Your AI Stops the Second You Close the Tab: Gemini Spark Changes That for $20 a Month
Everything you get out of AI right now, you get by being there. You open the app, you type, it answers, you close the tab, and the whole thing stops existing until you come back. Every hour you are not sitting in front of it is an hour it does nothing for you. That ceiling has nothing to do with how smart the model is.
Gemini Spark is Google's answer to that, and it landed on my account here in the Middle East this week, which is the only reason I am writing about it now instead of in six months. Spark is an agent that runs permanently in Google's cloud instead of on your device. You hand it a task, or a schedule of tasks, and it keeps executing while your phone is locked and your laptop is shut. It lives in its own tab inside the Gemini app. Before it acts it writes a plan you can read, during execution it stops at checkpoints and asks, and afterwards it leaves a log of what it did.
Until July 30 it was an Ultra feature at $99.99 a month. Now the entry price is the roughly $20 Google AI Pro plan that most people reading this already pay for, in more than 160 additional countries. Google took a capability its rivals are selling as premium and dropped it into the plan everybody is already on.
It connects to Gmail, Calendar, Docs, Drive, Keep and Tasks, and reaches outside Google through MCP connectors to Canva, OpenTable and Instacart. It can watch your inbox for a specific kind of update and act when it appears, put together a brief before every meeting on your calendar, follow your recurring charges, and run any of that on a schedule you set once and then forget.
Two things will surprise you if nobody says them first. On the Pro tier Spark works in English. It reads and handles your Arabic files and emails fine, but you write your instructions to it in English, because full local-language conversation is still an Ultra feature. And Chrome auto browse, where Spark uses your logged-in sessions to run web errands like bookings and forms, shipped US-first and has not reached the new countries.
Neither of those is the headline. The headline is that the agent Google demoed at I/O is now a $20 checkbox for anyone here, and this guide is about which task to give it first.
Follow for more:
Course Registration: https://halaqa.app/enrollment?course=start-with-ai
Your AI Has a Ceiling, and It Is Not the Model
Think about what your AI actually did for you yesterday. Whatever it was, it happened inside a window where you were sitting in front of it, typing. That is the shape of every chatbot: you ask, it answers, you leave, it stops. The model can get twice as capable next year and that shape does not change, because the bottleneck was never the intelligence. It was that you have to be present for any of it to happen. It is why buying a more expensive plan rarely changes how much you actually get done. You are still the thing everything waits on. An agent breaks that ceiling by removing the requirement that you be there, which is a different achievement from making the model smarter, and a more useful one for most people. The unit of work stops being a question and becomes a standing responsibility. "Summarize this email" is something you say every day for the rest of your life. "Watch this inbox, and when a supplier invoice arrives, file it and tell me" is something you say once.
What Changes When the Agent Lives in Google's Cloud
Gemini Spark first showed up at Google I/O in May 2026, and the thing that makes it work is infrastructure, not intelligence. Spark runs in Google's cloud, permanently. Your device is not part of the loop. A task you hand it at midnight keeps running while your phone is locked in a drawer, and you read the result in the morning. Inside the Gemini app it has its own tab, separate from chat, and the workflow there is deliberately slow in the right places. You describe what you want, and Spark writes a plan before it touches anything, which you read and approve. Then it executes. At the points where an action carries a cost, sending an email, spending money, deleting a file, it stops and asks you. When it finishes there is a log of every step it took, and that log is usually where you find out your instruction was ambiguous. At launch it is connected to Gmail, Calendar, Docs, Drive, Keep and Tasks, and it reaches services outside Google through MCP connectors, with Canva, OpenTable and Instacart among the first. Treat the plan and the log as the actual product. An agent you cannot audit is one you will quietly stop trusting after the first surprise.
The Price Drop Is the News: Who Gets It, and When
For its first weeks Spark was an Ultra feature: $99.99 a month, United States only. The wall came down in pieces. Ultra subscribers outside the US got it next, with local language support. On July 23 it reached Google AI Pro subscribers in the United States. Then on July 30 Google opened it to Pro subscribers in more than 160 additional countries, this region among them. Five conditions decide whether you have it today. The account has to be a personal Google account, because work and school accounts do not qualify yet. It has to carry a Pro or Ultra subscription. You have to be 18 or older. You have to be in a supported country, and the exclusion list is now short: the EEA, the UK, Switzerland and Nigeria, all for regulatory reasons. And Keep Activity has to be switched on in your Google account settings, because that is where Spark holds context between runs. One more thing that is not a condition but will shape how you use it: on Pro you write to Spark in English. It reads your Arabic documents and emails without trouble, but full local-language conversation is still Ultra-only. If you meet all five and the tab is not in your Gemini app yet, do not cancel and resubscribe. It is a phased rollout, and two people on the same plan in the same city can get it days apart. Check that the account you are signed into is the one holding the subscription, update the app, try gemini.google.com, and look again in a few days.
