> My hunch: vibe coding is a lot like stock-picking – everyone’s always blabbing about their big wins. Ask what their annual rate of return is above the S&P, and it’s a quieter conversation
LLMs are architected to aim toward the center of the bell curve of their training data. You shouldn't expect them to produce innovative ideas, but the upside is they also won't produce terrible ones.
The same applies to design. Most of the time, you get something that "doesn't suck", which is perfectly fine for projects where using a designer isn't worth it, like internal corporate pages. But consumer-facing pages require nuance to understand the client and their branding (which clients often struggle to articulate themselves), and that's not something current models can capture.
Internal corporate tooling can be greatly improved by promoting scheduled "water cooler chat on steroids" calls between engineers/designers/product/tpms/managers and the actual users.
What's nice is that these sessions bleed into everything. You don't need to look through users' eyes that many times to find great improvement in UX sensibilities.
Since scheduling is the biggest pain point here, I just built a scheduler and a signup form at work. Everyone who gets a session walks away with positive feedback, the company as a whole becomes more interconnected, and now I'm working to get more and more folk on board.
My goal is to unleash a whole flotilla of white collar workers who understand the value of talking to the users such that they too push for lightweight, no action item, scheduled sessions which becomes standard practice as part of our careers and we all end up with better software as a whole.
I think it can be a false comfort to think of LLMs being trained to the center of the bell curve. I think it's closer to true that there's no real "average" (just like there isn't an "average" human) because there's just too many dimensions.
But what LLMs do, in the absence of better instructions, is expect that the user WANTS the most middling innocuous output. Which is very reasonable! It's not a lack of capability; it's a strong capability to fill in the gaps in its instructions.
The person who has a good intuition for design (visual, narrative, conversational) but can't articulate that as instructions will find themselves stuck. And unsurprisingly this is common, because having that vision and being able to communicate that vision to an LLM is not a well practiced skill. Instructing an LLM is a little like instructing a person... but only a little. You have to learn it. And I don't think as LLMs get better that this will magically fix itself, because it's not exactly an error, there's no "right" answer.
Which is to say: I think applying design to one's work with AI is possible, important, and seldom done.
I've had great success using Claude to produce a new landing page which is much more stylish than something I would produce myself. It's also nowhere near the standard expected from a professional designer, but for a FOSS app, that's just fine with me :)
Yeah, my experience is that when they do work they'll get the job done but the design is going to be unmaintainable garbage unless I force my own design on it.
I think the big issue with design is the ux to LLMs at the moment: it’s really hard to iterate on a design, see the output, make changes etc. I’ve had terrible luck getting good design from ChatGPT/Codex, but V0 is probably one of the most impressive AI UX experiences I’ve encountered — I often show it to non-technical friends who are ai skeptical.
LLMs don't produce ideas they implement them, anyone who is relying on LLM produced ideas basically doesn't matter the whole point is they make people who want to design interesting stuff have a lower barrier to entry of actually producing stuff.
Tangent: Google keeps doing the same with GCP/Firebase/etc. Every year they launch a bunch of really well-crafted services that make it easy to build the most average kind of product in whatever area is trendy that year. Every year I am left intrigued but needing more to actually make use of them. The next year it's something else.
I guess this is primarily a business pattern. Or anti-pattern?
Assuming you mean cloud platforms in general, I don't even think it's that tangential. In fact it may cut to the heart of the matter: if React-over-REST-over-SQL-plus-some-background-jobs was all we needed, cloud platform innovation would've stopped at Heroku and Rails 20 years ago, and AI could probably make a run on replacing SWE jobs entirely.
But as it's played out, there are a ton of use cases that don't fit neatly into that model, and each year the cloud platforms offer new tools that fit some of those use cases. AI will probably be able to string together existing tools as IaaS providers offer them, perhaps without even involving an engineer, but for use cases that are still outside cloud platform offerings, seem like things that require some ingenuity and creativity that AI wouldn't be able to grok.
From time to time I look at the explore page of various AI design tools, and they are as corporate-depressing as I didn't expect them to be [1]. It's not even a bell curve. It feels like they are overindexed on bland corporate aesthetic. Getting them to output anything but a Linear clone is an exercise in frustration.
