Google shifts to agentic AI

1/ Google’s May 19 I/O keynote made a product-positioning change that matters: the company talked less about AI as a chatbot and more about AI as an agent that can keep working across products, sessions and devices. Google framed the event as the “agentic Gemini era.” (blog.google) 2/ The clearest product signal was Gemini 3.5 Flash. Google described it as the first model in its latest family to combine “frontier intelligence with action,” while TechCrunch reported Google pitched it as its strongest model yet for coding and autonomous agents. (blog.google) 3/ TechCrunch reported Gemini 3.5 Flash can execute coding pipelines, manage research projects and run for multiple hours, with Google saying the model is built for cases where several agents work on long-running tasks at once. DeepMind CTO Koray Kavukcuoglu told reporters it outperformed Google’s 3.1 Pro on nearly all benchmarks he cited. (techcrunch.com) 4/ Google also used I/O to describe how the model stack is supposed to work. According to TechCrunch, Tulsee Doshi said the forthcoming Gemini 3.5 Pro is meant to act as an “orchestrator” or planner, while Flash can serve as the sub-agents doing tool-heavy work. That is a different architecture from a single chat interface answering one prompt at a time. (techcrunch.com) 5/ The consumer-facing expression of that strategy is Gemini Spark. Google said Spark is a “24/7 personal AI agent” designed to proactively manage tasks and help users navigate their digital lives “under your direction.” It sits alongside a new Daily Brief feature that pulls together a personalized morning overview. (blog.google) 6/ That matters because Google is not limiting the agent pitch to one app. In its I/O roundup, the company said it is “unlocking agents and agentic experiences across our products,” including Information Agents in Search, Spark and Daily Brief in Gemini, and a Universal Cart for shopping. (blog.google) 7/ Search is part of the same story. Google’s I/O collection page described “a new era for AI Search,” and TechCrunch separately characterized the broader launch as a move away from AI as a conversational layer toward systems that plan and execute work. In practice, that means Google is trying to move from answering queries to handling workflows. (blog.google) 8/ The hardware side is important too. Google said Gemini-powered intelligent eyewear is coming this fall on Android XR, built with Samsung and Qualcomm, with frames from Gentle Monster and Warby Parker. Google said the first products will be audio glasses, followed later by display glasses. (blog.google) 9/ Google’s description of those glasses is notably agent-like, not just assistant-like. The company said users will be able to ask Gemini for directions, send texts, summarize missed messages, translate speech and writing, take photos and get help with multi-step tasks in the background. (blog.google) 10/ The operating-model change is the key point. A chatbot waits for a prompt and returns an answer. The products Google described at I/O are meant to retain context, trigger proactive updates, coordinate subtasks and act across software and hardware surfaces. That reading is based on Google’s own language around “24/7 help,” “under your direction,” and agents that help users “act.” (blog.google) 11/ That also raises the obvious trust and control questions. TechCrunch reported Doshi said Flash will sometimes pause to ask for user input when it reaches a decision point or permission issue requiring human judgment. Google, in its Android and Gemini materials, also emphasized that users stay in control and that proactive help is meant to operate under user direction. (techcrunch.com) 12/ The bigger takeaway from I/O is not one model name or one gadget. Google is trying to make Gemini the layer that works across Search, the Gemini app, shopping and wearable devices, with different models and interfaces handling planning, execution and ambient assistance. That is the company’s latest public blueprint as of May 19, 2026. (blog.google)