Behrad Mirafshar

UX designers, I just vibe-coded a full Booking.com homepage — and I never even touched Figma. Yup, not once. Instead, I used a trio of tools to compare results: Lovable, Bolt, and my own Frankenstack — Windusrf + Claude Sonnet. The result? Ridiculously fast. And fun to play with. But here’s the twist: vibe coding comes with its own constraints. Tools like Lovable are engineered for dopamine. You’ll get to the “WOW!” moment quickly. But after that high? The limitations hit hard. You can’t tweak microinteractions. You can’t finesse the details. You’re locked into a magic box. On the other hand, developer-led environments are messy. Slower. Less “magical.” But infinitely more customizable. You can stitch inspirations together. Swap libraries. Adjust flows. And coax your AI to interpret your references. Is it perfect? Not at all. Is it flexible enough for viral-level UX? You bet. ⚡ Vibe coding is a fun way to ideate. But if you’re serious about shipping delightful, layered UX — you’ll still need to wrestle with the chaos. P.S. We ship MVPs a lot faster for our clients by vibe-coding them into existence. Clients love it because, finally, they’ve found a product studio that speaks their language. What do they want? To be in the market today! Given our deep experience in Figma and knowing how to use the vibe-coding platforms, we can ship an app with viral UX without compromising on the nuances. #ux #uxdesign #vibecoding #lovable #bolt #claude

@mvanhorn @PabloFuente @gregisenberg I'm using it like a charm. will send you over what i've built soon.

@B_AFSH @PabloFuente @gregisenberg Isn’t there no open sign up for Reddit api? Have to shoot and wait?

I sat down with matt van horn (@mvanhorn) and watched him turn claude code into a real-time research engine with his /last30days claude code skill he "fixes" claude code in 30 seconds. this skill pulls what’s actually working right now from x, reddit, and the web, then feeds that context straight into your prompts so you stop building off stale advice. we went from trending rap songs → cold email frameworks → researching clawdbot → planning and building a competitor live, with almost zero hand-written code. pretty nifty little claude code skill share this with a friend / full ep available on @startupideaspod where i will give you ideas/tools/tutorials to make your dreams a reality i will not hold back any alpha and this claude skill is alpha forsure you can install this claude code skill in 30 seconds dream big my friends

@PabloFuente @gregisenberg @mvanhorn yes soon. I made it operational and turned into a telegram bot. now you can do a lot. @mvanhorn the proxy chatgpt is unecessary. just use the direct reddit api works much better and it's for free.

@B_AFSH @gregisenberg @mvanhorn Can you share? Thanks

CAVE mode claude code. Blocking everything. will get out after hitting rate limits

@kloss_xyz lol

You’ve reached your Claude usage limit. Your limit will reset in 5 hours.

I just went to sleep and RALPH did deliver 70% of tickets. Just crazy. Claude code is the future of work. I can see companies prefer to hire someone master Claude code over a cto with years of experience.

Took me 3 days fromst start to finish to complete 2100 lines of PRD, tables.md and userflows.md. Now my ralph loop created with over 100 tickets. Ready to run it overnight and expect to play around with with my app tomorrow morning. claude code counting on you.

Claude Code skip your skill if it sees the prompt asks for something straightforward. It made me pissed but now thinking about it it makes sense. I would do the same think. I don't need my years of financial expertise for a simple algebra. don't try to hack it that it always trigger your pre-defined skill. Just create a better prompt.

done installing all the skills i need on claude code to conquer the world. Hello new brave world, i'm coming

@udaysy have you tried to RAG it? overtime accumulates and claude tends to overlook things. makes sense what you're suggesting. so one project?

yep this is the tricky part. what worked for me: one top level CLAUDE.md that has the cross-cutting context (company goals, shared conventions, key decision log) and then department specific ones that inherit from it. for truly shared info i also just keep a /docs folder that all projects can reference. not perfect but better than total siloes

Claude desktop is great. but I want to move my entire work with claude code. is relying on skills the way to go? any tips?

@0xgaut yup. that hits hard. right to the bone

realizing you don’t need to know how to code to build good ideas anymore but you have no good ideas

@udaysy Uday I have different projects like ops department, marketing department, website design department, following your approach more or less. but the information is siloed and other departments can't use it. any experience with this?

@B_AFSH skills are nice but not required to start. the big thing is CLAUDE.md in your project root. put your project context, coding conventions, and architecture there. claude code reads it automatically. start simple and add skills as you find repetitive workflows

Claude code skills at the user level, interlinked together, are the future of work. Loop them into RALPH, and now we’re talking full-flavour Iranian cuisine.

@outboundman Or let it crash and mediate among the ruins?

@B_AFSH Gotta hit pause before the system crashes for good.

my work-life balance is defined when i've hit my daily limit on claude code, then i rest till it rests

I'm having fun with Gemini 3 Pro. Right prompt could open many doors.

A founder came to me with a €420K proposal from another vendor. 12-month roadmap. We shipped the same scope for €75K in 90 days. When I looked at the scope for the first time, told them: "You're paying for 9 months of work you'll throw away." Here's what they were charging you for that you never needed. 👉 Most dev agencies sell you setup. OAuth. User roles. Component libraries. Data pipelines. They scope it. Bill it. Build it from scratch. Every single time. We boilerplated it. When you look at a SaaS vertical, 70% of the infrastructure is identical. Authentication. Permissions. Admin panels. Notification systems. Why would we charge you to rebuild plumbing? That's the first unlock. 👉 The second is harder to swallow. We could've built the founder's full vision. Every feature on the roadmap. Every "nice to have" stakeholders added in meetings. We didn't. We asked different questions: + What unlocks the operational bottleneck right now? + What gets you to the next funding round? + What lets you scale users without scaling headcount? When you look at development through that lens, 80% of what you thought you needed becomes noise. We ended up building internal tooling with a thin client-facing layer. That's it. Ruthless omission isn't cutting corners. It's refusing to build things that won't move the needle for another 6 months. 👉 Third: we mapped the user journey and business model before touching code. Every founder says they know this cold. Every time we dissect it together, we find different insights. Not because they're wrong. Because they're too close. We map it to prioritize, not to create another deck to store in google drive. A 3-day user journey and a 12-month user journey require completely different products. Founders copy Notion and Linear because they're cool. Context be damned. 👉 Fourth: AI where it actually matters. Not "sprinkle some ChatGPT on it." RAG architecture. MCPs. Systems that ingest thousands of documents and retrieve exactly what you need, when you need it. The founder now runs reports that used to take a week. Takes an hour. 💎 That's not a feature. That's buying back 40 hours a week. Permanently P.S. If you've got a proposal sitting on your desk with a six-figure number and a 9-month timeline, send it to me. I'll tell you what you actually need to ship in 90 days.

I tested Gemini 3 at 2am last night. By 3am, I forgot I was supposed to sleep. Most LLMs understand prompts. Gemini 3 understands creative intent. The gap between those two things is massive. I fed it a creative brief. The kind that usually takes one hour of back and forth with any model. It nailed the tone, the structure, the nuance on the first passes. My theory: Google trained this on a mountain of YouTube content. Not just transcripts. Context. Emotion. How humans actually communicate when they're not writing for algorithms. Claude is powerful. But for understanding creative briefs, vocabulary richness, and capturing tone? Gemini 3 is far better. Not even close. This was one hour of exploration. Imagine what's possible with a week of focused work. If you're thinking about rebranding or rebuilding your website, you can't ignore gemini 3 and Nano Banana. P.S. We've been pairing Gemini 3 with our design system to ship website copy and landing pages in days, not weeks. If you want to see what's now possible for your next rebrand, book a fit call.