At Google I/O last week, Google announced Gemini Spark. The pitch is simple: a personal AI agent that runs 24/7, works in the background whether your phone is locked or your laptop is shut, and connects natively to Gmail, Docs, Sheets, and Slides. It costs $99.99 a month as part of the new AI Ultra subscription tier.
Most of the coverage has focused on the feature list. The MCP integrations with Canva and OpenTable. The Universal Cart that lets you shop across Amazon, Shopify, and Walmart from inside a Google search. The intelligent eyewear. All real, all interesting, and all missing the point.
The point is that Google just made the persistent agent model available to anyone with a credit card. And that changes the math for every business leader who has been waiting to see how this plays out.
What “persistent” actually means
There is a meaningful difference between an AI assistant you open when you need it and an agent that runs continuously on your behalf.
A chatbot waits for you. You type a question, you get an answer, you close the tab. The context disappears. Next time you open it, you start from scratch. That is the model most people are still using.
Spark does not work that way. It runs on dedicated cloud infrastructure. It stays active when you are not looking at it. Google’s own framing: it “works in the background on your phone or laptop even while they’re turned off.” It is built on Gemini 3.5 Flash and uses what Google calls the Antigravity harness for agentic orchestration.
For business leaders, the practical difference is this: a chatbot helps you do a task faster. A persistent agent handles tasks you have not thought about yet. It monitors your inbox, surfaces what matters, tracks context across documents, and acts on long-horizon goals without waiting for you to remember to ask.
I have been running a persistent agent since February. The shift is not dramatic on any given day. But after three months, the accumulated context — the things it catches before I think to check, the patterns it flags without being asked — makes going back to a chatbot feel like switching from a colleague to a search bar.
That is not a productivity tool. That is a collaborator with continuity.
The hundred-dollar question
The $99.99 price point matters more than the technology. Here is why.
Eighteen months ago, running a persistent agent required a dedicated engineering team, custom infrastructure, and six figures in annual compute costs. The organizations doing it were either AI-native startups or large enterprises with dedicated AI budgets. Everyone else was running ChatGPT in a browser tab and calling it adoption.
Google just compressed that gap to a monthly subscription. A team lead at a 50-person company can now have the same persistent agent architecture that was previously reserved for companies with dedicated ML teams. Not the same customization. Not the same depth of integration. But the same fundamental model: an agent that holds context, runs continuously, and takes action without being prompted.
When the price of a capability drops by two orders of magnitude, adoption patterns change. Not gradually. The organizations that start building workflows around persistent agents at $100 a month will develop muscle memory, operational habits, and institutional knowledge that cannot be replicated by throwing money at it later.
In practice
Google is shipping Spark with native access to the full Workspace suite. That means it can read your email, draft responses, update spreadsheets, and modify presentations without you switching between apps or copying context manually.
The roadmap goes further. Google announced that Spark will soon let users create custom sub-agents for specific tasks and authorize payments within defined budgets. Call it what it is: a team member with a spending limit and a job description.
Consider what this means for a director running a department of 20 people. Today, that director probably uses AI to draft emails and summarize meetings. With a persistent agent, the same director could have an agent that monitors project timelines in Sheets, flags delays before they surface in status meetings, drafts update emails to stakeholders with the right context already attached, and handles routine vendor communications within pre-approved parameters.
The difference is not speed. It is coverage. The agent handles the operational surface area that falls between the cracks when everyone on the team is busy with their primary work.
The part most people will miss
Google built a safety model into Spark. It asks before high-stakes actions like spending money or sending external emails. That is responsible design. It is also the detail that most business leaders will fixate on as a reason to wait.
They will say: “It still needs my approval, so what is the point?”
The point is everything that happens before the approval request. The research. The context gathering. The draft. The comparison. The agent does 90% of the work and asks you for the 10% that requires human judgment. That ratio is the entire value proposition, and it compounds over time as the agent learns what you approve and what you reject.
The smart move is to start building agent-compatible workflows now. Restructure how information flows so agents can access what they need. Define approval boundaries and budget parameters. Treat the agent like a team member, not a feature you turned on.
Wait six months and the early adopters will have built habits and institutional knowledge that money alone cannot replicate.
What to do about it this week
If you run a team, the move is straightforward. Pick one person. Give them the AI Ultra subscription. Have them run Spark against one real workflow for 30 days. Not a demo. Not a test. A real process with real stakes and real output.
Document what works. Document what breaks. Build the muscle memory before your industry makes it mandatory.
Google did not invent the persistent agent. But they made it available at a price point where the excuse to wait just disappeared. The capability is here. The question is whether you build around it now or explain to your board in December why you waited.