To me, the key quote is the simple "If you had told me in late 2022 I’d be saying these things 3 years later, I would’ve been pretty surprised." As someone with little exposure to the design industry, seeing how quickly AI could generate images, I'd been under the assumption that the AI takeover was already well underway there, so was surprised to learn that it's not.
If anything, that gives some comfort around the future of engineering job prospects. While there's still room to worry, "yeah but design is fundamentally human, while engineering is mostly technical and can be automated", I'm sure, just as design has realized, that when we get to a point where AI should be taking over, we'll realize that there's a lot of non-technical things that engineers do, that AI cannot replace.
Basically, if replacing a workforce is the goal, AI image generators and code generators look like replacement technologies from afar, but when you look closer you realize they're "the right solution to the wrong problem", to be a true replacement tech, and in fact don't really move the needle. And maybe AI, by definition of being artificial and intelligence (as opposed to real common sense) as a whole, is fundamentally an approach that "solves the wrong problem" as a replacement tech, even as AGI or even ASI gets created.
That quote stood out to me as well, but mostly because the 3 images shown by the author have nothing to do with product/interface/communications design.
I guess they’re vaguely cool looking images? If the author had used them to talk about how “concept art” in games/movies was going to get upended by AI there would be a point there, but as it stands I find it very puzzling that someone who claims to teach design would use them as key examples of why design - a human process of coming up with specific solutions to fuzzy problems with arbitrary constraints - was headed in any particularly direction.
I think there's some benefit of hindsight in that perspective though. I can imagine how, at the time, you see the advancement, and it's not obvious what the barriers for AI takeover are. Similar to software now, plenty of SWEs have a nagging feeling about AI encroachment. But in all likelihood, eventually it'll become clear that most SWE work involves coordinating with other teams, planning incremental delivery and various testing and review phases, working with CS when users face issues, etc. The boundaries will be a lot clearer, and looking back at the current FUD because of a better autocomplete, it'll seem ridiculous by then. (At least I hope so!)
The number one thing that AI is lacking is taste, precisely the reason we need designers in the first place. An engineer (most engineers are not designers) alone, or an LLM is not thinking about good design principles and doing the things needed to develop good taste.
It's the reason why LLMs are horrible at writing, and the reason why good design is really hard to get out of an LLM. Figma Make and Claude Code are really just using the out of the box CSS from shadcn that's why everything looks the same.
> After 2.5 years of insane hype, there’s no evidence that current AI is making the design process faster
I am designing prototypes faster today with LLMs this is just flat out wrong. And it's not really 2.5 years it's more like the last 6 months, GPT5, sonnet 4.0 and 4.5 have made this stuff viable to seriously use.
> I am designing prototypes faster today with LLMs this is just flat out wrong
One thing I've wondered and not been sure of, is that you can see productivity boost in parts of the process, but overall the end to end process doesn't seem to be getting done faster.
I'm not sure of this, but it's what I've observed at work.
You might be designing a lot of prototypes faster, but are you landing on something good quicker? Are you getting the final product out faster?
> You might be designing a lot of prototypes faster, but are you landing on something good quicker? Are you getting the final product out faster?
Yeah, but not because of the LLM, more because I have a strong opinion and multiple design options I want to get in front of customers asap. Instead of coding a UI myself which I can do, or working with an engineer, I can try a bunch of stuff, show it to customers early and see what resonates.
And this speaks to the whole AI hype and tendency to shove it into everything. It’s just a tool. When we do build stuff that uses AI it has to be because we find a problem worth solving and the solution characteristics happen to align with LLM’s, not because we actively want to use the tech and shoehorn it in.
Perhaps it is finally true this time, but the AI hype machine has been making this very same claim for years now.
You'll have to understand if we insist on tangible results before buying that Kool-Aid in bulk.
I wonder how much of the stuff that actually takes up the time of the people we call "designers" to do their jobs is something our current crop of LLMs is good for. If it's 90% then LLMs could make you a 10x designer. If it's 10%, LLMs could make you a 1.11x designer (minus the time it takes to fiddle around with the LLM, of course).
Basically replacing showing a UI made in figma with a UI made with claude code or figma make, to show a customer see what they think and then throw away, not to put in production ever.
I think it raises the bar on what’s needed to be a designer, but I don’t think it replaces the need for a designer.
I’d draw parallels to web design. Yes some frameworks made it incredibly easy for anyone to whip up a decent looking professional site. But then 95% of websites were the same boring single page scroll nonsense with some fancy CSS. It checks a box but isn’t notable. If you hire a designer to do something truly original then you can stand out.
