The Bugged Dev

Holy moly, I finally vibe coded a website I’ve always dreamt of building 🥹 Nah, it’s not a Blender model. Not even for 3D podium. It’s pure geometry built using @threejs with color maps (albedo), spring-based physics, a hybrid DOM, real-time planar reflections, and more. Completely vibe coded all using Gemini 3 Pro in @GoogleAIStudio!! Live: basketball-beryl-nine.vercel.app

@itsshara_ai @threejs Yep, fable is best at 3D 💣

@thebuggeddev @threejs A fully interactive 3D board from one prompt is a genuinely impressive one shot result.

This is one of the wildest things I have built with AI so far 😱 Asked Claude Fable 5 to create a fully interactive 3D crime investigation board using @threejs and it did it in ONE Prompt. It took one more prompt to give it a polished UI inspired by a reference design that I created with GPT Image. What surprised me most was not how it looked. It was how it worked. Fable built a complete Verlet rope physics system with 46 particles in every thread. It handled gravity, damping, distance constraints and all other features of file upload, text and everything. The threads/ropes even lift naturally with pinned evidence instead of clipping through it. I never asked it to implement any of that. I only described the experience I wanted. Maybe that's how we should work with AI. Instead of writing long long specs, we should explain what we want as an experience and let AI do its thing. Anyway, every day Fable 5 raises the bar for what I think AI can build. More to this Live: casebook-phi.vercel.app Code: github.com/thebuggeddev/c…

@MoezZhioua @threejs I'd say it could even take months, not just weeks,.. while building this entire website didn't even take me 10 minutes :)

@thebuggeddev @threejs Impressed. Letting Fable generate the rope system under a single prompt cuts weeks of physics tweaking. Describing the desired interaction beats writing line by line specs. Outcome first prompting wins

@SahilPanhotra @threejs I know but it's one shot. Fable is king 👑

@thebuggeddev @threejs that is really good bro😅 and i can't believe this is one shot

@jacalulu Another cool integration would be to allow people to share their profile as custom URLs, kind of their portfolio showing their projects on Google AI Studio

Sometimes, it's the small improvements that feel the biggest. Love seeing this update 🔗♥️

@kyumoni Yeah, tried it one shot. Was expecting it to get the design system more closely than replicating app as is

@thebuggeddev Is it one shotted ?

Eh, i'm really not impressed by GPT 5.6 sol lol. Asked it to recreate a 3 screen app design with image generation skill. Given how well it performs on benchmarks, I expected a lot more but the results were pretty ehhh. Not even close. Maybe I need to try harder but then... wasn't it supposed to do better?

@panos_dev_ I thought OpenAI said something like that... that it's the best coding model out there, with an index score of around 80 on coding benchmarks?

@thebuggeddev On your figma file the ice cream literally flies out of the phone. You will need gpt 18pro max for that to work

@GanjaRedNight See above shared prompt. It's not one sentence at all lol

@thebuggeddev idk depends on your harness, setup, prompt, skill, if you said one sentence than that's stupid and i have to seriously debate following you

@K7335807334520K It was already done using phased execution, but I get your point. I wouldn't call it vibe coding anymore, though. At that stage, it's more like agentic engineering or maybe even vibe engineering.

@thebuggeddev Separate work into multiple phases that can implement in different pass. Trying to one shot always making all models trying to rush/cheat.

It was GPT 5.6 sol with High effort. That alone costed me whole session for just ONE prompt. If I'd gone with Ultra, I probably would've had to sell my soul.

@thebuggeddev What effort level? Have you tried ultra?

@Sogni_Protocol Hm, what's token cost for running so many agents?

@thebuggeddev you will be even more impressed with Sogni Creative Agent inside 🔥 x.com/Sogni_Protocol…

@gripe1918 what? I was referring to how GPT 5.6 didn't show any improvement over GPT 5.5 when it came to coding layouts and UIs :)

@thebuggeddev That’s imagegen dude not dependent on 5.6

@franzbruckhoff I didn't get you but I'll try to get you in future

It’s funny how we get so used to magic. Heard someone in the future saying: “Sorry I’m late buddy, the portal was opening a bit slow today and then it was glitching, I had to wait an hour for a better sync before flying through. Not impressed with their maintenance. Will get an X Warp upgrade for my ship instead, way better anyway. At least I brought you something from the Andromeda galaxy, they call it C5-B9-Beta and it tastes unlike anything else!”

No, not at all. The goal was never to build a fully functional app. The goal was to respect the design language. These models are supposed to be good at visual understanding, especially OpenAI's models with GPT Image 2.0, so I was hoping it would extract the design system from the reference, pick up all the subtle details, and implement them consistently. If you compare both designs, you'll notice differences in the color system alone. In fact, my master prompt produced a better result. For example, the title "Better ice cream better planet" had the middle words highlighted in a different color, but Plan Mode didn't even preserve that. Those are the kinds of small but important details I was looking for. It's fine if it can't perfectly recreate layouts. I can always refine those myself. But if it struggles to understand the design system and implement it consistently, that's a much bigger issue, especially considering the benchmarks OpenAI has shared :)

@thebuggeddev I guess maybe I don't understand the goal, was it supposed to build a fully functional app based on the one image? I am curious though,where do you feel it fell short based on everything you provided to it? @thebuggeddev

@Przemys96038874 It's by GPT 5.6 sol, i had switched to Luna later

@thebuggeddev You said this was made by GPT5.6 Sol, but your video says it was made by GPT5.6 Luna, the weakest of the new ChatGPT models. Or maybe you switched to Luna later in another chat and then made the video. If it really was made by Luna, then wow this is a really great app design.

@Rawnukes why would it be?

@thebuggeddev Is this bait?

@raulalhi I tried it in Google AI Studio with their "Build Android App" feature using same prompt... and ah you would not like it

@thebuggeddev I found out that to recreate images UI Gemini 3.5 Flash is really good

@ogHashbuddha That switch happened after the recording. I was just checking whether Luna could do something with the remaining tokens, but the session ended before I could. I even lost the history showing what it was thinking while developing.

@thebuggeddev ?

@MarkYourMaker Show me how you'd do it better 👀

@thebuggeddev Skill issue