like Anthropic that's not even the cost of a single developer for a year. It's closer to what they pay a chef on premises.
That the showstopper for a better Claude agent is that they'd need to pay $168k or event $1M or even $10M in costs, can't be used as an excuse.
The argument I responded to argued that cost might be the prohibiting factor. I'm saying that for them, the costs mentioned are negligible.
What I argued was that you seemed to have been arguing an extreme point i.e. LLMs can one-shot stuff in any PL and that it does not matter in what PL they'd rewrite Bun.
If I misunderstood -- my apologies.
What I said is, those AI companies push LLMs as the replacement of programmers, that can do everything code-wise, say their programmers "don't write code anymore" etc.
So how come they can't have the LLM write something better than "react for cli" for their flagship agent product?
Claude Code is likely not well-specified and it's a product of rapid iteration and now nobody really knows how the whole thing works. LLMs don't do well when requirements are muddy.
I can understand people being more sold on Rust compared to Zig -- there are legitimate differences on the memory safety front and I can see how people will prefer one or the other.
Plus I believe Jarred (Bun's main dev) went out of his way to explain why he chose Rust over Zig in his blog post.
Of course the marketing angle might indeed be the explanation; not arguing against that, I simply followed with interest and Jarred seemed very reasonable to me.
they rewrite themselves, Bun team is under Antrophic umbrella
Does anybody really believe that it only took 11 days, one engineer and $168k in tokens?
Jarred used many agents and they worked 24/7. It tracks for me.
So yes, I do believe it. What makes you skeptical?
Cheaper, but not free (if you don't buy into the marketing promises too much). The bigger the project, the bigger the cost, even with a discount.
At the same time, the early versions weren't very good and you can be sure that any rewrite will also need to be similarly iterated upon until it is also good enough and polished.
If you do that and don't spend enough effort on making it be something polished --> your competitors have a better product and you lose.
If you pause feature development to give enough effort to the initiative, you don't get to add new features quickly enough --> your competitors have a better product and you lose.
Maybe their priorities lay elsewhere, like how I've noticed that the desktop app version of Claude Code has gotten both faster (no 2-4 second lag when switching conversations), more stable and usable over time, to where I enjoy using the models because of it, not in spite of it (though not that they haven't had bumps along the way, like that one cache invalidation issue, or how people didn't like the auto accept timeout thing). I don't doubt that you can get pretty far with gradual patches and improvements, instead of only big rewrites.
Honestly it's really cool for me to see Kimi having their own CLI too, same with OpenCode, Pi, Hermes (well more of an agent than just a coding harness but you get the idea) - there's so many competing solutions out there, each good or bad in unique ways.
Just wish we'd see similarly many GUI solutions, for now OpenCode GUI seems like the one I've settled on (cross platform and supports most models), though it's not exactly ideal either (feels a bit barebones, especially in regards to sub-tasks and progress/plan tracking, even ZCode seems a bit better in that regard, it was actually surprisingly good after they pushed out some updates).
In 2 years? A well-engineered TUI could be the difference between the Blackberry and the iPhone for all we know.
This can be true, while "nullifies such "business" concerns" can be false. Both are not dependent on each other.
Automation doesn't "nullify business concerns", it just changes them.
Eventually? I am sure they would agree. Currently it's you (and lots of people like you) who are doing the supposing, not Anthropic.
It's like "why did you go all in on buying scamcoin 3.0 as your investment strategy?" -- "I 5xed my money! End of story! It was fine!"
Or those choices just don't matter, it's fundamentally just "tabs vs spaces".
I am 46 y/o and still excited to code and solve problems with code and I still have trouble to admit to myself sometimes that people make dozens of millions with PHP and the tech hardly matters when you can throw bodies at the problem (and when that throwing of bodies at the problem actually solves it, of course).
Or many billions. eg. Facebook was originally written in PHP.
As a 52 y/o who cares deeply about software development as an art I don't have trouble admitting that the tech hardly matters relative to the economics, that was obvious to me very early in my career.
I have just never really cared because there has always been a path where you can have a decent career being paid to work on and with good technology in spite of the fact that the technology wasn't important to the economics.
