Lovable definitely seems to have a bit of a secret sauce.
I've been trying to figure out how they do it. First, Claude3.7 is probably their backend model. Gemini 2.5 Pro is definitely getting there but I'm pretty sure Claude is still king for this kind of work. Second, if you break up the design and then implementation, you get significantly better responses. Finally, you have throw in a bit of what I'm calling stable-diffusion-prompt-isms where you almost excessively drop references to known brands or design philosophies to trigger those 'latent' memories and steer away from the more-basic stuff that seems to otherwise surface.
It's so good that I think it kills off a bunch of businesses.
Anybody selling templates - dead.
Anybody selling design services to pre-seed startups and small businesses - dead.
Squarespace - probably dead soon too.
Any startup not taking advantage of these tools is stuck in the past. This is the new "going fast". You can test your ideas so quickly with these tools.
As a full stack engineer, I can build and design myself to a pretty reasonable degree. I'm not going to do that anymore. These tools are faster than me.
> It's so good that I think it kills off a bunch of businesses.
I disagree with the premise that somehow now all startups and small business would like to do their own webdesign simply because LLMs can help with some of the process, same goes for the rest of your points.
Those business won't disappear, but instead they'll be able to make more with less, just like in the past when automation been improved. People don't suddenly get fired, but instead pick up new skills and can suddenly do much more.
The dream was always that we could automate humanity enough so we can all have more free-time, but turns out that just have the same amount of free-time as before but now we're pushed to do even more in less time.
> People don't suddenly get fired, but instead pick up new skills and can suddenly do much more.
What idyllic utopia are you living in?
It cannot be the same world I'm living in, we're just seeing mass layoffs here
> The dream was always that we could automate humanity enough so we can all have more free-time,
The dream of the workforce maybe.
I think the dream of the wealthy and owner class is that they no longer need a workforce at all and can safely grind us workers into fertilizer without losing any quality of life for themselves
zero of those jobs have been replaced by AI though. We're seeing layoffs because of a one-two punch of insane overhiring during/after covid followed immediately by an economic pullback and return to "lean" operating strategies.
Those are a few of the reasons. There are more. End of ZIRP, monopolies realizing they don't have to employ all the talented engineers to prevent competition anymore, etc. But you can't exclude AI.
Chegg and StackOverflow certainly beg to differ with your hypothesis. And they're only the first to fall.
There’s a difference between your job being replaced by AI and your employer being disrupted by AI. Stack Overflow might have fewer employees, while the AI companies expand. That’s not a good explanation for a job market bust.
It’s relevant to a discussion about whether AI is causing unemployment in the tech industry. Basically anything can cause unemployment if the bar is that some company somewhere is affected in a way that leads them to have to lay someone off.
> zero of those jobs have been replaced by AI though.
Hard disagree on "zero", and I've read articles about businesses laying off people specifically to replace them with AI (Washington Post had an article about people getting let go shortly after ChatGPT first came out, e.g. copywriters), so this is just demonstrably false.
That said, I agree the larger reason for the current layoffs are (a) massive Covid overhiring, (b) end of the ~decade long ZIRP era, (c) at least in the US, much more outsourcing now that video conf tech is good and everyone is used to remote work. Long term, though, I think we've reached a state where, for a ton of jobs, technology is destroying jobs a lot faster than it's creating new ones. Lots has been written about how many startups can now execute quickly with a team half the size or less than what was required just a couple years ago. Many forms of labor have just become devalued in an incredibly short time span.
It was an article from the point of view of people who actually lost their jobs due to AI. People are simply in denial if they think this isn't happening. And this article is from nearly 2 years ago.
> It cannot be the same world I'm living in, we're just seeing mass layoffs here
I'm seeing mass layoffs all around too, I believe it's because they added too many people in the past though, not because LLMs will suddenly replace a bunch of roles that suddenly won't need humans. Of course the companies won't admit to hiring too much before, but instead find the most convenient scapegoat.
Having a different perspective is not living in a different world, it's just a different perspective.
> I think the dream of the wealthy and owner class is that they no longer need a workforce at all and can safely grind us workers into fertilizer without losing any quality of life for themselves
Certainly true in some countries like the US where the working and middle class basically has given up, but absolutely not true in other places. I'd love to see them try though, long time ago we rebalanced the scales so about time.
Why do you think your upper classes have a different dream just because your working classes are better at pushing back? They want the same thing as American ones so, and it's dangerous to forget that.
