Jensen Huang Just Dropped an AI Bombshell

"My AI assistant just asked if I wanted to refactor my prompt engineering workflow. I think we've officially reached the point where our tools are more organized than we are."

Hello Builders! 👋

Jensen Huang just dropped some real talk about Perplexity's multi-model approach, calling it "completely genius."

We've been stuck in this "pick your LLM and pray" mindset, but Perplexity's been quietly routing queries to different models based on what each does best. It's like having a toolbox instead of just one hammer. Need fast answers? Use a lightweight model. Complex reasoning? Bring out the big guns.

Google's FunctionGemma Goes Tiny - Run a 270M parameter model fully offline on mobile to parse natural language into OS actions like calendar events and contacts.

Amazon Finally Brings Alexa Web - Alexa.com launches as a web interface for Amazon's AI assistant, only about a decade late to the party.

xAI's Grok Goes Enterprise - Grok Business and Enterprise tiers now available for $30/month, targeting companies who want their AI extra spicy.

AI Coding Battle Royale - ChatGPT, Claude, and DeepSeek build the same Tetris game so you can see which model actually writes working code.

Context Engineering Deep Dive - Stop letting your LLM apps degrade over time by learning how to treat context windows like the precious resource they are.

Every Framework Showdown - Someone actually built the same app in every major frontend framework and compared performance, dev experience, and real-world viability for 2026.

Canvas: AI Knowledge Base - Open source UI kit built on Memvid's video-based memory format, treating videos as first-class knowledge objects for AI.

Roast AI Rates Outfits - Claude-coded app roasts your fashion choices in 3 hours of vibe coding, hit 200+ users with a brutal low score of 55.

Vibe Coding Every Hour - Cosmo Scharf asks if anyone else is literally vibe coding all day, sparking the question we're all thinking about.

Understanding Your Vibe Code - Madhur Prajapati asks the real question: how do you actually understand the project you just vibe coded into existence?

Making Agents Composable - Khazar Ayaz argues everyone's rebuilding the same agent infrastructure, pitching agent:// as a way to make agents public and addressable.

All AI Videos Are Harmful - Hot take generating 307 HN comments about whether AI-generated video content is fundamentally problematic regardless of quality or use case.

Nadella: Stop Calling It Slop - Microsoft's CEO wants you to think of AI as a helpful assistant, not a job-killing slop machine.

NVIDIA's Alpamayo Brings Reasoning to AVs - Open-source vision-language-action models give autonomous vehicles chain-of-thought reasoning to think more like humans behind the wheel.

NVIDIA's Rubin Chip Architecture Drops - Jensen Huang unveiled Rubin at CES, claiming it's the new state-of-the-art for AI computing infrastructure.

NVIDIA Goes All-In on Robotics - Full-stack robotics ecosystem with foundation models, sim tools, and hardware aims to become the Android of robots (bold move).

AMD's New AI PC Chips Land - Latest AI-powered processors target gaming, content creation, and multitasking as the AI PC wars heat up at CES.

Wild that we're talking about offline mobile AI and agent deployment nightmares in the same newsletter. FunctionGemma running locally while production agents are quietly breaking feels like the perfect snapshot of where we're at, the tech is simultaneously getting smaller and more unpredictable.

What are you shipping this week? Building agents, testing those Tetris code generators, or staying safely offline?

Keep shipping,

P.S. If you're deploying agents to production, maybe start with a strong drink.