I do fear that the rapid adoption of LLMs will probably cause the path I took to narrow considerably. There will likely be less companies willing to pay people to work on elegant technology when they are all competing against endless code generation machines (and this is assuming software companies can exist at all in this future). But at my age that's more of a concern for younger people than me.
But IMO the endless churning of code will slow down, if it has not already. Quantity does not equal quality as we know, it even erodes it. There are already consulting companies that are "vibe-code fixers".
I am also not sure it's about "elegant" code. What we might call elegant has multiple measurable axii f.ex. well-isolated effectful code layer so you can easily static-code-check it, domains and enforcement of boundaries between them, bespoke linting to enforce business processes and standing code discipline, and many others.
This isn't a minor nitpick, it's a pretty major UX issue.
Not saying it's like a massive business downside because I'm just one of a few users, maybe this affects their bottom line a little bit, but probably not by much.
Regardless, switching to pi has been a nice breath of fresh air. It just renders well and smoothly and handles terminal resizes well, which is especially important when used in a terminal window in PHPStorm.
Truthfully, the terminal in Jetbrains IDEs just isn't that great. It's gotten a lot better with their rewrites (yes, plural) of the last years, especially on Windows, but it can still be pretty dodgy.
Tabs vs spaces makes no tradeoffs. Tech choices do
However, that's not the same as the technology choices not mattering at all. As others have mentioned the product is buggy, and it appears Anthropic are struggling to continue to release new features within the existing architecture.
That's also not to say that Anthropic made bad choices - there's always going to be a tension between building quickly to capture the market now vs moving slower and building something more maintainable in the long-term.
They couldn't even fix flickering in their TUI and their new renderer is complete garbage.
They succeeded despite their tech.
It is rarely the case that technology choice is the make-or-break when it comes to whether a product is successful and achieves widespread adoption. Some choices are less ideal than others, but at the end of the day if you manage to make something that people want, the rest won't matter much.
Most people would choose to swim a lap through a pool of pee to get a billion dollars at the end. The pool of pee itself wasn't the once in a lifetime opportunity (for most).
Sorry, I'm bad at analogies.
This is just blatantly ignoring all of the glaring issues Claude Code has had over its lifespan.
Any kind of scrolling back, copying text, using their menu system - basically anything that isn’t typing characters has had/still has unaddressed bugs.
OpenAI shipped a competitive model and I’m over in Codex now. I have yet to hit a bug.
If you’re holding the SOTA crown, people will put up with your buggy mess. As soon as that crown slips your pile of trash becomes a huge liability.
Oh, that's buggy too. I just tried Claude Code on win10 powershell, and the first typed character goes in the wrong spot and can't be backspaced.
It is by far the the least reliable program on my machine, and every time I have to interact with it I feel like walking in eggshells.
So weird that the same exact people telling you that programming careers are now obsolete are the same group who haven't been able to fix screen flickering bugs for like a year...
Edit, the sonnet question shouldn't be taken as proof, it knows I'm a rust dev.
Nobody but a JS dev would choose react-on-terminal shenanigans over it
I could see CC getting a ratatui port eventually actually
the era in which the tech industry built software to last was over long before LLMs, especially VC-backed startups.
I'm pretty sure "built to last" was only ever more true in specific contexts, and wasn't particularly more true than today.
And the question of “if AI is so amazing, shouldn’t it enable easy development of a native TUI even if JS is easier at first?”
Don’t get me wrong I find great value in coding agents daily. Just finding the hype cycle tiring.
Programs which need more stability and performance than exploration are better done in statically typed languages like Rust. Therefore Bun is a platform for JavaScript programming, but itself is written in Zig or Rust.
The engineer is suggesting that it could be done cheaper and maybe with better outcome. Ironically, this is a classic business case.
If you have unlimited access to the magical development tool, then why would you not?
But they haven't.
A ringing endorsement!
You can make anything work when you have enough money to buy and radically change the entire runtime you’re relying on.
One must suspect that if they did not have insane amounts of money to burn, they could have tried other approaches to fixing the problems. Maybe engineering, perhaps.
I'd argue these tech companies got popular because they have good models, nothing else.
Even OpenAI took two years to fix basic chat scrolling.
Could it be that they aren't being entirely honest?
