From dictionary fame to smarter database queries—AI's having a moment

Hello Builders! 👋

"Vibe coding" just went from Silicon Valley slang to dictionary-official—turns out guessing your way through AI-assisted code is now a legitimate career skill.

Collins Dictionary crowned it Word of the Year while the AI world kept shipping: OpenAI added mid-query interrupts so you can course-correct without starting over, Google launched File Search for fully managed RAG, and developers are deep-diving into multi-agent SQL assistants that compare RAG strategies head-to-head.

Well, it's official: "vibe coding" is Collins Dictionary's Word of the Year. Look, I get why this won. We've all done it. "Make this button look better" to Cursor. "Fix this async bug" to Claude. But calling it "coding" may feel generous—it's more like architectural hand-waving with extra steps.

The real question is whether this is a paradigm shift or just 2025's version of "no-code." My take? It's genuinely useful for boilerplate and prototypes, but the moment you need to debug why your React app is re-rendering 47 times, you still need to know what's actually happening under the hood. Still awesome for anyone to start building, shipping and making lots of money.

OpenAI adds interrupt-and-refine to ChatGPT - Stop long-running queries mid-stream and add context without losing progress. Actually useful for deep research instead of watching it hallucinate for three minutes.

Google launches File Search for Gemini API - Fully managed RAG system lets you ground Gemini in your own data. Google calling it "simple, integrated and affordable" which means it'll cost $47/month somehow.

Gemini Deep Research goes full creepy - Deep Research can now read your emails, Drive docs, and chats. Most-requested feature apparently, though I don't remember asking AI to know everything about me.

Vidi: P2P whiteboard with no server - Real-time whiteboard collaboration that's completely distributed. Just copy the URL and you're off. Finally, a whiteboard tool that won't randomly paywall you next quarter.

CodeMachine CLI hits 0.4.0 - 500+ GitHub stars in two weeks. The update "hits different" according to the creator, which is either genuine excitement or terminal Twitter brain.

BeeBot: Foursquare founder's AI neighborhood app - Dennis Crowley's new iPhone app talks to you about what's nearby. Because what we really needed was Foursquare but it won't shut up.

Multi-Agent SQL Assistant, Part 2: Building a RAG Manager - Hands-on comparison of Keyword, FAISS, and Chroma RAG strategies. Actually useful if you're building real SQL agents instead of demos.

The Reinforcement Learning Handbook: A Guide to Foundational Questions ⚠️ - Comprehensive RL primer that won't make your brain hurt. Good for understanding what's actually happening under the hood.

AI Papers to Read in 2025 - Curated list of essential papers from classics to cutting-edge. Saves you from drowning in ArXiv's daily paper flood.

GitHub Agent: What It's Really Like to Code with an AI Teammate - Real talk about coding with GitHub's new agent. Spoiler: it's not replacing you yet, but it's getting interesting.

Running multiple Claude Code CLIs on the same project - Dennis on the cognitive load of juggling 3, 4, 5+ Claude terminals. Context switching between instances gets messy quick.

I am approaching opus usage limit - Ian hit his Opus limit after ONE prompt in plan mode on the max plan. Those tokens burn fast when you're shipping.

The trust collapse: Infinite AI content is awful - Arnon's take sparked 183 HN comments. The slop is real and people are noticing the quality degradation everywhere.

The hardest part of programming is naming variables - Kasif spent 20 minutes on userData vs userInfo. We've all been there, staring at the cursor, knowing both options suck.

Apple planning custom Google Gemini for Apple Intelligence - Apple's reportedly paying Google ~$1B annually to fix Siri with Gemini. Remember when they said privacy was their whole thing?

Perplexity paying Snap $400M to power Snapchat search - $400M in cash and equity for search integration. That's a lot of money to help people find old snaps.

Inception raises $50M for diffusion models for code - Applying image generation techniques to code generation. Either genius or another way to produce bugs with extra steps.

Subtle Computing raises $6M for voice isolation models - Finally, voice commands that work in coffee shops. Hardware device coming next year.

Amazon doing more mass layoffs - Claims it's about being "nimble," but timing suggests economic fears. The AI efficiency gains are starting to show up in headcount.

From prototype to production: What vibe coding tools must fix for enterprise adoption - Enterprise requirements for AI coding tools

We Didn't Invent Attention — We Just Rediscovered It - Historical context of attention mechanisms

I've been watching this pattern for months now, and I'm calling it: the next wave of breakout products won't come from AI features bolted onto existing tools. They'll come from people who treat AI like plumbing—invisible infrastructure that just makes their thing work better. The winners will be the ones we don't even realize are AI-powered.

Keep shipping,

P.S. If you're procrastinating on launching because your landing page isn't perfect, just remember: Craigslist still looks like it was built in 1997 and does $1B+ in revenue. Ship the thing.