"I asked AI to fix my code and it suggested therapy instead. Fair point."
What's up builders 👋
Today's serving includes OpenAI finally giving us conversation branching (about time!), Google quietly becoming a top 10 AI site, and apparently older devs are better at vibe coding than the youngsters. Plus we've got a heartwarming story about using ChatGPT to help a disabled brother surf the web, and enough new AI tools launching to make your VS Code jealous.
Let's dive into what's shipping in the AI builder world...
🔥 Today's Top Story
OpenAI launches conversation branching in ChatGPT - the feature we've all been copy-pasting our way around finally arrives. This is bigger than it sounds. Every builder knows the pain of losing a good conversation thread because you wanted to explore a different approach. Now you can fork your ChatGPT conversations like Git branches, keeping your original context while exploring alternate solutions.
Why this matters: Think about debugging sessions where you're testing multiple hypotheses, or when you're iterating on prompts for code generation. Instead of opening 5 browser tabs or losing your perfect setup, you can now branch off and compare approaches side-by-side. This is especially huge for vibe coding where you're rapidly iterating on ideas. The feature is live now for logged-in users on web - mobile coming soon. Time to refactor those prompt engineering workflows.
🚀 Ships & Launches
• Google's EmbeddingGemma - Open-source embedding model optimized for on-device AI. Perfect for building local RAG systems without API costs
• Kimi K2-0905 Update - 256k token context window and enhanced front-end coding capabilities. Works with Claude Code and Roo scaffolds
• v0 + ElevenLabs Templates - Build AI music generators with pre-built templates. They're running a competition for best theme song generator
• SupersonikAI Launch - First AI that can run live product demos. Backed by a16z, currently in early access
• djangoforai - Django + local LLM demo with server-side events and HTMX. Clone it and have a working AI app in minutes
• AI Agent Marketplace by Cysic - New marketplace for deploying and monetizing AI agents. Early access available now
📺 Learn & Build
• Build a Voice Assistant with LangGraph and MCP - 14-day build diary for a fully local voice assistant with no API keys required
• Real-time Recommendation Engine Tutorial - Step-by-step guide using LLMs + Graph DBs for product recommendations
• Free AI Agentic Design Patterns Book - Google engineer releases comprehensive patterns guide for building AI agents (this one is viral)
• Agentic AI Crash Course - Four types of agentic systems and real-world implementation strategies
• Python Dataclass Deep Dive - Write less boilerplate code using Python's underused dataclass feature (very technical)
💬 Builder Conversations
• Vibe Coding Success Story - Developer creates web browsing tool for quadriplegic brother using ChatGPT. Community rallying with suggestions
• AI Coding Setup Debate - Pat Walls shares fix for Cursor crashes by running Claude Code in separate terminal. Devs sharing their multi-tool setups
• Claude 4 vs GPT Migration - Developers discussing mass migration from Claude to other tools for vibe coding. Split opinions on which performs better
• Time-of-Day AI Performance - Developer claims Claude Code quality varies by time of day. Others confirming similar observations
• Older Coders Better at Vibe Coding - Study suggests experienced developers adapt better to AI-assisted coding. Younger devs pushing back
📰 Industry Moves
• Marc Andreessen: "AI coding biggest earthquake since software invention" - a16z founder declares golden age of AI coding in Lex Fridman interview
• Google AI Studio hits Top 10 globally - Developer tool ranks among consumer sites, showing massive adoption
• Grok 4 claims #1 on FutureX benchmark - Reportedly beats GPT-5 Pro in real-world predictions (take with salt)
• Replit joins Google Cloud Marketplace - Enterprise customers can now purchase directly through GCP
• DuckDuckGo adds AI models to subscription - $9.99/month for privacy-focused access to GPT-4o, Claude, and Llama
• OpenAI announces LinkedIn competitor - AI-powered hiring platform launching mid-2026 (I knew they wont just stick to LLMs)
• Visa adds MCP support for AI payments - AI-enhanced payment options coming to more apps via Model Context Protocol
• Business Idea Validator with Alex Hormozi metrics - Free tool built with Claude that validates startup ideas using proven frameworks. Try it before building or build a validator yourself :)
• TikTok Viral Slide Generator - Agentic AI app creates viral slide posts in 5 seconds. Uses proprietary prompt engineering
• Cloud Coding Agents Deep Dive - Comprehensive comparison of Devin, Codex, and Cursor from real-world usage
• Top 7 Small Language Models Guide - Run powerful models on consumer hardware. Includes setup instructions for each
• DeepMind's Gravitational Wave Control - Deep Loop Shaping method improves observatory control. Code available for physics simulations
⚡ Quick Links
• Google DevRel hosting SF Bay Area hackathon this Saturday - Win Gemini API credits
• GitHub Spark enables vibe coding workflows - InfoWorld explores the new paradigm
• LLMs vs Custom Algorithms comparison - When to use LLMs vs writing specific code
• Gemini Canvas hits 200M+ edited images - Templates now available for quick generation
• Anti-AI sentiment discussion goes viral - Community debates AI's impact on creativity
One More Thing...
Here's a wild thought: if older developers really are better at vibe coding (as that TechRadar study claims), it might be because they've already learned to debug by reading tea leaves and compiler errors that just say "Segmentation fault". When you've spent decades deciphering cryptic error messages, having an AI that actually explains what went wrong feels like cheating.
What are you building this week? Hit reply and tell me about your latest vibe coding adventure - especially if you've tried that conversation branching in ChatGPT.
Keep shipping,
P.S. - Tried to teach my rubber duck about pair programming with AI. Now it just sits there judging both me AND Claude's code suggestions. Some traditions shouldn't evolve.