It will be the same with AI where “good enough to check the box” becomes easy, but going beyond that still requires skill and experience.
Design moves at the speed of culture; not technology. It took 3 years of people messing with mobile phones before it occurred to someone to implement "pull down to refresh" and much longer for it to be common practice that people just expect from UX. I think people are still learning what they want from an AI experience.
I do think you have to be pretty targeted with your predictions, though. Consumer product design seems to be evolving differently from B2B and at a different pace. Growth curves are different for each.
One of the bigger design battles at a prior company was designers insisting on pull to refresh, and the researchers insisting on removing it due to customer feedback.
Lots of these discussions are simplifying design to 'making things look pretty'. That's just not true for even the more visual-based design disciplines like graphic design. And the 'regular' product design (ux/ui/ixd) happening in most tech companies has very little of this compared to the rest of the scope of what a designer really does.
Product design isn't a layer that you apply. It's not an output of some prompt. It's a difficult-to-define process of crafting the interface between the user and product's functionality.
I’m a product designer and this is what I’ve noticed:
- When I use AI to vibe-code, it gives me a usable result but I personally have no idea if the output is “correct”. I am not sure if there are security vulnerabilities, I don’t know what is actually happening under the hood to make it work, etc.
- When my engineering friends use AI to vibe-design, I notice the same pattern. It looks “designed” but there are obvious usability issues, pattern mismatches for common user goals, and the UI lacks an overall sense of polish.
Basically, my takeaway is that AI is great for spinning things up quickly but it is not a replacement for fundamentals or craft.
I think it's trivially true that (at present moment) AI can deliver an average result, which makes it useful in domains you are below average but not useful for domains in which you are good at.
Design is one of those things that succeeds or fails in subtlety and both are difficult to quantify and back propagate through any sort of process, let alone training a model. The same way we figured out that the microwave can make approximations to good food quickly, so too shall we see that AI can do the same with tasks that rely heavily on a connection to people's aesthetics.
I'd add a #3 to the "explainable with" list: the trend towards relentlessly outsourcing work, often in situations where you get 1/2 the quality on 2x the timeline, because it's still cheaper than good engineers and the ostensible relentless time pressure turns out to not exist if offshore workers are cheap enough.
I’ve designed interfaces that have stood the test of time and the criticality of their use. I’ve created before there were design systems/tokens, the bootstraps, and the tailwinds. They are used spanning decades in clinics, hospitals, publishing houses, banking, and by many other MNCs to deploy interesting products.
The current AI and design is in a honeymoon period, focused more on experimentation and functional components, using working design patterns, philosophies, templates, and components in abundance. I trust that it is indeed a good thing to accept “Good Enough Design” and layer in the better and best ones later. Right now, we have good enough designs everywhere.
"After 2.5 years of insane hype, there’s no evidence that current AI is making the design process faster"
This is antidotally untrue. I work for a small startup. We don't have the money / aren't willing to pay for a full-time designer. So, let's just say, our UI design has always been pretty terrible. With AI, using claude, to generate design and HTML / CSS based on requirements, the design that has been generated has been heads and tails better than anything we ever came up with alone.
> After 2.5 years of insane hype, there’s no evidence that current AI is making the design process faster
This does not match my experience. I have been using Claude for speeding up design. Describe your page and ask it to use Tailwind and it will come up with some interesting designs. You still need a designer because some designs it comes up with are over the top, and need to be moderated.
Most designers I've worked with don't want Tailwind slop as a starting point.
At most AI prototypes and images serve the role that a whiteboard drawing or wireframe did before: that's a win, but it's not an monumental change in efficiency.
I ironically think AI is already there in terms of being capable of more, but no one has built the right harnesses for AI.
I've heard some anecdotal evidence that designers are starting to use AI tools to help them build working interactive prototypes of their designs, but I don't have enough day-to-day contact with professional design teams to have a good feel for how widespread that is and what impact it's having on the process.
I thought a huge part of the perceived value of ChatGPT etc. is the ability to bypass the work of designers. The whole dream of AI is not having to deal with what someone else thought the solution to your problem was, and instead just skipping to the end.
SHAPIRO: OK, so you've spent your career creating television without AI, and I could imagine today you thinking, boy, I wish I had had that tool to solve those thorny problems...