The point is that if you were starting a new TUI LLM harness today, you would basically use CC's architectural decisions as a guide for what not to do.
Much less fragmented compared to "write the app 3 times in 3 different languages".
But the "coding is mostly solved" narrative kinda doesn't match right? If good, correct, high-performance software is like, free now? Wouldn't you want it to be slick as hell, really reliable, all that? Even a little breakage costs a lot of money at that scale and pricing, it would be better than a wash if you put the magic code thing on the case.
"Claude. Do all employee work. Make no mistake. Notify in slack when revenue is double."
I'm sure there was some logical reason for shoehorning web technology into this stack given that we have a good 40 years or so of experience with interactive terminal programs that use curses/ncurses, alongside emacs, vim, almost the entirety of MS-DOS, and so on.
With js you get 1 thread at 100% utilization. Power usage and heat scale non-linearly with cpu utilization and 100% utilization on a single threaded js app means you will have ui lag. Other languages like golang would split work across 8 threads and have 8 threads at 20% utilization and this would result in less power usage. Claude Code would have been better off performance wise being an electron app because it would be offloading rendering to the browser and gpu.
Also, the architecture of Open Code is actually a lot better here than Claude Code. ClaudeCode does everything, including rendering in a single thread and it's all in js. OpenCode has a zig based tui renderer it offloads that work onto.
But I will also say these coding agent tuis do get unfairly maligned because they launch subprocesses and those subprocesses tend to be expensive. If you are using Rust analyzer with claude code, it's rust analyzer that's causing the majority of your problems.
> they need a high powered game engine rendering loop
Don't believe the bs on twitter. Claude code source was leaked, there is no high powered game engine rendering loop.
> It needs to stream text to stdout
These apps are doing real work taking text that is streaming in and doing syntax highlighting, calculating diffs, and rendering markdown. If the work was split across threads they would not use anywhere close to as much power and there would not be a slow ui.
If you think this is actually nothing, prove it. Put the code up on github that shows taking a constant random stream of markdown, code, and diffs and displays it in the terminal can be done cheaply on a single thread.
The reason the computer gets hot and uses more power is because the cpu is getting close to 100% utilization.
> These apps are doing real work taking text that is streaming in and doing syntax highlighting, calculating diffs, and rendering markdown.
The first argument is that these are expensive operations. They are not. And the second argument is the assumption that these are desirable things for an agent system to be doing. That's a personal preference but, personally, I don't want them and would appreciate a way to disable all of that to reduce CPU use.
A lot of people go “js is bad” reflexively and they are right for the wrong reasons. Its not because js (specifically v8) is slow. I have many websites and web based apps open and none fuck up my laptop as hard as claude code.
The UI itself requires laughably simple tech. The idea that a markdown renderer that can produce HTML at 50MB/s would use much CPU to process incoming hundreds of bytes per second is insane. CPU should be idle even if doing that constantly. Nevermind that between turns it has nothing to do whatsoever and should use zero CPU at all.
I picked the P90 because I remember running Linux ~1.0 and/or NT on it, with real multitasking and services etc. Those soaked up many of the cycles.
Another story, around Y2k/Win2k was asked to write a C program to record all the files/sizes on the C: in to file. Then something would be installed, my program run again, and the results diffed. Despite thousands of files the program was done before the enter key came back up. Boss told me that he thought the program was cheating.
That’s just how fast modern computers are these days, I said. When running C code anyway. That perf is still available today, compare astral’s uv to slow competitors. Just takes some investment, though probably less when boosted by VC bionics.
— Henry Petroski
Casey Muratori addressed this years ago with the windows terminal drama, https://github.com/cmuratori/refterm
Put up the code and prove it. We already know exactly how fast golang is in comparison to nodejs on basic numeric computations, it's roughly 40 to 50% faster.
I don't doubt this, but I would love details and citations here.
Bad coding is just bad coding.
It's also all cpu bound work. The majority of the stuff on your screen is being rendered by the gpu.
You are really, really not. Not at LLM inference speeds at least. Frontier models output maybe 200 bytes per second on the fast end. I guarantee you that idle BSD box is still processing more than 200 bytes of whatever at any point in time. Streaming 200 bytes of LLM output through syntax highlighting, markdown rendering, and diff formatting, will still leave the box basically idle if done correctly. Claude Code has not done it correctly.
