"My rubber duck filed for unemployment last week. Said Claude was taking all its hours."
This week's lineup is wild: someone vibe coded an entire MMORPG (Massively Multiplayer Online Role Playing Game), free and open source, another builder turned their ceiling into a live flight tracker for the planes flying over their house, and we're digging into why most engineers use AI but few actually engineer with it.
Hello Builders! 👋
🔥 Today's Top Story
Bob Starr shipped Boomberg, a slick site tracking US tax dollars flowing to tech companies, the moment his AI finished generating it. Months later he discovered what every security researcher has been screaming about: his vibe coded baby was leaking like a sieve. The full breakdown is required reading before you push another weekend project live.
Here's the uncomfortable truth nobody wants to hear at the demo party. LLMs are trained to make code that runs, not code that survives contact with the internet. Exposed API keys, broken auth, SQL injection holes you could drive a truck through. The model will happily generate all of it and tell you the app looks great.
If you are shipping anything that touches user data, payments, or auth, treat the AI output as a first draft from a confident intern. Run a real security scan. Your users are not your QA team.
🚀 Ships & Launches
Vibe-Coded MMORPG Goes Open Source - A fully playable MMORPG built entirely by prompting AI, free to play and surprisingly not broken.
Ceiling Becomes Live Flight Radar - Dev living near an airport vibe coded a projector setup tracking overhead planes in real-time, peak weekend energy.
ATM-Style Privacy iOS App - Place your hand on the screen to reveal only the area underneath, a years old idea AI finally unlocked.
📺 Learn & Build
Engineer With AI, Not Just Use It - The difference between copy-pasting Claude output and actually architecting with AI as a real tool.
25 Years of Coding Friction - A veteran's confession on what AI quietly removed from his work, and what he wishes it hadn't.
How Agents Decide What's Next - Demystifies tool calling so you stop treating LLM agents like a black box that occasionally works.
Prompts vs RAG vs Agents - Practical breakdown of which architecture you actually need before your team over-engineers another chatbot.
Fix Vague RAG Questions Smartly - Clever pattern: ask one clarifying question, save the default, stop annoying users on repeat queries.
Connect AI Agents to Databases Safely - Real talk on giving agents live database access without handing over the keys to production.
Full-Stack AI Agent Roadmap - 110 tutorials, 580k words, 400+ curated projects taking you from zero to shipping agents.
💬 Builder Conversations
Cursor Acquisition Spooks Loyal Users - Long-time Cursor user warns the acquisition puts proprietary code uncomfortably close to new corporate owners.
Ultra Credits Vanished Mid-Migration - Dev burns through Cursor Ultra credits 90% through a Next.js to Nuxt migration, then gets a confusing email.
Claude Code Update Goes Wild - Vibecoders losing their minds over the latest Claude Code drop, calling it a step-change in agent quality.
Ten Terminals One Dev - Tokenmaxxing thread shows one builder running opencode across 10 parallel terminals because why not.
Just Cancel Claude Already - HN's 247-comment brawl over whether open models are now good enough to ditch Anthropic entirely.
AI Is Rotting Our Brains - Nature study with 315 HN comments suggests early evidence that AI tools are measurably degrading developer skills.
The Slop Refactor Cometh - Alistair Cockburn's hot take that the next big consulting wave will be cleaning up vibe-coded codebases.
Claude Wants Your ID - Anthropic rolling out identity verification, and HN is not thrilled about handing over government IDs to chat.
📰 Industry Moves
SpaceX's $60B Cursor Grab? - Reddit rumor mill says Musk is buying Cursor for $60B, file this one firmly under "believe it when SEC filings drop."
iPhone AI Apps Leaking Keys - Hundreds of iOS AI apps shipping with hardcoded API credentials, because apparently env vars are too hard in 2024.
Deno Goes Desktop - Deno launches native desktop app runtime, taking direct aim at Electron's bloated 200MB hello-world legacy. HN is intrigued (247 comments).
Switzerland's Sovereign AI Play - Apertus drops as an open foundation model built for "sovereign AI," because every country now wants its own LLM apparently.
The SaaSpocalypse Cometh - Bolt's Taylor Bornstein argues AI flipped build-vs-buy economics, naming the four forces gutting traditional SaaS moats.
Anthropic's AI-Native Playbook - Claude team drops a founder's guide for building AI-native startups, sparking 167 HN comments of mixed reactions.
AI Democratizes SaaS Instead - Counter-take argues AI won't kill SaaS, just lower the barrier so solo devs can ship vertical tools overnight.
Boris-Style Pipeline Orchestrator - TUI that supervises headless Claude Code agents through a gated delivery pipeline, pointed at any repo.
Offline Memory for Claude - Stop re-explaining your project every session , durable, entirely offline context for Claude Code.
Obsidian as AI Brain - Turns your Obsidian vault into persistent AI memory using plain markdown, zero vector database nonsense.
Agent Memory Wiper - Lets your AI agent nuke its own context and replay the first message when things get gnarly.
Massive MoE on Macbooks - Run MoE LLMs via MLX on Apple Silicon that your Mac genuinely has no business running.
Security Hardening for Agents - Defensive workflow toolkit that helps AI agents detect what a system does and harden what matters.
9-Stage Delivery Pipeline - AI project director that wraps Claude Code, Codex, or OpenCode in a governable commercial-grade workflow.
One More Thing...
Bob Starr's "Boomberg" disaster and a fully vibe-coded MMORPG dropping in the same week is peak 2024 energy , turns out the gap between "delightful demo" and "production-ready" is exactly where most of us are living right now.
What are you shipping (or un-shipping) this week? Hit reply, I read everything.
Keep shipping,
P.S. Before you push that vibe-coded side project live, maybe ask your AI to actually engineer with you instead of just for you. Your future self (and your users) will thank you.