SIMON: What?
SHAPIRO: ...Or saying...
SIMON: You imagine that?
SHAPIRO: ...Boy, if that had existed, it would have screwed me over.
SIMON: I don't think AI can remotely challenge what writers do at a fundamentally creative level.
SHAPIRO: But if you're trying to transition from scene five to scene six, and you're stuck with that transition, you could imagine plugging that portion of the script into an AI and say, give me 10 ideas for how to transition this.
In some sense it is happening. But as alluded to in the article its going to be from the bottom up. I use AI designs for my slides and videos. Are they as good as a professional designer? No. Not even close really. Is it better than me? Yes. Not even close really.
The thing is on a computer today is that most design is prepackaged and not customized to what I'm doing. AI gets me a big step closer to that so much so that it is preferable to most of the prepackaged work done by a professional designer. Again, if I had a professional designer next to me to do slides and videos then that would be better, but very few people have that.
I totally agree with this post. The AI only helped the scammers who sell courses, claiming they made so much money from their crappy mobile apps or their other types of web apps. It is just a social media hype. Can a newbie without any understanding of database structures or modular design create a perfect flow? With the current technology, it is not possible. People are so impatient to understand what it takes to make a functioning app, to create a structured app flow, so the vibe coding project gets messy in a few prompts. It is just a hype, at least for coding, and people will soon realize it.
Suno and image generators are totally different and riding the AI wave at the moment. YouTube turned into crap, and Instagram content as well. AI fatigue is here.
Because art, commercially viable art, is 1% inspiration and 99% perspirarion. AI can generate the first inspirational image but cannot do the 99%, all the hard work, to turn that image into a viable product.
I was in the grocery store the other day, needed to use the restroom, and saw an AI-generated picture, framed, hanging on the wall in the restroom stall. It was even "signed" by "Meta AI".
Well, it is still much more efficient (consumed energy, outcome, and mental sanity-wise) to fill a form with a textbox and a datepicker to book a flight, than trying to negotiate it with a chatbot.
Everyone assumes that the perfect communication medium is human (or human like) conversation. If that was the case, couple and marriage therapists wouldn't exist.
Developers who use AI think they're quicker and better, but they're actually slower and worse. Not surprised to hear it's similar in other fields. AI in general, just like drugs, makes people feel good but without any actual substance behind the feeling.
(If feeling without substance is all you need, then it's okay to use AI. AI Dungeon, for example, was pretty cool. Or slide backgrounds that would have otherwise been solid colours because they're worth $0 and you wouldn't have paid a designer.)
This story is not about developers, but setting that aside: The reason it's not damning is that the results can't be generalized. It's mostly self-reported "minutes per issue" anecdata by 16 experienced OSS maintainers working on their own, mature repos, most of whom were new to Cursor and did not a chance to develop or adopt frameworks for AI-assisted development.
That has not been my experience at all. Whenever I tried asking the AI to do something, it took an inordinate amount of time and thought to go through its changes.
The mistakes were subtle but critical. Like copying a mutex by value. Now, if I would be writing the code, I would not make the mistake. But when reviewing, it almost slipped by.
And that's where the issue is: you have to review the code as if it's written by a clueless junior dev. So, building up the mental model when reviewing, going through all the cases and thinking of the mistakes that could possibly have happened... sorry, no help.
Maybe 10% of typing it out but when I think about it, it's taking more time because I have to create the mental model in my mind then create the mental model out of the thing that AI typed out and then have to compare the two. This latter is much more time consuming than creating the model in my mind and typing it out.
I think that programming languages (well, at least some of them, maybe not all) have succeeded in being good ways of expressing programmatic thought. If I know what I want in (say) C, it can be faster for me to write C code than to write English that describes C code.
I guess it depends on what you use it for. I found it quite relaxing to get AI to write a bunch of unit tests for existing code. Simple and easy to review, and not fun to write myself.
> Developers who use AI think they're quicker and better, but they're actually slower and worse.
You responded that this is a "gross overgeneralization of the content of the actual study", but the study appears to back up the original statement. To quote the summary:
> When developers are allowed to use AI tools, they take 19% longer to complete issues—a significant slowdown that goes against developer beliefs and expert forecasts. This gap between perception and reality is striking: developers expected AI to speed them up by 24%, and even after experiencing the slowdown, they still believed AI had sped them up by 20%.
