Google turns Gemini into a 24/7 proactive agent

For the past few years, the default way to use an AI chatbot has been simple: you open a prompt box, type a question, get an answer, and close the tab. If you walk away, the AI stops working.

At Google I/O 2026, Google made it clear that those days are over. Announcing that the Gemini app now serves over 900 million monthly active users across 230 countries (more than double its 400 million footprint last year), the company rolled out a sweeping, agentic overhaul aimed at turning Gemini from a passive conversational tool into an always-on background worker.

An agent running while your laptop is closed

The centerpiece of the keynote was Gemini Spark, a 24/7 personal agent built on the new Gemini 3.5 foundation and Google's Antigravity harness.

Unlike traditional chatbots, Spark runs on dedicated virtual machines in Google Cloud. That architecture allows it to keep chewing through complex, multi-step workflows even when your laptop is shut or your phone is locked. In a press briefing, Alphabet CEO Sundar Pichai framed Spark as a partner that takes action on your behalf while remaining under your direction.

Because Spark is wired natively into Gmail, Docs, and Slides, its real-world utility is immediate. You can instruct it to scan incoming emails from your kids' school, pull out crucial deadlines, and send a daily summary to your partner. It can sift through monthly credit card statements to flag sneaky subscription increases, draft kickoff emails from messy meeting notes, or take incoming tasks through a dedicated Gmail address. Spark also integrates directly with Chrome to browse the web, and connects to third-party tools like Canva, OpenTable, and Instacart via MCP (Model Context Protocol).

Spark rolls out to trusted testers this week and lands as a Beta for U.S. Google AI Ultra subscribers next week, with a macOS desktop integration scheduled for later this summer.

For lighter morning routines, Google also launched Daily Brief. Evolving from the CC experiment in Google Labs, this agent aggregates unread Gmail threads and upcoming Calendar events into an actionable morning digest. Daily Brief starts rolling out immediately to U.S. subscribers across the Plus, Pro, and Ultra tiers.

Conversational video editing and a visual redesign

On the creative front, Google introduced Gemini Omni, rolling out its first model, Gemini Omni Flash, to the Gemini app, Google Flow, and YouTube Shorts. Google described Omni as "Nano Banana for video," allowing users to mix text, images, audio, and video inputs into cinematic clips grounded in physical and cultural reasoning.

Editing in Omni functions like a chat thread: you can prompt the model to swap backgrounds, adjust angles, or change specific object textures without resetting the scene. Users can also generate realistic digital avatars that mirror their own likeness and voice. All generated videos embed Google's SynthID invisible watermarks.

To house these tools, Google launched Neural Expressive, a dynamic redesign rolling out globally today across web, iOS, and Android. The interface scraps plain walls of text in favor of rich imagery, interactive timelines, and animated explainers. Gemini Live has been baked into the main app with an improved mic that prevents accidental interruptions while you think out loud.

Why it matters

Google just presented the industry with a brutal distribution advantage: it already lives where all your data resides.

While rivals like OpenAI and Anthropic are also building capable autonomous agents, third-party developers face immense friction convincing users to grant permissions and wire up scattered APIs. Google doesn't need to ask for permission to bridge the gap between your calendar, your documents, and your inbox; the plumbing is already there. For regular professionals and small business owners, an agent that can triage customer emails and summarize files right out of the box is genuinely useful from day one.

However, this deep integration comes with sharp trade-offs. As early reviews from outlets like The Verge pointed out, handing autonomous control of your digital life to an always-on cloud agent sparks legitimate privacy and hallucination anxieties. Google insists Spark will ask before taking high-stakes actions like spending money or hitting send, but trusting an agent with your entire inbox requires a massive leap of faith. Putting Spark behind the top-tier Ultra subscription also signals that dedicated virtual machines won't come cheap.

We think Google made a brilliant tactical move by leveraging Workspace as its agentic moat. It shifts the AI race away from pure benchmark trivia toward practical, low-friction utility, putting immense pressure on standalone productivity tools to justify their existence.