AI Mornings with Andreas Vig
Your daily AI news briefing in under 10 minutes. New models, product launches, research breakthroughs, and industry shifts, explained clearly, no hype.
AI Mornings with Andreas Vig
Google's Agent Era Begins & Karpathy Joins Anthropic
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Hey, welcome to AI Mornings with Andreas Vig. It's the 20th of May 2026, and Google just dropped an absolute avalanche of AI announcements at their I.O. conference yesterday. Let's get into what matters. The biggest story is Gemini 3.5 Flash, a new model that represents a fundamental shift in how Google is thinking about AI. Rather than building better chatbots, they're building agents. The model is optimized for coding and autonomous work, and in demos at I.O. Google showed agents collaborating to build an entire operating system from scratch inside their anti-gravity development platform. What's remarkable is the speed. Gemini 3.5 Flash runs four times faster than other Frontier models, and there's an optimized version that's 12 times faster while maintaining the same quality. Korai Kavukwoglu, DeepMind's chief technologist, said it outperforms their latest Frontier model on nearly all benchmarks. This is now the default model in the Gemini app and in AI mode in search globally. The message is clear. Google believes the future isn't about having conversations with AI. It's about having AI actually do things for you. And that brings me to the second major announcement. Google search, as we've known it, is effectively over. The era of the 10 blue links has officially ended. Google unveiled a complete AI-powered overhaul centered around an intelligent search box that expands for conversational queries and builds custom interactive experiences on the fly. You can now create what Google calls information agents that run 24-7 in the background, tracking changes across the web and delivering synthesized updates when conditions you specify are met. Think Google Alerts, but actually intelligent. AI overviews now serve 2.5 billion monthly users, and their conversational AI mode has hit 1 billion monthly users. For comparison, ChatGPT has around 900 million weekly active users. Everything rolls out free to all users this summer, which means publishers should probably be very nervous about referral traffic. Google also launched Gemini Omni, a new family of models that can create video from any combination of images, audio, text, and existing video. Instead of just stitching inputs together, Omni reasons across all of them to produce coherent output with an understanding of physics, culture, and science. You can give it a prompt like a claymation explainer of protein folding, and it renders a stop-motion video with accurate narration. Users can create digital avatars after going through a verification process, and every video gets Google's synth ID watermark baked in. Gemini OmniFlash is rolling out today to the Gemini app YouTube Shorts and Flow. In a major talent move that caught a lot of attention, OpenAI co-founder Andrei Karpathi announced he's joined Anthropic's pre-training team. Carpathi is one of the few researchers who can bridge the gap between LLM theory and actual large-scale training practice. He previously led AI at Tesla for their full self-driving program, founded Eureka Labs for AI Education, and created the popular Neural Networks Zero to Hero course. At Anthropic, he'll start a team focused on using Claw to accelerate pre-training research, which signals Anthropic's belief that AI-assisted research, not just raw compute, is how they stay competitive with OpenAI and Google. Alright, a few more things worth knowing about today. Google connected its Genie World model to Street View, which means you can now simulate real-world environments from 280 billion images across 110 countries. You can adjust weather conditions, see what a street looks like in different seasons, or explore hypothetical scenarios. Waymo is already using this for self-driving simulations to prepare for rare events like tornadoes or unusual obstacles. OpenAI is adopting Google's synth ID invisible watermarking technology for AI-generated images, along with C2PA content credentials. They launched a public verification portal where you can upload an image and check if it contains AI metadata or watermarks. Nvidia, Kacau, and 11 Labs are also joining this cross-industry standard, and Google is bringing the same verification to Chrome and search. French AI company Mistral acquired Vienna-based EMI AI, a startup specializing in physics-based AI for industrial engineering simulations. EMI's 35-person team, including co-founders, will join Mistrol. This is their second acquisition in recent months and positions them for industrial clients across Europe. And briefly, Google introduced Gemini Spark, a 24-7 personal AI assistant integrated with Gmail that can manage your inbox. They announced audio-powered smart glasses with voice control and real-time translation. AI Studio now builds fully functioning Android apps from natural language in minutes. And a startup founded by a former teen hacker who later worked on Israel's Iron Dome just raised 28 million US dollars for AI phishing detection aimed at national security. That's it for today. See you tomorrow.