(I realise newer models have been released since the study, but you didn't claim that the findings have been superceded.)
Sure! The study focused on experienced devs working in complex codebases they already knew thoroughly. This was and is the worst case for using AI tooling from a cold start, _particularly_ AI tooling as it existed at the time.
There were also only 16 developers involved in the study.
Time has passed since the study and we've had an entirely new class of tool introduced (the agentic CLI a la Claude Code) as well as two subsequent generations of model improvement (Sonnet 3.7 to Sonnet 4 to Sonnet 4.5). Given that the results of the METR study were stated as an eternal, unqualified truth, the fact that tooling and models are much superior now compared to when the study was conducted is worth noting as well.
Which would be non-news if the developers also thought it wasn't going to be helpful because they already knew their codebases thoroughly. Or at least if they did the task, and then reported that AI made it harder. But in reality, they expected it to be faster, and then after doing it slower, said they'd done it faster. That's weird.
> You responded that this is a "gross overgeneralization of the content of the actual study", but the study appears to back up the original statement.
It doesn't, and the study authors themselves are pretty clear about the limitations. The irony is that current foundation models are pretty good at helping to identify why this study doesn't offer useful general insights into the productivity benefits (or lack of) of AI-assisted development.
Best summary I've heard so far
vibes coding is massively saturated with early model-access poasters who define the narrative on each release.
not complicated.
The same applies to design. Most of the time, you get something that "doesn't suck", which is perfectly fine for projects where using a designer isn't worth it, like internal corporate pages. But consumer-facing pages require nuance to understand the client and their branding (which clients often struggle to articulate themselves), and that's not something current models can capture.
What's nice is that these sessions bleed into everything. You don't need to look through users' eyes that many times to find great improvement in UX sensibilities.
Since scheduling is the biggest pain point here, I just built a scheduler and a signup form at work. Everyone who gets a session walks away with positive feedback, the company as a whole becomes more interconnected, and now I'm working to get more and more folk on board.
My goal is to unleash a whole flotilla of white collar workers who understand the value of talking to the users such that they too push for lightweight, no action item, scheduled sessions which becomes standard practice as part of our careers and we all end up with better software as a whole.
But what LLMs do, in the absence of better instructions, is expect that the user WANTS the most middling innocuous output. Which is very reasonable! It's not a lack of capability; it's a strong capability to fill in the gaps in its instructions.
The person who has a good intuition for design (visual, narrative, conversational) but can't articulate that as instructions will find themselves stuck. And unsurprisingly this is common, because having that vision and being able to communicate that vision to an LLM is not a well practiced skill. Instructing an LLM is a little like instructing a person... but only a little. You have to learn it. And I don't think as LLMs get better that this will magically fix itself, because it's not exactly an error, there's no "right" answer.
Which is to say: I think applying design to one's work with AI is possible, important, and seldom done.
I've had great success using Claude to produce a new landing page which is much more stylish than something I would produce myself. It's also nowhere near the standard expected from a professional designer, but for a FOSS app, that's just fine with me :)
Design by committee is known to produce terrible designs. The best an LLM can do is to completely copy a common decent design.
And ideas born without knowledge about their "implementation" are, by definition, quite low resolution.
I guess this is primarily a business pattern. Or anti-pattern?
But as it's played out, there are a ton of use cases that don't fit neatly into that model, and each year the cloud platforms offer new tools that fit some of those use cases. AI will probably be able to string together existing tools as IaaS providers offer them, perhaps without even involving an engineer, but for use cases that are still outside cloud platform offerings, seem like things that require some ingenuity and creativity that AI wouldn't be able to grok.
[1] https://x.com/dmitriid/status/1953750443248562431
If anything, that gives some comfort around the future of engineering job prospects. While there's still room to worry, "yeah but design is fundamentally human, while engineering is mostly technical and can be automated", I'm sure, just as design has realized, that when we get to a point where AI should be taking over, we'll realize that there's a lot of non-technical things that engineers do, that AI cannot replace.
Basically, if replacing a workforce is the goal, AI image generators and code generators look like replacement technologies from afar, but when you look closer you realize they're "the right solution to the wrong problem", to be a true replacement tech, and in fact don't really move the needle. And maybe AI, by definition of being artificial and intelligence (as opposed to real common sense) as a whole, is fundamentally an approach that "solves the wrong problem" as a replacement tech, even as AGI or even ASI gets created.
