Google enters 'agentic era' with AI overhaul across search, coding, and personal assistants

Sundar Pichai stood on stage at Google's annual developer conference in Mountain View and declared it plainly: the company had entered the agentic era. For two hours, he and his team walked through a vision of artificial intelligence so thoroughly woven into Google's products that the line between tool and autonomous agent had begun to blur. Agents in search. Agents for shopping. Agents that would work on your behalf while you slept, 24 hours a day, seven days a week. The message was unmistakable—automation was no longer a feature. It was the strategy.

The conference itself was a parade of AI announcements, each one a small proof of concept. There were nervous demonstrations where the technology cooperated as intended, and moments of relief when it did. There were also glitches: the Gemini assistant, at one point, interrupted a speaker mid-sentence with a question about her daily affirmations. But the glitches were footnotes. The real story was the systematic push to embed intelligent agents everywhere Google operated.

Much of what Google announced was a direct response to competitors who had moved faster in recent months. Anthropic, the five-year-old startup founded by the Amodei brothers, had become a genuine rival. Google unveiled Antigravity 2.0, an agent-centered coding platform that could be instructed through plain language—a direct answer to Anthropic's Claude Code and OpenAI's Codex. The platform had demonstrated its capability by building a functional operating system from scratch in 12 hours using 93 sub-agents working in parallel. Google also introduced CodeMender, a tool that automatically finds and fixes security vulnerabilities in code, mirroring the function of Anthropic's Mythos model. The company said it was inviting a limited group of partners to test CodeMender's application programming interface, but declined to name them.

The most ambitious personal product was Gemini Spark, which Pichai described as a personal AI agent. Unlike tools that respond when you ask them something, Spark works continuously in the background, handling professional and personal tasks simultaneously. It can be accessed through the Gemini app, email, or chat, and soon through voice commands on both Android and iOS devices. On Android, a new interface called Android Halo would let users track task progress in real time. During the summer, Spark would integrate directly into Chrome. For now, it was available only to a trusted testing group, with a beta version coming the following week to subscribers of Google's AI Ultra plan in the United States. The product was positioned as Google's answer to OpenClaw, the viral platform that had captured attention by promising round-the-clock AI work.

Google also refreshed its core technology. The Gemini 3.5 model was four times faster than competing systems and would power much of what the company had announced. For video generation, Google revealed Omni, a multimodal model capable of generating video from any input—text, image, or existing video. Unlike its predecessor Veo, Omni could take a video you provided and modify it based on your instructions. If you uploaded a video of a sculpture, you could ask the model to reimagine it made of bubbles. The model had been trained to understand physics more accurately than previous versions, grasping gravity, kinetic energy, and fluid dynamics in ways that produced more realistic scenes. The Flash version became available immediately through the Gemini app, Google Flow, and YouTube Shorts.

Search itself was being remade. Liz Reid, the vice president overseeing Google's search division, took the stage to announce what she called the largest update to the search box in more than 25 years. The interface would now adapt dynamically based on how users were searching—whether by text, voice, or image. The changes began rolling out that Tuesday. A user starting with a simple search could transition into AI mode and ask follow-up questions. Beyond the interface, Google was introducing information agents—personalized AI agents that could run in the background continuously, configured to hunt for specific information. One agent might alert you to newly released sneakers by particular artists. Another might find houses matching your criteria. These agents would search constantly, sending alerts when they found relevant news or listings. Information agents would launch that summer for subscribers to Google's AI Pro and Ultra plans.

Google also showed off smart glasses, though not yet for sale. The company displayed design mockups of frames developed with Warby Parker and Gentle Monster. The display-equipped glasses were being held for the end of the year. But for autumn, Google promised its first audio-enabled smart glasses—frames without a screen that could respond to voice commands and deliver information through a small speaker. A long press would activate Gemini or search your phone's data. The glasses would work with Uber, Google Maps, and Google's translation service. It was a vision of AI literally in front of your eyes, always listening, always ready to act.