This does not even speak to it pegging my CPU and keeping the fans running on full blast even when the laptop is supposed to be idle overnight.
You could test this by running "yes > /dev/null" in Ghostty. In Activity Monitor on my Macbook it shows `yes` using 100% and Ghostty using 0.5% of cpu.
Never had to dig through the xterm bug tracker to figure something out. I don't even know what version I'm running, or even what the versioning scheme looks like. I haven't changed the configuration since the first time I set it up. It just works.
Then something is waking it up and it's not asleep at all. Asleep the CPU shouldn't be running.
Is this because Codex is written in Rust and not JS? I dunno. I think it's more just "lighter" in general, or it certainly feels that way. It's probably possible to make something with a similar feel in JS, just perhaps not with the big honking mess they've created.
Claude Code seems to be an insanely complex program will all manner of cutesy features and telemetry features. The net result is approximately the same as Codex, but it’s pretty common in software engineering to find a simple thing and a complex thing that do more or less the same thing.
Proves LLMs don’t help with taste.
My question is moreso “why was it ever JS in the first place”
& the cost of a native rewrite would be cheaper than acquiring Bun no matter how you shake it
I hear some version of this anecdote constantly
yet despite using all these prompting tools since 2019 I have never once heard a lab or professional say “x model can do anything”
Can you provide me whatever specific source you heard for this?
Yeah, they "merely" say that they're about to break the singularity in a couple of years, that it's too capable to be safe to release, that it's going to take 50% (or all, depening on the day and audience) of programmers out of work, that it's smarter than Nobel prize winners, that their programmers don't write anything themselves anymore, and things to that effect.
Again… provide me a quote from someone doing this every day that says: “they're about to break the singularity in a couple of years”
The only people saying this stuff are random Internet commenters, breathless reporters and reactionaries.
The only person on record for a statement like this was Geoff Hinton about radiologists and he has since walked back his statement
Dario Amodei said in January that he expected software engineers to be replaced [0], so I suppose the quote should be extended to “can do anything… a software engineer can do”
And I think we can all agree that SWE is on the more complex end of all white collar jobs. So, my thinking is that if SWE can hypothetically be replaced, so can many other jobs. Hence, the claim the models can “do anything” intending to capture the majority of white collar work.
This, of course, is if you take their claim at face value - which I certainly do not.
[0] https://www.entrepreneur.com/business-news/ai-ceo-says-softw...
If I missed your point it was unintentional. Instead of being rude, try gentle clarification.
But now I’ve lost interest. Have a good day!
So then its misleading in the extreme to say “I’ve been told AI can do anything”
It amplifies hype when people make claims like this by including hyperbolic statements as though they are reflective of the state of the discussion.
That said, with respect what I clarified, I’m curious your thoughts.
I suspect user facing/fast moving code (UX layer) will move to dynamic systems with fast iteration times. Infra layer will move towards safe systems level environments like Rust. I'm not sure where Java/C# lands in all of this - it's kind of the middle ground between these two worlds but the tradeoffs change drastically with LLMs - my gut feeling says that TS/Python is good enough for UX work and Rust is better for systems work so it gets less popular going forward.
[1] https://avi.press/posts/2026-07-10-after-7-years-in-producti...
Also GHC has an interpreter. It powers the REPL. No need to compile everything if you’re just experimenting. I do find that in the Haskell world the REPL is criminally underused compared to Python.
And the rest of your comment doesn’t make sense; this shop is rewriting their infra (not UI) from Haskell to Python.
Because React is the only UI framework that takes the problem seriously. Everything else is stuck in the dark ages.
How HN gets this so badly backwards I'll never understand. Everyone on this site talks a big game about "there shouldn't be so many competing tech stacks, why can't everyone work together on one framework that does things right", and then as soon as that framework actually appears this site hates it more than anything.
Yeah, good question. OpenAI decided to rewrite Codex in Rust about a year ago[0].
In fact, since rewrites are that easy, what do we need Bun for? Why doesn't everybody just port all of their Javascript code into Rust?
I've read and heard in videos that Javascript is a pretty good language for AI to write code in. Apparently this is because there is so much training data out there. Also Javascript avoids problems of multi threading and memory management that can mess up the AI in other "more performant" languages.
