codecowboy’s Substack

codecowboy’s Substack

i tested all 5 major generative UI providers:

WIP (work in progress)

codecowboy's avatar
codecowboy
Jun 26, 2025
∙ Paid

Criteria:

Zero-shot [x/10]

Followup-shots[x/10]

Code tinkering [x/10]

Architecture choices[x/10] + Library Choices[x/10]

Pricing $-$$$/Worth it?

TL;DR- (example, the scores below don’t matter, though I have come to conclusions for onlook)

Current state:

1st?(bolt), 2nd?(lovable), 3rd?(onlook), 4th?(v0), 5th builder.io But really who cares about the points?

a man in a suit and tie is giving a speech in front of a crowd ..

What matters is the UIUX. Do you want to rely simply on vigorously shaking a magic 8-ball. Or, would you like to interface with surgically modify your creation? Personally, as cathartic it is to bully an AI into submission, I’d prefer to inject precise prompts applied exactly to the intended element without being limited to only interfacing with the LLM as a wall of text.

Future state(Prediction):

1st(Onlook), 2nd lovable, 3rd v0, 4th (bolt)

Don’t call it a comeback just yet. Onlook has the greatest longterm UIUX potential and I have the highest hopes for that startup(pre-seed!). The others have more funding and are already impressive. I just think that the room for refinement is really high for Onlook, where the room for improvement in the other 3 will take quite the UIUX overhaul. (or maybe they can generate it lickety-split?)

That’s my love letter and aspirations for Onlook. It gives you multiple dimensions to interface in the same src code. Love that it let’s you seamlessly leverage your IDE of preference. I’m unaffiliated with them, but I do like advocating for startups that ought to exist come hell or high water. The big guys sometimes wont sweat the little things, it’s just human nature. The Onlook team is cracked for launching this and I’m here for it. All that said, they have some serious work to do to catch up. As of now I’m getting better output with less precision with the major providers so… Swim little guppy! 🎣


Let’s get into the nitty gritty shall we?

The Assessment:

Note-

I gave each providers the same two screenshots and prompt:

I want to create a "cooking mode" for recipes that makes scrolling through the steps, as you complete them, feel like  guitar hero. As you scroll vertically, the instruction+ingredient information is horizontally scrolling past (right to left) in a gantt chart progress bar of sorts as seen depicted in the image above

Onlook(still have highest hopes for, despite lowest ranking): localhost:3000 :)

bolt: https://bolt.new/~/sb1-mbhl7bc3

lovable: https://lovable.dev/projects/21d91e42-3068-4bf9-aab0-909ea408ebb8

v0: https://v0.dev/chat/cooking-mode-concept-DbQZy0hB5hA

Zero-Shot Intent Capture

this is what i found out about their zero-shot architecture decisions:

Followup-Shot Intent Crafting

these inherently varied because each gave a totally unique output.

[could they capture my intent]

Code Tinkering

[only 80% typically, this is why we will still have to roll up our sleeves at the end of the day and why AI, if at all, will only briefly take your job before your employer comes groveling back. This is a really mute topic at this point people get ahold of yourselves and stop with the existential dread and sensationalism. Are we cooked? Or, can you just not cook?]

[manual override. could I mold them back on track manually? Would they infer the direction I was suggesting or gloss over it willy nilly]

Architecture Choices + Library Choices

[These could’ve been manually over ridden. For the sake of study. I wanted to see an unadulterated glimpse with the intuition of these products]

[technically I suppose you could ask for it to use certain libraries, refraining from doing so lets you see what it defaults to and what kind of taste it has intuitively. consider that it knows best. consider your self righteous notions inferior. release control until it goes into a tail spin. then you can wad up that work, throw it in the bin, start from a blank slate and detailed prompt with strong foundational choices led by you]

Pricing/Worth It?

It’s in the eye of the beholder. I don’t know your budget or motivations. Everyone has different ROI aspirations.

I’ve only paid for bolt. I will probably test the waters with each. In the meantime I’m going to hit my daily rate limit for each of them and really have them work for their money. The most annoying paywall for free users is Onlook, but I’m not mad at them for it. They have the lowest budget and are probably having to throttle the LLM to a cheaper version rather than sonnet 3.7/gemini 2.5(speculation)

Paid Content (bonus outlook on product strategy comparisons & a rage fueled rant. you’ve been warned)


This post is for paid subscribers

Already a paid subscriber? Sign in
© 2026 codecowboy · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture