Google's Gemini Spark Runs Your Inbox 24/7 Even When You're Offline
Google's monthly Gemini Drops adds an always-on agent called Spark, a new video model, a daily briefing system, and a $100 Ultra tier.
- Gemini Spark is a 24/7 cloud-based personal agent that runs even when devices are off.
- Spark runs on Gemini 3.5 Flash and Google's Antigravity persistent-workload infrastructure.
- Rolling out to trusted testers, then US Google AI Ultra subscribers at $100/month.
- Gemini Omni turns text, image, and video inputs into cinematic video outputs.
- Daily Brief auto-compiles a morning briefing from Gmail, Calendar, and Drive.
- Neural Expressive redesign integrates Gemini Live directly into the main chat flow.
Google just bundled a small mountain of updates into its monthly Gemini Drops, and the headline shift is unmistakable: the Gemini app is moving from something you talk to into something that quietly works in the background while you do other things. The centerpiece is Gemini Spark, a persistent personal agent, but the drop also includes a new flagship-speed model, a redesigned interface, a video generator that takes mixed inputs, and a fresh $100 per month subscription tier.
An agent that keeps working after you close the laptop
Spark is the clearest signal yet that Google wants Gemini to behave less like a chatbot and more like a delegated assistant. Google describes it as a 24/7 personal AI agent designed to proactively manage tasks and help you navigate your digital life, all under your direction. Give it a task and it works in the background 24/7, even if your phone and laptop are turned off. It operates autonomously, but always under your direction. You choose to turn it on and it's designed to check with you before taking major actions.
The technical setup is what makes the always-on behavior possible. Gemini Spark runs continuously on cloud virtual machines, connecting to Gmail, Calendar, Drive, Docs, and third-party tools. It stays active even when your phone is off and your laptop is closed. The agent operates on Gemini 3.5 Flash and Google's Antigravity platform, a dedicated infrastructure layer built specifically for persistent AI workloads. On the enterprise backend, each task runs in a strictly isolated, ephemeral virtual machine on Google Cloud, ensuring zero data overlap or persistence between separate sessions.
Out of the box, Spark plugs into Workspace and a handful of partner services. Spark is completely opt in: it's up to you whether to turn it on and use it, and you can also choose which apps it can connect to. In addition to Workspace apps, it will be able to link to Canva, OpenTable and Instacart right now, with more partner apps coming in the following weeks. Google will also release more features for the AI agent in the coming weeks, including being able to give it the ability to send texts and emails and the ability to operate your browser.
Practically, that means recurring jobs you would normally schedule as cron-style automations can be expressed in plain language. You can teach it to perform various tasks, such as creating a list of critical deadlines in your Gmail and sending it to you, or writing up a summary of ongoing updates in lengthy email threads. It's also possible to set it do to recurring tasks, like making it spot hidden fees in credit card bills every month. You can program it to complete several interconnected tasks for complete workflows, as well.
Daily Brief and a redesigned shell
Sitting alongside Spark is Daily Brief, which is essentially a lighter-weight, opinionated version of the same agentic plumbing. Once you opt in, Gemini works across your connected apps in the background. It gathers urgent updates from your Gmail inbox, tracks upcoming events from your Calendar and compiles relevant follow-up details into a skimmable briefing. It goes far beyond a simple summary. Daily Brief actively organizes and prioritizes based on your specific goals, even suggesting immediate next steps. You can easily steer it by giving responses a quick thumbs up or down over time.
The app itself also got a visual overhaul. Google has redesigned the entire Gemini experience from the ground up, introducing a stunning new design language it calls Neural Expressive. The interface now features fluid animations, vibrant colors, new typography and haptic feedback. More interesting for power users, the Gemini Live conversational experience is now integrated directly into Gemini. You can seamlessly switch from typing a quick question to diving deep into a free-flowing conversation, and back again, without missing a beat. The mic was also re-engineered so you can tap and talk through a complex idea at your own pace without getting cut off mid-thought.
A faster flagship model and a multimodal video generator
The model lineup also moves forward. Gemini 3.5 Flash is described as the first in the next generation of models, combining frontier intelligence with lightning-fast action. It is the same model powering Spark, which matters because an agent that runs continuously needs cheap, low-latency inference rather than a heavy reasoning model on every step.
On the creative side, Gemini Omni is a new model that can seamlessly transform text, images and video prompts into cinematic, high-quality video outputs. It is available to Google AI Plus, Pro, and Ultra subscribers on the web and in the mobile app. Google is also pushing visual learning further, letting you zoom into details inside generated answers and pull up video overviews for specific topics rather than reading walls of text.
Why the Workspace integration matters
The real story here is positioning. Spark is purpose-built for one ecosystem, and that is both its strength and its constraint. For an individual AI agent to offer optimal performance, it must understand its human users. You give it an objective, and it takes actions to accomplish that on your behalf. For it to be good, it needs intelligence, context and information relevant to those actions. Gemini Spark is integrated into Google's widely used suite of digital tools, including Gmail and Google Calendar, giving the AI access to important contextual information about a user.
That tight coupling is the differentiator versus competitors. The clearest competitor is Microsoft's Copilot, which does similar things inside the Microsoft 365 stack. The difference is that Spark is designed to run without user interaction, while Copilot is more frequently triggered by the user within an app. OpenAI's Operator is more capable as a general-purpose web agent but doesn't have the same tight integration with a personal productivity stack. For developers building on top of Gemini, the takeaway is that Google is using its data gravity, your inbox, your calendar, your Drive, as the moat rather than raw model capability alone.
Pricing, rollout, and what to actually expect
Access is staged and tied to subscriptions. Google said it is testing Gemini Spark this week and rolling it out to Google AI Ultra subscribers, who pay $100 a month for the highest level of access to its AI tools, next week. The desktop story is on a slower track: later this summer, Google will integrate Spark into the macOS desktop app, allowing the agent to manage local files and automate on-device workflows natively.
A few caveats are worth keeping in mind before handing over keys to your inbox. Autonomous agents are not infallible. Spark can and does make errors, misunderstanding instructions, acting on incomplete information, or failing to escalate when human judgment is needed. Google includes confirmation prompts for high-impact actions, but users should approach autonomous task delegation with appropriate oversight, especially for critical workflows. Google said Gemini Spark will need to obtain a user's permission before executing high-stakes actions like spending money or sending emails.
If you live inside Workspace and you have been waiting for an AI that handles the recurring drudgery, this is the first version that genuinely looks like it can. If you do not, Spark may feel like a feature locked behind a $100 subscription and a specific stack, and the more interesting watch is whether the underlying agent harness eventually shows up in the Gemini API for developers to build on directly.