Read-and-Report First, Write-Actions Later
The first week with Spark should look like the first week with a new assistant who already has your passwords. Give it one standing job, read its output every day, and add the second job only once the first one has become boring. Start where a mistake costs nothing, which means read-and-report: a morning brief assembled from your inbox and calendar, a weekly digest on a topic you follow, a watcher that flags any new recurring charge. Those tasks produce text you check, and nothing else. After a week of that, move to write-actions, where it files invoices into Drive, drafts agenda notes into Docs, updates a tracking sheet. The reason for the staging is not caution for its own sake. Spark acts on your live data, and the checkpoint system protects you only to the degree that your instruction was specific about what needs approval. Vague instruction, vague checkpoint. One capability not to count on yet: Chrome auto browse, where Spark uses your logged-in sessions and saved passwords to run errands on the web like starting a booking or filling in a form, shipped US-first and has not arrived in the new countries. If you see someone demonstrating it from here, they are ahead of the rollout. You will get it later. It is not a reason to hold off on everything else. Three standing tasks you actually trust is a realistic month. Five configured and forgotten is what most people will end up with.
Prompt
# GEMINI SPARK: ACTIVATION AND YOUR FIRST STANDING TASKS # Opened to Google AI Pro in 160+ countries on 2026-07-30. # ─── BEFORE THE TAB APPEARS: 5 CONDITIONS ─── # 1. A PERSONAL Google account (work and school accounts do not qualify yet) # 2. Google AI Pro (~$20/mo) or Ultra ($99.99/mo) on that same account # 3. Age 18 or older # 4. A supported country. Still excluded: EEA, UK, Switzerland, Nigeria # 5. Keep Activity ON in Google account settings. This is where Spark # holds context between runs. # On Pro you write to Spark in English. It reads your Arabic content fine. # ─── WHERE TO FIND IT ─── # Gemini app or gemini.google.com, a Spark tab beside the chat tab. # Not there? Phased rollout. Do not cancel and resubscribe. # - confirm the signed-in account is the one holding the subscription # - update the app, or try the web version # - look again in a few days # ─── WEEK 1: READ AND REPORT ONLY (copy-paste, English) ─── "Every morning at 7:30, check my Gmail and Calendar and send me a brief: top 3 priorities, meetings with prep notes, anything urgent." "Every Sunday, review my Calendar for the coming week and tell me which meetings have no agenda and what each one appears to be about." "Scan my inbox monthly for statements and alert me about any new or increased recurring charge. Report only, change nothing." # ─── WEEK 2 ONWARD: WRITE ACTIONS ─── # Only after a week where the read-only tasks were boring. "Monitor my inbox for invoices and receipts. File each one in my Drive folder Finances/2026 and add a row to my expenses sheet." "For every meeting on my calendar with no agenda, draft agenda notes in a Doc and link it to the event. Ask me before sending anything." "Research [topic] across the week and compile the findings into one Doc every Friday, organized by theme." # ─── HOW TO SUPERVISE IT ─── # 1. You describe the task. Spark writes a PLAN before touching anything. # 2. You approve. It executes in the CLOUD, your device can be off. # 3. It stops at checkpoints before costly actions and asks. # 4. It leaves a LOG. Read the log. That is where you learn your # instruction was ambiguous. # Connected at launch: Gmail, Calendar, Docs, Drive, Keep, Tasks # Outside Google via MCP: Canva, OpenTable, Instacart # ─── NOT AVAILABLE HERE YET ─── # Chrome auto browse (logged-in web errands, bookings, filling forms) # shipped US-first and has not reached the new countries. # Full local-language conversation is still Ultra-only. # ─── THE ONE RULE ─── # One standing task at a time. Add the next only after the current one # has been boring for a week. Three tasks you trust beat five you # configured and forgot.