As a designer myself (and writer of this post) I'm not really worried about AI taking my job. If you notice, all designs look extremely similar between each other and I believe uniqueness will be something very valuable in the future. I do believe AI makes processes much faster (I dont code, so I could skip asking a developer to create these webs and just have the AI create them) but customization, branding and brand's tone & voice are not really at risk as for now. We'll have to see how this evolves in the future.
I definitely agree we all need to learn these tools and use them in our favor, but I'm not sure they'll directly take our jobs as for now.
I remember having the same kind of fear of loosing my job around 10yrs ago when templates and came up, but we just learnt how to include them in our workflows and continued working from there.
I build and sell templates for a living. Like shadcnblocks.com or zerostatic.io. How long do you think I’ve got? It’s seriously an existential threat for me, but my sales don’t seem affected yet.
Cursor has gone to the next level with Gemini 2.5. The reasons it gives for what it’s doing are well thought through and far more in context.
Gemini seems to now advise you when you’re telling it to do something that may not make sense - first time I’ve really seen a non-Yes Man LLM. It’s more like a Yes-but-are-you-sure man.
Considering all of these probably use either ChatGPT or Claude, is this entire business basically which company sends the best system prompt with your ask?
No. Cursor, for example, let's you choose your model, and you can even use your own OpenAI API key for example if you want.
But it's a lot more than just "the best system prompt". E.g. Cursor uses RAG (it indexes your repo in a vector DB) so it can use the right parts of your repo as context when calling the model. The biggest benefit I initially got from Cursor was asking it questions about my own repository, not a generic "how do I do X in Python" type of question
I'd also say UI and workflow mean a ton here. The agentic mode of Cursor is very well integrated.
Right, because the defining factor between Facebook and Friendster was just ‘who compiled their code better.’ Amazing take. Meanwhile, LLM wrapper startups are out here acting like sending a slightly fancier paragraph to OpenAI is the new gold rush—until OpenAI just builds the feature themselves.
No clue what you're on about unless you seriously couldn't get the irony of my comment, in which case I'm not sure it's worth trying to discuss anything more nuanced than a sledgehammer in the current company.
On a slightly related note, I have created a tool for generating web pages based on any subject using different "themes". You can create one here with a "Mario-Bro's" theme for example: https://thedeadweb.eu/?q=honey&style=mario-bros
This is so weird! I entered "German health insurance" because that's what I'm working on at the moment. Every link I clicked had the faintest whiff of usefulness, but was ultimately completely worthless. It was like a pointless maze that never had any satisfying payoff. The name is really appropriate.
yeah, it usually functions like a simplified wikipedia. It is more useful for simpler subjects in its current form, but ultimately its just a bit of fun
I tried Lovable and it made a really lovely looking landing page, but it uses god-alone-knows how many frameworks and dependencies, so now I have to throw it into something else to make it vanilla.
That's a good catch - sorry we missed including the links initially and thanks for including them here. The blog post has been updated to include links in a dedicated section.
Slightly meta. I was trying to understand your product. Does Codeyam simulate how the software looks/works without actually running it?
That seems impossible for any non-trivial project. I may have misinterpreted the idea, but calling it a "simulator" and the video leads me in that direction.
"Software simulations" is a working title but it is sort of a new kind of thing we're trying to do and describe - definitely room to improve.
Our goal is to simulate every function and aspect of your application and in doing so provides both a test suite and demonstration artifacts. it's akin to an intelligent, semantically aware fuzzer that is creating both good and bad results (vs. a fuzzer just creating bad results with the intent of breaking things).
It is connected to your code via a GitHub app with read permissions and the ability to comment on PRs when analysis happens.
Let me know if this helps (or doesn't!). I think it's a little different from what you're envisioning but you're not far off.
I also found this surprising! I think a lot of UIs are sort of still just riffs on ChatGPT type products but with some exceptions/improvements (v0 stood out here).
I've been trying to figure out how they do it. First, Claude3.7 is probably their backend model. Gemini 2.5 Pro is definitely getting there but I'm pretty sure Claude is still king for this kind of work. Second, if you break up the design and then implementation, you get significantly better responses. Finally, you have throw in a bit of what I'm calling stable-diffusion-prompt-isms where you almost excessively drop references to known brands or design philosophies to trigger those 'latent' memories and steer away from the more-basic stuff that seems to otherwise surface.