"I asked Claude to remember my coding style. Now it writes bugs exactly the way I like them."
What's up builders 👋
Today's a big one. Claude just dropped memory features that'll change how we work with AI assistants, UAE shocked everyone with an open-source reasoning model that's beating closed alternatives, and the vibe coding revolution just got a $3 billion validation stamp from Replit.
We're also diving into Google's new Gemini CLI extensions that'll speed up your deploy workflow, Anthropic's engineering guide on building LLM tools, and some wild experiments in "production vibe coding" that'll either inspire or terrify you.
🔥 Today's Top Story
Claude now has memory - and this changes everything for how we build with AI assistants. Rolling out to Team & Enterprise plans starting today, Claude can now remember your coding preferences, project context, and development patterns across conversations. No more copy-pasting the same context every time you start a new chat.
Why this matters: Think about how much time you waste re-explaining your project structure, preferred frameworks, or coding style to AI assistants. Claude's memory feature means you can say "use my usual auth setup" and it knows you mean Supabase with NextAuth. It remembers your API patterns, your variable naming conventions, even your specific error handling approaches. This isn't just a convenience feature - it's the difference between AI as a tool and AI as a true coding partner.
The cherry on top? They're also launching incognito mode for when you need to experiment without polluting your memory context. Perfect for those "what if I built this completely differently" moments at 10 PM.
🚀 Ships & Launches
• Gemini CLI Extensions - Deploy straight to Cloud Run with /deploy
and scan for vulnerabilities with /security:analyze
. No more context switching between terminal and browser.
• K2-Think AI Model - UAE's 32B parameter open-source reasoning model that's reportedly outperforming everything else. Claims to be world's fastest - grab it before the hype train arrives.
• ChatGPT API Wrapper - Free ChatGPT access via API instead of paying for official API. Perfect for prototypes and experiments where you don't need production reliability.
• Gemini Batch API Updates - Now supports embedding models and OpenAI SDK compatibility. Process thousands of requests asynchronously at lower cost.
• v0 State of Vibe Coding Report - v0's analysis on how vibe coding is changing development. Includes trends, risks, and predictions for where this is heading.
• Qwen3-Next - New model release generating serious discussion (141 HN comments). Another open-source alternative worth testing.
📺 Learn & Build
• Complete Vibe Coding Guide - Build 12 apps of varying difficulty with Claude Code and GPT-5. Full walkthrough from setup to deployment (Seems paid).
• MCP Server Tutorial - Connect traditional programs to Gemini CLI for data retrieval and video generation. Practical implementation guide by @lukeschlangen.
• Anthropic's LLM Tools Guide - Engineering best practices for writing effective tools for LLM agents. Direct from Anthropic's team - this is the real insider knowledge.
• LangChain Complete Series - Everything from basics to building agents, connecting ChatGPT to custom files, and extracting document insights.
• Best Fine-Tuning Guide on arXiv - Covers NLP basics, PEFT/LoRA/QLoRA techniques, Mixture of Experts, and seven-stage pipeline. (If you like technical stuff)
• OpenAI's 32-Page Agent Masterclass - Official OpenAI documentation on building AI agents. Dense but worth the read.
💬 Builder Conversations
• AI Ruined My English - "what error, please fix" - How AI coding has changed developer communication. We're all becoming prompt engineers in daily conversation.
• Check Your Model Versions - PSA about coding agents defaulting to old models. One dev wasted serious money on GPT-4 file ops when newer models were available.
• Vibe Coding Reality Check - "Only things that worked are small AI wrappers/chrome extensions." Honest discussion about vibe coding limitations for complex apps.
• Ship Broken Code to Learn - "Nothing motivates like fear." Controversial take on the fastest way to learn coding.
📰 Industry Moves
• Replit Hits $3B Valuation - Vibe coding startup's massive valuation validates the entire paradigm shift. Expect more investment in AI-first dev tools.
• Golden Age of Robotics - Costs down, market matured. New opportunities for AI developers in robotics space.
• a16z AI Office Tools Benchmark - Weeks of testing dozens of AI-native apps. Essential research for choosing your productivity stack.
• Gauss AI Solves Math Theorem - AI agent completes in 3 weeks what took humans 18+ months. Shows current state of reasoning capabilities.
• Movie Modding Experiments - VFX supervisor creating Superbad x Star Wars remixes. Early glimpse at AI-powered creative tools that let you mod movies like games.
• Weekend Project Ideas - 10 open source projects perfect for AI builders. Each one designed to be shippable in 48 hours.
• AI Drug Discovery Breakthrough - Harvard's cellular-level healing AI. Opens new opportunities for developers in biotech space.
• Real-Time AI Agents Winners - See what won the n8n/Bright Data challenge. Great examples of production-ready agent implementations.
⚡ Quick Links
• Context is the new currency in AI - Why RAG and context engineering matter more than model size.
• Vibe coding tattoo commitment - Developer considering 'CODE' tattoo if hitting 10k MRR.
• 20-minute waitlist page with v0 - Proof that anyone can ship with current tools.
One More Thing...
Here's the wild part: that developer vibe coding in production on a Hetzner server? They're not wrong. We've reached a point where AI can write better error handling than most of us, and the speed of iteration beats perfect planning every time. The future isn't careful commits - it's YOLO deployments with an AI safety net.
What are you shipping this weekend? Hit reply and tell me about your most chaotic production deploy.
Keep shipping,
P.S. - If vibe coding is worth $3 billion, my collection of half-finished side projects must be worth at least $12. Accepting term sheets via DM.