I guess they’re vaguely cool looking images? If the author had used them to talk about how “concept art” in games/movies was going to get upended by AI there would be a point there, but as it stands I find it very puzzling that someone who claims to teach design would use them as key examples of why design - a human process of coming up with specific solutions to fuzzy problems with arbitrary constraints - was headed in any particularly direction.
It's the reason why LLMs are horrible at writing, and the reason why good design is really hard to get out of an LLM. Figma Make and Claude Code are really just using the out of the box CSS from shadcn that's why everything looks the same.
> After 2.5 years of insane hype, there’s no evidence that current AI is making the design process faster
I am designing prototypes faster today with LLMs this is just flat out wrong. And it's not really 2.5 years it's more like the last 6 months, GPT5, sonnet 4.0 and 4.5 have made this stuff viable to seriously use.
One thing I've wondered and not been sure of, is that you can see productivity boost in parts of the process, but overall the end to end process doesn't seem to be getting done faster.
I'm not sure of this, but it's what I've observed at work.
You might be designing a lot of prototypes faster, but are you landing on something good quicker? Are you getting the final product out faster?
Yeah, but not because of the LLM, more because I have a strong opinion and multiple design options I want to get in front of customers asap. Instead of coding a UI myself which I can do, or working with an engineer, I can try a bunch of stuff, show it to customers early and see what resonates.
Its a combination of having a strong opinion while taking into account what the customer says and does. Otherwise we’re just building faster horses (https://hbr.org/2011/08/henry-ford-never-said-the-fast)
And this speaks to the whole AI hype and tendency to shove it into everything. It’s just a tool. When we do build stuff that uses AI it has to be because we find a problem worth solving and the solution characteristics happen to align with LLM’s, not because we actively want to use the tech and shoehorn it in.
Perhaps it is finally true this time, but the AI hype machine has been making this very same claim for years now.
You'll have to understand if we insist on tangible results before buying that Kool-Aid in bulk.
I wonder how much of the stuff that actually takes up the time of the people we call "designers" to do their jobs is something our current crop of LLMs is good for. If it's 90% then LLMs could make you a 10x designer. If it's 10%, LLMs could make you a 1.11x designer (minus the time it takes to fiddle around with the LLM, of course).
I’d draw parallels to web design. Yes some frameworks made it incredibly easy for anyone to whip up a decent looking professional site. But then 95% of websites were the same boring single page scroll nonsense with some fancy CSS. It checks a box but isn’t notable. If you hire a designer to do something truly original then you can stand out.
It will be the same with AI where “good enough to check the box” becomes easy, but going beyond that still requires skill and experience.
I do think you have to be pretty targeted with your predictions, though. Consumer product design seems to be evolving differently from B2B and at a different pace. Growth curves are different for each.
Better to just show progress instead.
Back then people were similarly incredulous of the entire idea of the internet and apps.
Product design isn't a layer that you apply. It's not an output of some prompt. It's a difficult-to-define process of crafting the interface between the user and product's functionality.
- When I use AI to vibe-code, it gives me a usable result but I personally have no idea if the output is “correct”. I am not sure if there are security vulnerabilities, I don’t know what is actually happening under the hood to make it work, etc.
- When my engineering friends use AI to vibe-design, I notice the same pattern. It looks “designed” but there are obvious usability issues, pattern mismatches for common user goals, and the UI lacks an overall sense of polish.
Basically, my takeaway is that AI is great for spinning things up quickly but it is not a replacement for fundamentals or craft.
The current AI and design is in a honeymoon period, focused more on experimentation and functional components, using working design patterns, philosophies, templates, and components in abundance. I trust that it is indeed a good thing to accept “Good Enough Design” and layer in the better and best ones later. Right now, we have good enough designs everywhere.
This is antidotally untrue. I work for a small startup. We don't have the money / aren't willing to pay for a full-time designer. So, let's just say, our UI design has always been pretty terrible. With AI, using claude, to generate design and HTML / CSS based on requirements, the design that has been generated has been heads and tails better than anything we ever came up with alone.
This does not match my experience. I have been using Claude for speeding up design. Describe your page and ask it to use Tailwind and it will come up with some interesting designs. You still need a designer because some designs it comes up with are over the top, and need to be moderated.
