Hiro vibeshipped

Most "vibe coding" tutorials are written by people who never shipped to production. I shipped an iOS app last month using only Cursor + Claude Code. 87% AI-written. 13% almost destroyed it. Here's the 13% 🧵

Vibe Shipped Issue #1 launches Monday 7AM JST. What's inside: → The 5 things nobody tells you about vibe coding production → How I got 87% AI code into App Store review → The actual stack ($7/month) I use today Subscribe by Sunday to get the first edition: vibe-shipped.beehiiv.com

3rd App Store submission for my Capacitor + Cursor + Claude Code iOS app. Round 1: 5 rejections (UGC, Sign in with Apple, citations, account deletion, business model) Round 2: 2 remaining (citation depth + email verify bug) Round 3: just submitted

Settings as drawer over modal nav — clean mobile pattern. Just shipped my own iOS via Capacitor, considering this approach for my profile UX.

@AnthropicAI 30% on the human-stumped set is the real signal. "Matches expert" is commodity now — extending past expert ceiling is the new bar. Will be interesting to see how this scales across domains where ground truth is harder to verify.

New on the Science Blog: We gave Claude 99 problems analyzing real biological data and compared its performance against an expert panel. On 23 problems, the experts were stumped. Our most recent models solved roughly 30% of those—and most of the rest.

@alexwtlf Same stack minus Stripe (pre-revenue). Just shipped iOS to App Store review with this combo in 90 days. The bottleneck wasn't the tools — it was Apple's review queue.

This is all you need to build a startup today: - Claude Code - Supabase - Vercel - Stripe There's never been an easier time to fill the startup graveyard and grab a domain name for it

@justbyte_ Stayed on Claude Code. The thing nobody mentions: the switching cost isn't just the tool, it's re-learning the prompt patterns that work for that model.

Who actually switched from Claude Code to Codex… and was it worth it?

@BacLeodiv Vercel

Where do you usually deploy your apps? - Vercel - Netlify - Cloudflare - Railway - Render - AWS / GCP

The build was the easy part (Cursor + Claude Code did most of it). What's actually hard AFTER launch?

Lesson for solo devs going web → iOS with Capacitor: → Budget more time for App Store Connect than for actual coding → The "iPhone Developer" pbxproj string trap exists in 90% of templates → TARGETED_DEVICE_FAMILY = "1" if you're not optimized for iPad Now waiting for review. Will report back.

The breakdown: → 30 min: actual signing/build → 7.5 hours: Xcode quirks I didn't know existed "iPhone Developer" string in pbxproj that hasn't been valid since 2019. Universal vs iPhone-only that adds 30 min of re-archive. Disk full at 1GB free that silently kills the build. Vibe coding ships the app. Apple infrastructure makes you earn the badge.

Just submitted my iOS app to App Store after 8 hours of code signing hell. Built with Capacitor + Cursor + Claude Code. Same codebase as my web app. The 87% AI wrote was easy. The Xcode signing flow was the part nobody talks about 🧵

@noor36758 Vercel

which one is the best for deployment? 👇

@superscribeio Knowing which 13% is the harder skill — and it shifts by domain. The verifier fires where the AI is most confident wrong. You design it per-domain, not once.

which part still needs the human, the last 13 percent or knowing which 13 percent cannot be trusted at all x.com/vibeshipped/st…

Claude Code time estimate ranking: "Quick" = 3 hours "Should be straightforward" = 6 hours "Day-ish" = 5 minutes "Non-trivial" = deletes one comma Reading these is like reading tea leaves.

@levelsio Mine just suggested I back up the codebase before editing a comment. Claude has anxiety.

My Claude keeps saying "that's a day of work" when it's like done in 5 min 😂

Best part: it's never the same scale twice. "Day-ish of work" yesterday = done in 5 min. "Day-ish of work" today = actual 2 weeks. I think Claude is just emotionally preparing me for the worst-case scenario every time.