What Gemini Spark Is and Why I Let It In
Gemini Spark is Google’s new AI personal assistant that runs as a 24/7 background agent, using your emails, calendar, and documents to plan, organize, and carry out multi-step tasks across your digital life. It promises autonomous AI agent capabilities through three pillars—Tasks, Skills, and Schedules—that link to Gmail, Calendar, Docs, Sheets, and Slides so it can do more than answer chat prompts. According to PCMag, Gemini Spark runs on the Gemini Flash 3.5 model and lives on Google’s Antigravity platform, which lets it execute actions like booking a flight or hotel. Access is restricted to Google AI Ultra subscription customers, with plans that start at USD 99.99 (approx. RM460) per month. With that context, I switched Spark on, handed it access to my inbox, calendar, and Drive, and treated it like a real 24/7 personal agent for a week.
Setting Up Spark: Expensive Buy-In, Smooth Cross-Device Access
Getting Gemini Spark running starts with the Google AI Ultra subscription paywall, which begins at USD 99.99 (approx. RM460) per month and can go up to USD 199.99 (approx. RM920) for higher usage limits. Once subscribed, Spark appeared in my Gemini sidebar on the web and as a dedicated tab between search chats and daily briefs in the mobile app, mirroring Android Police’s description. I liked that my tasks followed me from laptop to phone without extra setup; tweak a task at your desk, and it quietly continues in the background on Google Cloud even if every device is powered down. Spark’s description in the Gemini interface stresses that it “operates autonomously, but always under your direction,” and in practice that translated into prompts for approval whenever it wanted to send an email or touch my calendar.
Putting Spark to Work: Party Planning and Inbox Drudgery
My main test was a medium-sized birthday party that needed venue ideas, price comparisons, and coordinated invites. I created a Spark Task and connected it to Gmail and Calendar, telling it to track quotes, suggest dates, and prepare an invite draft. Spark combed through existing threads with restaurants, listed options, and produced a neat table of availability and basic costs—very close to PCMag’s example of tracking vendor prices for weddings or renovations. It also suggested a date that avoided obvious conflicts in my calendar and drafted an email in my voice by turning my past messages into a “ghostwriter” Skill. This part was impressive: it saved me an hour of copy-pasting and cross-checking. When I approved, Spark proposed calendar holds and formatted my invite in a Google Doc, ready to send.
Where the ‘24/7 Agent’ Stumbles on Real-World Context
Despite its deep access, Spark still missed important context. For the party, it never noticed a note in a Google Doc outlining my strict budget, even though the document lived in the same Drive folder as the event plan. It prioritized venues that looked convenient on my calendar over places I had starred in a separate research document. In another test, I asked it to summarize the week’s most important emails and create a to-do list, mirroring Google’s own Schedules example. The recap captured deadlines, but it ignored a short, emotionally loaded message from a close friend because it lacked a clear action item. Having a 24/7 personal agent sounds like handing off judgment, but Spark is still pattern-matching around dates, numbers, and keywords. It understands logistics better than nuance or relationships.
Who Gemini Spark Is For—and What to Expect Next
After a week, Gemini Spark felt less like a sci-fi butler and more like a diligent, slightly literal assistant for email-heavy knowledge work. If you already live in Gmail, Calendar, Docs, and Sheets and you are willing to pay for the Google AI Ultra subscription, Spark’s autonomous AI agent capabilities could be valuable for recurring logistics: weekly email summaries, scheduling, travel options, and basic research checklists. Android Police notes that Spark will gain the ability to spend money on your behalf, and PCMag reports upcoming features such as texting or emailing Spark, building sub-agents, and controlling your local browser. Those upgrades could move it closer to the “24/7 personal agent” Google pitches. For now, it is great at structured digital chores, but you must stay in the loop and supply the human context it cannot infer.