At most AI prototypes and images serve the role that a whiteboard drawing or wireframe did before: that's a win, but it's not an monumental change in efficiency.
I ironically think AI is already there in terms of being capable of more, but no one has built the right harnesses for AI.
Loose tailwind classes and shadcn is not it.
Can’t fit all that in a million tokens.
> I am utterly disgusted. ... I strongly feel this is an insult to life itself.
https://www.youtube.com/watch?v=ngZ0K3lWKRc
or David Simon:
SHAPIRO: OK, so you've spent your career creating television without AI, and I could imagine today you thinking, boy, I wish I had had that tool to solve those thorny problems...
SIMON: What?
SHAPIRO: ...Or saying...
SIMON: You imagine that?
SHAPIRO: ...Boy, if that had existed, it would have screwed me over.
SIMON: I don't think AI can remotely challenge what writers do at a fundamentally creative level.
SHAPIRO: But if you're trying to transition from scene five to scene six, and you're stuck with that transition, you could imagine plugging that portion of the script into an AI and say, give me 10 ideas for how to transition this.
SIMON: I'd rather put a gun in my mouth.
https://www.npr.org/transcripts/1177569966
The thing is on a computer today is that most design is prepackaged and not customized to what I'm doing. AI gets me a big step closer to that so much so that it is preferable to most of the prepackaged work done by a professional designer. Again, if I had a professional designer next to me to do slides and videos then that would be better, but very few people have that.
Suno and image generators are totally different and riding the AI wave at the moment. YouTube turned into crap, and Instagram content as well. AI fatigue is here.
The interesting conversation is around art not as a product to be consumed but a medium of human expression with intent behind it.
Everyone assumes that the perfect communication medium is human (or human like) conversation. If that was the case, couple and marriage therapists wouldn't exist.
(If feeling without substance is all you need, then it's okay to use AI. AI Dungeon, for example, was pretty cool. Or slide backgrounds that would have otherwise been solid colours because they're worth $0 and you wouldn't have paid a designer.)
This first chart should be absolutely damning: https://metr.org/blog/2025-07-10-early-2025-ai-experienced-o...
I'd love to see a larger study on more experienced users. The mismatch between their perception and reality is really curious.
It's quite possible that it took less overall mental effort from the developers using AI, but it took more elapsed time.
The mistakes were subtle but critical. Like copying a mutex by value. Now, if I would be writing the code, I would not make the mistake. But when reviewing, it almost slipped by.
And that's where the issue is: you have to review the code as if it's written by a clueless junior dev. So, building up the mental model when reviewing, going through all the cases and thinking of the mistakes that could possibly have happened... sorry, no help.
Maybe 10% of typing it out but when I think about it, it's taking more time because I have to create the mental model in my mind then create the mental model out of the thing that AI typed out and then have to compare the two. This latter is much more time consuming than creating the model in my mind and typing it out.
OP stated that:
> Developers who use AI think they're quicker and better, but they're actually slower and worse.
You responded that this is a "gross overgeneralization of the content of the actual study", but the study appears to back up the original statement. To quote the summary:
> When developers are allowed to use AI tools, they take 19% longer to complete issues—a significant slowdown that goes against developer beliefs and expert forecasts. This gap between perception and reality is striking: developers expected AI to speed them up by 24%, and even after experiencing the slowdown, they still believed AI had sped them up by 20%.
(I realise newer models have been released since the study, but you didn't claim that the findings have been superceded.)
Sure! The study focused on experienced devs working in complex codebases they already knew thoroughly. This was and is the worst case for using AI tooling from a cold start, _particularly_ AI tooling as it existed at the time.
There were also only 16 developers involved in the study.
Time has passed since the study and we've had an entirely new class of tool introduced (the agentic CLI a la Claude Code) as well as two subsequent generations of model improvement (Sonnet 3.7 to Sonnet 4 to Sonnet 4.5). Given that the results of the METR study were stated as an eternal, unqualified truth, the fact that tooling and models are much superior now compared to when the study was conducted is worth noting as well.
It doesn't, and the study authors themselves are pretty clear about the limitations. The irony is that current foundation models are pretty good at helping to identify why this study doesn't offer useful general insights into the productivity benefits (or lack of) of AI-assisted development.