So maybe Javascript is not the worst choice for writing software fast with AI
The latest frontier models are competent at Rust and Swift and all manner of other less widely used languages.
The more important factor is how good they language's compiler is at kicking out actionable error messages, since one-shot code generation isn't as important once you have a coding agent loop.
Like Ada or Ocaml or Haskell or something.
Or even require like MC / DC testing or MISRA C verification.
Or even some of the languages that apply Hoare checks, like Ada SPARK or FRAMA-C or VALE or whatever.
I don't have the budget to test how that would work but it seems interesting.
It was unpleasantly surprised when I learned the hard way that LLMs are not much better at translating than writing from scratch.
The more you look into how they work, the more you see that it doesn't really give them a huge advantage, if the classes are big enough. You can tell them to break each function up and translate them one-by-one, but the errors compound, you can't test most of it until you have a lot done, and in the end, it really isn't much faster (and sometimes it seems to be a lot slower) then just telling them to start over from scratch.
The downside is... If you have a system that already works, you don't want to start from scratch and test everything all over again...
To me it comes back to the premise I hear a lot from AI hype men (of which I do not subscribe but it’s worth clarifying that I find great value in using it for code) that the code doesn’t matter anymore. I just don’t know how to square that claim with the idea that we’re limited still by JS/TS being easier to use or approachable despite inferior TUI toolsets.
Heck, a buddy and I once chatted about the likelihood of k8s running as the control plane in a prototype autonomous vehicle.
top/btop/htop on the mac are always fun to run and see what's up.
Unless you use an ancient teletype, you don't have to redraw everything. That would make any interactive applications way too slow/flickery.
You can move the cursor arbitrarily in the terminal and start overwriting characters from there. So you need to track state to know what is "dirty" and needs refreshing. Occasionally you issue a full redraw to catch missed artifacts left behind or when the terminal is resized (SIGWINCH).
This is close to the same "why Spotify is a chromium embed?" question. Because it works, and users are ok with it.
"I'm using technology I know that will get us there" is not the same as treating every problem as a nail. It's making a practical choice that probably also had time constraints and other factors we don't know.
How does this support your argument? The tech didn't even exist in the 80s or 90s.
Saying "They wouldn't have made the same choices 40 years" is meaningless in this context. There's a lot of things they wouldn't have done back then that we do today. And this is a field that has been changing by the week recently, which makes even less appropriate.
This idiom now apparently means "I disagree with your tech choices" or in your case, "Someone 40 years ago would disagree with your tech choices".
I did not say they wouldn’t used react in the 90s (obviously). Anybody familiar with terminal uis (a technology which did exist) and perhaps any other ui system would not have chosen this design.
And no it’s not just me as evidenced by the many confused comments any time this is brought up.
It's hell to maintain for not much gain (for a use case like this at least). As much as it's become a meme, JS and web tech in general has become extremely portable and stable.
I also don't think Anthropic bought bun to make their TUI better. They could have forked it, they bought bun because it incidentally was excellent for the way agents prefer to work and they wanted to capture that audience.
Their new flagship product (earning them billions of dollars in revenue per month) was dependent on a platform maintained by a tiny startup.
Buying Bun was a very rational way to reduce that risk, epically since it also got them some top tier engineering talent.
Thinking about that further, I wonder if that was part of the rationale for switching from Zig to Rust that they haven't talked about?
Zig is a much riskier bet for your multi-billion dollar cash cow than Rust is.
But saying that out loud would be rude - they took steps to NOT openly criticize Zig, even after Zig's founder did not show them the same courtesy.
If this does not get Anthropic 'The Presidential Medal of Strategic Restraint' I don't know what will.
No, you are not crazy. They do crazy things with unlimited budget, and still their chat app flickers when using.
They should just port pi-tui from pi coding agent, since they have no clue.
It's like asking "why does everything run on Windows?" Because everything runs on Windows.
I have been trying to keep away in the last couple weeks and it was all win for me. I still come down here sometimes when I am stressed with real work since it is a strong addiction to see “how terrible the plebs are doing”.
This is especially useful when you're trying to evaluate behaviors while changing state surgically.