> In July 2025, the Securities and Exchange Board of India (SEBI) alleged that Jane Street used multiple entities for market manipulation and barred it from accessing the market.
As I understand Jane Street they have done great things for OCaml and they also do their own web framework (I guess that big money bin needs a lot of dashboards).
I think the designer here takes a wrong approach and sort of falls into engineer envy where he wants to make prototypes as deep and realistic as possible. And that is not really the most important part of the design job.
The most important thing is that the right thing gets build (why do we need a JSQL input box? what do we actually want? what are other ways to do this?). And this is often better done with pen and paper sketches, meetings, observation, discussions ... rather than too quickly narrowing on a particular design and spinning into discussions on whether buttons should be on the left or on the right, LLM intricacies etc.
People are much better at saying what they don’t want rather than thinking about what they do. It’s more effective to throw everything in using LLMs, then using human judgment to sculpt away.
Taking it slow with paper and interviews works if you think about what you're doing. The other option now is shut down brain and press the slot machine lever one more time. One of the designs has got to be good soon.
I think AI programming just has a bliss period for a lot of people where it feels like you can solve every problem with a prompt. And you can, for a time. Eventually the chickens come home to roost and you realise what a mess you made.
Give it some time. We’ll figure out what LLMs are good and bad at. I think vibe engineering will eventually go up on the wall next to static vs dynamic typing and vi vs eMacs.
At least, that assumes AI models won’t keep improving by leaps and bounds. We’re in a transition period. It’s gonna be chaos.
> eventually the chickens come home to roost and you realise what a mess you made
But a lot of times they don't. Devs building websites for small businesses used to be a thing, and it was almost universally a crap product. Practically every restaurant in my small town has been able to take control of their own website with AI, and it's a better experience for everyone.
Rapid digitisation meant a lot of low-value work wound up being done by high-value people. The economy is pivoting away from that configuration. The high-value folks getting displaced are pissed, partly rightly, partly out of spite. The folks who had to pay those bills are ecstatic, mostly overexuberantly, but in part correctly.
That era is already long gone. Those things have been built in Wix for over a decade. And since then, I haven't had any friends or family ask me for a simple site. The low-value work is already automated.
Your comment is valid, but not for the reason you think. What you should be talking about is the grunt work done in our field to non-trivial applications. Adding a search box to a table, or add an extra field to a form, etc.
So you think about a ticket that often might take a few hours, but in badly architected system might take a week, add a field to a class, edit the DB structure (maybe manually, maybe through via an ORM generated migration), add it to DTOs, add a validator, add it to the FE definitions, edit the page layout, etc..
Low value work that until now had to be performed by high-value employees.
> Low value work that until now had to be performed by high-value employees
The current trend is a proliferation of internal tooling. I've honestly found quite a bit of it useful. The rest is the usual should-have-been-a-Google-form nonsense.
I can use a local model to build a restaurant website. The model generates code fast enough that even my 10 year old server can generate all the code needed in about an hour. It will use about 50 cents of electricity and be better than 90% of the restaurant websites out there.
Yesterday I needed a tool for a specific task. In 1 hour I created a working tool with antigravity. Is it something I'd publish or sell or offer to others? Probably not, but I found that quite impressive. It is an extension of the idea "now I can program and create everything."
Besides I remember what kind of code we had when we first started coding with AI and now with all these coding agents etc. it has become quite impressive.
However, at the end it is still a tool and needs to be used accordingly.
You can create anything you need as long as what you need is a disposable script, a scotch-taped together single page app, or a complex problem and you have thousands of dollars to throw at tokens.
I've been playing with local models for some time, and I've been pleasantly surprised of late. A meager rtx 5080 with 16gb can give pretty good results now. The ecosystem is also improving pretty quickly.
I have a feeling at some point we will have a "Windows 95" moment (when computing really became personal for the masses) in AI, and things will significantly change shape again.
Hey, at least vi vs emacs is fun, no matter how crappy as an editor Emacs is, no one is getting paid by privacy invasive corporate, to sway people, to cash out more money
In between some old family history and recent US politics, I think there are other things I'd rather reserve the weighty word "cult" for... However, should it arise, one of the features that might convince me is this:
In the name of "loyalty" or "faith" cults require members to burn their bridges, actively cutting off their own potential to escape to any other social support network. That may mean creating enemies out of former associates, humiliating rituals or blackmail material, or simply ruining their reputations through obvious lies.
The utility of current-era AI may well be overstated, and the business models of certain companies might be doubtful, but that doesn't make AI it a "cult".
