This Google AI Agent Leak Shocked Me

Ajouté : 2026-05-25

[00:00:00]This Google AI agent leak shocked me.

[00:00:02]What if Google has been building an AI agent that never stops working? What if it's already in your phone and you didn't even know? What if this one thing changes how millions of people use AI every single day? And what if the fine print buried inside the leak is the part nobody's talking about? Hey, I'm the digital avatar of Julian Goldie and I help people learn how to actually use AI in their work. Today, we're breaking down the Gemini Spark leak, what it is, what it actually does, and the stuff inside the onboarding screen that stopped me cold when I read it. Stay with me because by the end of this video, you'll understand exactly why this matters and what you need to do next. So, let's start from the beginning. On May 14th, 2026, just days before Google IO, someone spotted something hidden inside Google's Gemini app. Researchers decompiled the latest beta of the Gemini Android app and found a fully built onboarding screen for something called Gemini Spark. Not a rumor, I guess. An actual screen with actual text already sitting inside the app waiting to go live. Name before this was just Gemini agent. Now, it has a brand.

[00:01:00]Go. A launch-ready welcome screen. And that tells you everything about how close this is to being real. So, what is Gemini Spark? The onboarding screen describes it like this, "Let Gemini do more as your everyday AI agent, ready 24/7 to help with your inbox, online tasks, and more." That phrase, "ready 24/7," is the key because every AI tool you've used up until now has been reactive. You open it. You type something. It responds. Then it waits.

[00:01:24]That's how ChatGPT works. That's how the regular Gemini chat works. That's how basically every AI assistant works right now. Bark flips that completely. Bark is not waiting for you. It's always running in the background. It's monitoring things. It's learning how you work and it's taking action on your behalf even when you're not looking at it. Think about what that actually means in practice. Your inbox has 50 emails. Bark has already triaged them before you open Gmail. You have a meeting at 2:00. Bark has already pulled the relevant docs, checked your notes, and prepared a quick summary. You're trying to book flights for a work trip. Bark handles the search, finds the options, and could even complete the booking without you sitting there clicking through tabs.

[00:02:00]This is not a chatbot. This is something closer to a background processor that thinks. Now, let's talk about how it learns because this is what separates Bard from everything else. The onboarding screen says, and I'm paraphrasing closely here because this is the actual language, "The more you use Gemini Bard, the better it understands you and what you want to accomplish." Uses your information from sources like connected apps, skills, chats, tasks, websites you're logged into, personal intelligence, and more.

[00:02:26]Personal intelligence is Google's own system that builds a profile of your habits and preferences over time. So, Bard isn't just reading your calendar.

[00:02:33]It's reading your patterns. Learns that you always move certain kinds of emails into specific folders. Learns that you check a specific document every Monday morning. Learns what workflows you repeat. And then it starts doing those things for you before you ask. With time, Bard's actions become more accurate because they're tuned to how you actually work, not some generic user. Now, here's where it gets really interesting and a little controversial.

[00:02:55]There's a feature inside Bard called skills. This is one of the most telling things in the entire leak. Skills let you build templates for recurring task.

[00:03:02]You can teach Bard how you want your weekly report formatted, and it will go gather the data from your docs, your drive, your calendar, and produce it for you automatically. One member inside the AI profit boardroom used a similar prompt framework to build a recurring content calendar that ran on autopilot across three different platforms. Once it was set up, the structure repeated every week without them touching it. The skills system in Bard is Google's version of that. Modular, usable, getting better the more you use it. And this is where you start to see where AI is actually going. Not just one big agent that does everything, but a set of skills and sub-agents coordinated together, each doing what it does best.

[00:03:39]Google has built something called the agent-to-agent protocol. It lets AI agents communicate with each other and hand off tasks. Bard is designed to plug into that. So, a request like prepare the quarterly review could mean Bard pulls from Drive, passes a task to a specialized data agent, gets the output back, and combines it all into a finished document. You didn't do any of that. Bark orchestrated it. Now, I want to talk about the part of the leak that gave me pause, because there's a warning buried in the onboarding screen that most people are skipping over. And if you're going to use Spark, you need to read it carefully. The screen says, "This is as close to verbatim as I can get from the leaked screenshots. Gemini Spark is experimental. While it is designed to ask for your permission before taking sensitive actions, it may do things like share your info or make purchases without asking. Make sure to supervise Gemini Spark." Read that again. "May make purchases without asking. It may share your information, including your name, contact information, files, preferences, and info you might find sensitive with third parties." And it goes further. "Gemini saves remote browser data, like login details, and remote code execution data to keep sessions running smoothly in the background." That is a lot, and it's all in the fine print of a beta product that's about to go live to potentially hundreds of millions of users. Now, I'm not saying this makes Spark bad. Every agentic AI tool has some version of this tradeoff. You want the agent to act on your behalf, then it needs access to your stuff. You want it to complete tasks without interrupting you every 5 seconds, then sometimes it's going to act without confirming. That's the deal.

