Google did not just polish its chatbot at I/O on Tuesday. On home turf near its Mountain View campus, the company tried to rebrand Gemini as a full-time digital sidekick that does things for you, not just answers questions. At the Shoreline Amphitheatre keynote, Google rolled out Gemini 3.5 Flash, a faster and cheaper model; Gemini Omni, a multimodal world model that can even generate and edit short video; and Gemini Spark, a 24/7 personal AI agent meant to operate across Search, Workspace and other apps. The message for locals and everyone else was clear: this is less about chat windows and more about always-on AI agents.
According to Google's blog, the Gemini app is getting a "Neural Expressive" redesign, along with model upgrades and new agent capabilities. The post describes Gemini Spark as a "24/7 personal AI agent" that runs on dedicated cloud virtual machines so it can keep working even when your devices are off. The blog also notes that Omni Flash and a Daily Brief feature begin rolling out to Google AI Plus, Pro and Ultra subscribers now, with Spark following a staggered schedule: trusted testers this week, then a U.S. beta for AI Ultra subscribers next week.
Omni, Flash and the new creative tools
The Associated Press reported that Gemini Omni can turn text, photos, audio and video into short, editable cinematic clips, and that every Omni output will carry Google's SynthID watermark to flag AI provenance. The outlet added that Omni Flash is available to paid Google AI subscribers and will also be offered at no cost on YouTube Shorts and the YouTube Create app to start. In other words, the same keynote that hyped flashier creative tricks also leaned on verification tools meant to curb misuse and help viewers tell when something was made by a model.
TechCrunch noted Omni's conversational editing controls, where you can ask the model to tweak camera moves, swap backgrounds or drop a custom AI avatar into a scene. Google is initially capping outputs at short clips while it works on longer durations. The outlet also reported that avatar creation includes an onboarding step that has users record a short phrase or numbers to help reduce deepfake risk. Omni is set to surface in tools like the Gemini app, Google Flow and YouTube Shorts.
Why the Bay Area should care
Google CEO Sundar Pichai pitched the announcements as the start of an agentic era and underscored the hardware behind it. The company told attendees it now processes more than 3.2 quadrillion tokens a month and showed off new TPU 8 chips meant to support larger, faster models. Local stakes are not abstract. Hoodline previously reported that compute scarcity and the race for custom chips have already reshaped hiring and created a Bay Area talent shuffle as engineers chase guaranteed access to training capacity. The new models' promise of faster, lower-cost inference, paired with capital spending that Alphabet has signaled to expand AI infrastructure, helped make the Mountain View keynote a bellwether for the region's tech economy, according to local coverage.
Rollout and what to expect
Coverage from 9to5Google and Google's posts say Gemini 3.5 Flash, Omni Flash and the Gemini app redesign are rolling out now to various paid tiers, while Spark is set to expand from trusted testers into a U.S. beta for AI Ultra subscribers next week. Google also previewed integrations slated for this summer, including Spark in Chrome and a macOS Gemini app with local-file workflows. Alongside those moves, the company is adjusting its subscription lineup so that agent features land in front of paying customers rather than remaining a pure demo item. Expect features to arrive in waves over the summer as partner integrations and safety checks are completed.
For Mountain View and the broader Bay Area, I/O signaled that Google's big bet is on agents that quietly handle tasks in the background instead of just chatting in a tab. Whether users, creators and regulators are ready to live with always-on assistants will be the question that lingers long after the keynote sizzle reels. In the near term, the new capabilities will land first with paid subscribers and test groups, with broader rollout and heavier scrutiny to follow as Google weaves agents into Search, Workspace and everyday apps.