It's also a huge ad for microprocessors. I mean don't people realize we did math before microprocessors? We even had mechanical machines which were much more elegant than these electronic abominations. /s
Maybe a random employees mildly interesting blog post is not exactly where we need to look for a psyop. But that is probably what they want you to think.
I'm not saying AI isn't a good tool. However the less you understand what you're using and what you're doing the further you stand to geopardize the business you're working for.
> free, unlimited iteration, unbothered when I changed my mind
I think they mean that specifically in working with 3rd party per-project / freelance designers you usually get a "first draft + one adjustment" price, then every modification costs more. Similar for small design shops. Prices aren't necessarily per hour, as you'd get from developers.
50 iterations of CSS / layout? Easy, not even pushing it. A freelancer will cuss you out after 3-4 rounds of re-doing everything, but an LLM is happy to keep generating.
The benefit here is designers learning to code. It was always weird to me that designers were shaping software without knowing how it was built. I'm a designer btw.
However, designing in code is technology-first. One could argue that the purpose of design - to shape the artifacts for human purpose - is better done NOT starting with the strict rules of code. Pen and paper is still hard to beat, not for anything that looks nice, but for helping your mind forward.
Certainly, it helps the architect to know the constraints of concrete and bricks, but you don’t really have to know every level of down to the chemistry.
I’ve been using Claude Design for my front ends. The output looks and feels good enough, but the designs often look very similar and generally adhere to contemporary web tropes.
Keen to hear if anyone has had unconventional creative adventures with it.
Most applications are years long projects that already have styling and code structure done by humans so if the mock up is not perfectly aligned it doesn't even matter because when you finally approve the design and get to building it; that's where the pixel perfect tweaking comes in. And again if it's a mature code base with an existing stylesheet it tends to come out with all the expectations you'd have of it because it's re-using components that already exist 99% of the time.
I've had that experience and I've started testing different prompts and inputs.
I find it funny about meeting requirements when you give them, and making safe choices when you don't give direction. So if you're going to rate the output aesthetics and UX/content, but you don't prompt especially much around the aesthetics, you're only getting the safe assumed defaults. It's good at making bootstrap/tailwind clone designs unless you work that angle. For simple web pages, I've started making this the only focus for initial iteration.
I’ve done a few different things, some different presentations, web mockups etc. They all look more or less identical out of the box, and in every single case of a presentation, Claude Design has zero concept of layout boundaries, happily making slides that expand 200% or more out of the visible viewport (I have a lot of content and it just can’t figure out nice ways of presenting it).
There’s still value in it for me, I get decent enough output to convey my intent and I sometimes manually tweak the HTML.
Defining fonts is a good way to at least not get the same typography as everyone else.
I use claude design too. It has been suggested to me by some very respected and experienced designers which now prototype almost exclusively in claude and then when happy refine in figma.
By the way, you can always tune your prompts to not be generic, if you ask for generic UIs without detailed styling prompts you will get generic designs.
Sort of, I've given it examples of unconventional UI's and then had it sort of create a mashup of them and it's been decent. But I feel like that's cheating.
I use the same approach a lot. Before AI I also did this manually. First sit down with a user and just paper and pencil, then hack together a frontend POC / demo, have them play with it and adjust until it works as they wanted.
For me building a quick (not production quality) frontend demo in code was already often faster than getting the right interaction working in Figma. And it allowed to make it fully interactive so you can catch much more edge cases on the UX side.
Now with Claude Code it's even faster to build the throw away prototype. But not a huge difference since discussing with the users and thinking about how it should work is 80% of the time. Claude maybe halves the other 20% compared to quickly doing it yourself. Faster to first version, slower to iterate if it didn't fully get it.
We are doing this on my team (I am the frontend engineer) and honestly I really miss the old way of doing things.
Written specifications are being reduced in favor of these working prototypes, and now there’s this extra cognitive burden of reading the code and trying to determine what were the intended changes, and what’s the slop that needs to be tossed aside.
We also have to figure out, should we take over this generated PR and make any needed changes? Or do we start over from scratch? There’s often a sense of friction either way.
There have been times where a bunch of unintended changes were generated and I took time to port them over on my reimplementation, and then later on it’s “oops! Sorry! We didn’t mean to change that.”
I get it’s empowering but it does take away from some of the joy I used to find in my work and replaced it with some headaches.
I am in a similar boat. Design and product use Claude to vibe design/code a feature/experience and rapidly prototype it, getting it in front of customers to get feedback with minimal engineering time. Amazing! However, you might be surprised to find out it hasn’t really helped us ship stuff faster overall.