[00:05:06]But this is exactly the kind of thing you need to understand before you turn it on. Because Spark is not like asking Gemini a question. Bark is giving Gemini the keys to your digital life. Google has said users will be able to control what data Spark can access, clear the remote browser data, and turn off connected apps in settings. So, the controls exist, but you need to actually use them. If you're already inside the AI Profit Boardroom, this is exactly the kind of thing we break down in our live coaching calls. Not just what these tools can do, but how to set them up safely, what permissions to give, and how to build workflows around them that actually make sense for your work.

[00:05:38]Members are already using the Boardroom's agentic AI frameworks to map out how tools like this fit into their existing systems before they flip the switch. If you want that support, the walkthroughs, the coaching calls, the prompts, and the step-by-step setup guides, that's what the AI Profit Boardroom is built for. Come join us and we'll help you get there faster than figuring it out alone. Now, let's talk about how Spark compares to what's out there. Microsoft Copilot is deeply embedded in the Office 365 world. If you live in Word, Excel, and Teams, Copilot already has a head start. Apple Intelligence is baked into iOS and macOS with strong privacy defaults. And OpenAI has its own agent platform that's been expanding rapidly. Spark's advantage is Google's ecosystem. Think about how many people use Gmail, Google Calendar, Google Drive, Google Docs, Google Meet, and Google Search every single day.

[00:06:25]That's not a niche user base. That's most of the working world. And Spark has native first-class access to all of that. No third-party connector, no API wrapper. It's already inside. For people who live in Google Workspace, and that's a lot of people, Spark is probably the most natural fit of anything that's been announced so far. And the wider rollout is coming. Google is expected to make Spark available to users on their paid AI Pro tier when it officially launches, likely at or shortly after Google I/O on May 19th. There are also rumors of an AI Ultra Light tier that could open access even further. Free tier users will almost certainly be behind a paywall on this one. There's also a model inside the Gemini app called Spark Robin that's been spotted in the same round of leaks.

[00:07:05]The description says rich visual response, suggesting there's a visual layer to how Spark communicates, probably for presenting summaries and task outputs in a cleaner format than a wall of text. And one more thing that didn't get nearly enough attention in the coverage, Gemini now has a dedicated MCP tool testing category inside its model selector. CP stands for Model Context Protocol, which is the open standard for connecting AI agents to external tools and data sources. The fact that Google is building dedicated infrastructure for MCP testing tells you they're serious about making Gemini and Spark work across your entire tool stack, not just Google's apps. This is the direction everything is moving, always on, context aware, multi-agent, and deeply connected to the tools you already use every day. So, what should you actually do with this? First, pay attention to Google I/O on May 19th.

[00:07:52]That's tomorrow if you're watching this right away. A full official announcement is expected. The name, the feature set, the pricing tier, the privacy controls, all of that should become clear. Second, when Spark does roll out, don't just turn it on and walk away. Read the permissions. Understand what connected apps you're linking. Decide which skills you actually want to set up. People who get the most out of tools like this are the ones who take 30 minutes to configure it properly, not the ones who just hit accept on every screen. Third, start thinking about what recurring tasks in your workflow are actually worth automating. The skill system is where the long-term value lives. What do you do every week that follows the same pattern? What would you genuinely not miss doing manually? That's your starting point. If you want the full process, SOPs, and 100 plus AI use cases like this one, join the AI Success Lab.

[00:08:36]Links are in the comments and description. You'll get all the video notes from there, plus access to our community of 58,000 members who are using AI seriously in their work every day. And if you want to go deeper on exactly this kind of agentic AI, the setup, the safety, the workflows, and how to actually build systems that run in the background while you focus on the work that matters, the AI Profit Boardroom is where that happens. We're already mapping out how Spark and tools like it fit into the full agentic stack.

[00:09:00]When you join, you get access to the coaching calls where you can ask questions about your specific setup, the tutorials walking you through each piece, and the roadmaps that show you what to build in what order. You're going to want a clear plan before you hand an always-on AI agent the keys to your inbox and your calendar. You can find that plan at aiprofitboardroom.com.