The reason, I believe, is we’ve lost something along the way. Thinking. A non-trivial amount of which is now outsourced to a language model. It paints over the cracks in the prompt, hallucinates how things should work when unspecified, what would normally make someone stop and say “this isn’t quite working”, “how should I communicate this idea” or “what happens when…” has gone. Now, these details are left for after it’s been built proper.
Sure, we can improve the process and reflect more on how to utilise this new technique … but is it better than how things used to be? Yeah nah
My experience has been there’s more “what did it make and does it work?” overhead. It’s like a junior developer throwing stuff over the wall and I’m responsible for seeing if it sticks.
I did this before AI though. I'd sometimes just build mockups directly in React using real components because it was quick and easy to put a form together with existing UI elements. I remember one project where the whole team was waiting on designs and I just came into stand up and was like "I built this yesterday. Is this what we wanted?"
It's much harder to RL out design taste because it's not self-grounding, and human labelers have no real skin in the game, so this (having a human with a vested outcome in the process directing a model's work) is the best way to get LLMs better at design/"taste"/aesthetic judgment themselves. We were working on the same thing 7 months ago and then I realized that winning over designers to do this would be a huge uphill battle setting up an inevitable fall from grace later on.
What makes me most suspicious of Claude Design is that when you disconnect and reconnect later, it loses context and nags you that the product doesn't work like that. Bullshit. It's at best an anti-abuse/implementation detail (to keep you from launching 10 at once and coming back to them later) or product shortcoming that just so happens to be optimized for keeping you from continuing your design in better tools than theirs for the inevitable followups.
It's great for one shots and it makes sense when you're trying to build a vertical product development stack like Anthropic but I'm disappointed it feels more like a tool optimized for keeping you in their product than for what you're working on. If a company other than Anthropic had shipped this - it's not that hard to build a visual self-eval loop, just use Chrome Devtools Protocol to run headless chrome and take screenshots -> feed into a judge LLM for feedback -> continue - I don't think it would really have seen much adoption.
That said, AI trained on Actor-Critic with a tight human feedback loop definitely seems like the right approach to solving the problem, just not something I want to spend my time training for someone else unless I can do so with higher "entropy" ie high parallelism/optionality
Even for the large products, figma is not the starting point for new concepts anymore. I start with a quick prototype on dev environment and then share it with designer for further improvements in figma or in the app itself. With every new model or agent improvement, going back to figma for polishing the ui is decllinig.
If we can find a way to keep the frontend code static templates without complex logic, need for this polishing in figma will go away completey as llms can understand it one context window. With the modern frameworks designed for client side rendering, keeping everything in one context is still tricky.
I use LLMs to design things and build wireframes. Product people can actually play with the wireframes and it's trivial to implement changes.
And same LLM generated files can be used to guide the LLM to build the actual pages too.
Sounds like desk strat RAD work is moving to LLM gen code at JS. My recentish experience of that kind of work has been Athena at JPMC and Quartz at BoA; both Python with functional style via DAG or pixie with py ui framework to match. Which enables quick dev of the parts of trading workflow that don't need to be quick, like booking tools or EOD risk. I know first hand Athena and Qz are crufty when you get into the weeds. The bonsai framework with Elm inspired ocaml impl sounds v cool. So I can see how this approach can accelerate a lot of trading tech dev. But does it have any traction over the hard problems where we turn to C++ or Rust: near real pricing and risk across multiple instruments and markets?
I worry about Figma stock, I know some who bought during the IPO who are now underwater. Figma launched their own design agent but not sure how well that's doing.
It’s not that I want them to suffer, it’s that them being rewarded incentivizes suboptimal behavior and outcomes. Markets don’t work anymore. Which results in even more suffering.
Figma make and gpt designer have a bunch of catching up to do. I couldn’t even import our brand guidelines into make which is already a .fig like what are we even doing here, guys? CD crunches through ungodly amount of tokens and is really slow on iteration but at least you can get some really nice prototypes extremely quickly there. GPT beats any Anthropic models on illustrations so they really should get a grip on multimodal. Overall, it seems like we’re still super early but you can already see glimpses of what may come
Even if this is ad only - are we really ready for a rug pull from Anthropic?
Maybe I’m completely unaware in this tools space, but I feel like it’s last tool that’s worth (and not pricey)…
Honestly, I'm not really pro or anti llm and I think there are a ton of limitations for using it to generate code, but UI has been probably the only thing I've been able to vibe code. It helps that I've done a lot of UI work over the years, but I think the combination of defects being easily visible through normal usage, the UI being a non-critical component of a system (bugs don't cause vulns or data corruption (usually), combined with the amount of churn that UI's see, make it a somewhat uniquely good candidate for vibe coding. Also a lot of UI toolkits are declarative, and I think language models do much better with declarative code.
In a way it's not much different from copy-pasting components from templates or whatever, just with more customisability. And for stuff that isn't HTML-based like React it does worse. It's also not great at building component libraries, I still write those myself with little LLM involvement, but that makes sense because the architecture is actually relevant with that, unlike generating CSS and xml-derived components, which is mostly just declarative templating anyways.
I've had decent success writing the core logic myself and then delegating the UI to AI. I think if I didn't write the core logic it would not work very well, but since it's designed well by myself the AI has a much smaller scope to work in which constrains it enough where vibe coding works. Pretty cool.
> In July 2025, the Securities and Exchange Board of India (SEBI) alleged that Jane Street used multiple entities for market manipulation and barred it from accessing the market.
https://www.bloomberg.com/opinion/newsletters/2025-07-07/jan... (https://archive.fo/0PQ2E)
I think the designer here takes a wrong approach and sort of falls into engineer envy where he wants to make prototypes as deep and realistic as possible. And that is not really the most important part of the design job.
The most important thing is that the right thing gets build (why do we need a JSQL input box? what do we actually want? what are other ways to do this?). And this is often better done with pen and paper sketches, meetings, observation, discussions ... rather than too quickly narrowing on a particular design and spinning into discussions on whether buttons should be on the left or on the right, LLM intricacies etc.
Give it some time. We’ll figure out what LLMs are good and bad at. I think vibe engineering will eventually go up on the wall next to static vs dynamic typing and vi vs eMacs.
At least, that assumes AI models won’t keep improving by leaps and bounds. We’re in a transition period. It’s gonna be chaos.
But a lot of times they don't. Devs building websites for small businesses used to be a thing, and it was almost universally a crap product. Practically every restaurant in my small town has been able to take control of their own website with AI, and it's a better experience for everyone.
Rapid digitisation meant a lot of low-value work wound up being done by high-value people. The economy is pivoting away from that configuration. The high-value folks getting displaced are pissed, partly rightly, partly out of spite. The folks who had to pay those bills are ecstatic, mostly overexuberantly, but in part correctly.
Your comment is valid, but not for the reason you think. What you should be talking about is the grunt work done in our field to non-trivial applications. Adding a search box to a table, or add an extra field to a form, etc.
So you think about a ticket that often might take a few hours, but in badly architected system might take a week, add a field to a class, edit the DB structure (maybe manually, maybe through via an ORM generated migration), add it to DTOs, add a validator, add it to the FE definitions, edit the page layout, etc..
Low value work that until now had to be performed by high-value employees.
The current trend is a proliferation of internal tooling. I've honestly found quite a bit of it useful. The rest is the usual should-have-been-a-Google-form nonsense.
Besides I remember what kind of code we had when we first started coding with AI and now with all these coding agents etc. it has become quite impressive.
However, at the end it is still a tool and needs to be used accordingly.
I have a feeling at some point we will have a "Windows 95" moment (when computing really became personal for the masses) in AI, and things will significantly change shape again.
In the name of "loyalty" or "faith" cults require members to burn their bridges, actively cutting off their own potential to escape to any other social support network. That may mean creating enemies out of former associates, humiliating rituals or blackmail material, or simply ruining their reputations through obvious lies.
Unless you invade a country again and we have to deal with the Fallout for years to come, that is.
For example: I have banned all news about Trump administration from my home.
It's simply not interesting. He is a toddler, and I have a toddler of my own.
It's boring and it's viral and if you refuse to participate, there are still people annoying you with their beliefs.
It's a fucking tool, like a fucking shovel.
I myself, I pump my investments at least twice a day. Once in the morning, right after I work out. And then once right after lunch.
I want to, that's not why I do it. I do it cause I fuckin' need to.
I'm not saying AI isn't a good tool. However the less you understand what you're using and what you're doing the further you stand to geopardize the business you're working for.
Do you not pay for Claude?
I think they mean that specifically in working with 3rd party per-project / freelance designers you usually get a "first draft + one adjustment" price, then every modification costs more. Similar for small design shops. Prices aren't necessarily per hour, as you'd get from developers.
However, designing in code is technology-first. One could argue that the purpose of design - to shape the artifacts for human purpose - is better done NOT starting with the strict rules of code. Pen and paper is still hard to beat, not for anything that looks nice, but for helping your mind forward.
It helps to understand the constraints of a medium, but you really don't need to know every level down to the electrons moving through the silicon.
Keen to hear if anyone has had unconventional creative adventures with it.
I find it funny about meeting requirements when you give them, and making safe choices when you don't give direction. So if you're going to rate the output aesthetics and UX/content, but you don't prompt especially much around the aesthetics, you're only getting the safe assumed defaults. It's good at making bootstrap/tailwind clone designs unless you work that angle. For simple web pages, I've started making this the only focus for initial iteration.
There’s still value in it for me, I get decent enough output to convey my intent and I sometimes manually tweak the HTML.
Defining fonts is a good way to at least not get the same typography as everyone else.
By the way, you can always tune your prompts to not be generic, if you ask for generic UIs without detailed styling prompts you will get generic designs.
For me building a quick (not production quality) frontend demo in code was already often faster than getting the right interaction working in Figma. And it allowed to make it fully interactive so you can catch much more edge cases on the UX side.
Now with Claude Code it's even faster to build the throw away prototype. But not a huge difference since discussing with the users and thinking about how it should work is 80% of the time. Claude maybe halves the other 20% compared to quickly doing it yourself. Faster to first version, slower to iterate if it didn't fully get it.
Written specifications are being reduced in favor of these working prototypes, and now there’s this extra cognitive burden of reading the code and trying to determine what were the intended changes, and what’s the slop that needs to be tossed aside.
We also have to figure out, should we take over this generated PR and make any needed changes? Or do we start over from scratch? There’s often a sense of friction either way.
There have been times where a bunch of unintended changes were generated and I took time to port them over on my reimplementation, and then later on it’s “oops! Sorry! We didn’t mean to change that.”
I get it’s empowering but it does take away from some of the joy I used to find in my work and replaced it with some headaches.
Do you look at generated assembly that comes out of your compiler? You don't. So why are you looking at this code?
We have pushed up the abstraction layer.
The reason, I believe, is we’ve lost something along the way. Thinking. A non-trivial amount of which is now outsourced to a language model. It paints over the cracks in the prompt, hallucinates how things should work when unspecified, what would normally make someone stop and say “this isn’t quite working”, “how should I communicate this idea” or “what happens when…” has gone. Now, these details are left for after it’s been built proper.
Sure, we can improve the process and reflect more on how to utilise this new technique … but is it better than how things used to be? Yeah nah
no thanks. BE boys do FE now
It's much harder to RL out design taste because it's not self-grounding, and human labelers have no real skin in the game, so this (having a human with a vested outcome in the process directing a model's work) is the best way to get LLMs better at design/"taste"/aesthetic judgment themselves. We were working on the same thing 7 months ago and then I realized that winning over designers to do this would be a huge uphill battle setting up an inevitable fall from grace later on.
What makes me most suspicious of Claude Design is that when you disconnect and reconnect later, it loses context and nags you that the product doesn't work like that. Bullshit. It's at best an anti-abuse/implementation detail (to keep you from launching 10 at once and coming back to them later) or product shortcoming that just so happens to be optimized for keeping you from continuing your design in better tools than theirs for the inevitable followups.
It's great for one shots and it makes sense when you're trying to build a vertical product development stack like Anthropic but I'm disappointed it feels more like a tool optimized for keeping you in their product than for what you're working on. If a company other than Anthropic had shipped this - it's not that hard to build a visual self-eval loop, just use Chrome Devtools Protocol to run headless chrome and take screenshots -> feed into a judge LLM for feedback -> continue - I don't think it would really have seen much adoption.
That said, AI trained on Actor-Critic with a tight human feedback loop definitely seems like the right approach to solving the problem, just not something I want to spend my time training for someone else unless I can do so with higher "entropy" ie high parallelism/optionality
Yeah you don't say. High of $124, currently $22. But hey: that's ~~gambling~~ stock trading for you.
https://au.finance.yahoo.com/quote/FIG/
That's pretty mean. I disagree with a lot of people on their investment choices and evangelism. That doesn't mean I want them to suffer for it.
What if they're choosing correctly? Like, I've been pretty consistently critical of crypto. Yet I'm glad my index funds have to buy Coinbase.
Klankers will fix everything. Right?
When they paywall hard, I’ll use a local model on my laptop processor instead,
or phone,
and wait a few hours as needed.
from 6 sessions and 5 projects only one template that I choose anything else is really really bad
This is a very disappointing post from Jane Street.
In a way it's not much different from copy-pasting components from templates or whatever, just with more customisability. And for stuff that isn't HTML-based like React it does worse. It's also not great at building component libraries, I still write those myself with little LLM involvement, but that makes sense because the architecture is actually relevant with that, unlike generating CSS and xml-derived components, which is mostly just declarative templating anyways.
I've had decent success writing the core logic myself and then delegating the UI to AI. I think if I didn't write the core logic it would not work very well, but since it's designed well by myself the AI has a much smaller scope to work in which constrains it enough where vibe coding works. Pretty cool.