Gemini Spark is the most impressive and terrifying AI experience I’ve had yet
Google's Gemini Spark built a deeply personalized Hershey weekend plan from the author's Google data. The result was useful, but it also felt invasive.
Intelligence analysis by GPT-5.4 Mini
David Pierce tests Google’s always-on Gemini Spark on inbox cleanup and trip planning, and finds it unusually good at turning scattered personal data into a polished plan. The same data-driven usefulness also makes the system feel unsettlingly intimate.
Spark is like a super-helpful robot that reads lots of notes about someone’s life and builds a perfect vacation plan. It helps a lot, but it also feels a bit like someone peeking into a diary.
Analysis
What Spark did
Pierce tries Google’s Gemini Spark on a few basic tasks first, like scanning Gmail for subscriptions and looking through Google Docs for unfinished work. In those cases, Spark performs well and even assembles a clean document with unsubscribe links.
The real test is trip planning. Pierce asks Spark to plan a family weekend in Hershey, Pennsylvania for his wife, two kids, and dog. Spark responds with a detailed itinerary in a Google Doc, including places to stay, eat, and visit.
Why it stood out
The itinerary is not generic. Spark seems to pull in information from across Google’s ecosystem: the author’s home address, his dog’s name, the ages of his children, his wife’s preferences, and even a concert already in his email. It also correctly handles practical details like free admission for an infant, a paid ticket for a toddler, and a nap window.
Pierce says the tool behaves more like a human assistant than a chatbot. It can attach the doc, draft a note, and send it to his wife using her email address.
Where it breaks
The one clear failure comes when Spark tries to book an Airbnb. It can open the site, but Airbnb blocks the login and payment step, so Spark falls back to listing available places and reminding him what he would still need to book manually.
The bigger point
Pierce’s conclusion is split in two. On one hand, Spark is one of the most impressive AI experiences he has seen: useful, personal, and genuinely well tailored. On the other hand, it feels deeply creepy because it turns private data into a service layer. The piece argues that this is the bargain modern AI is pushing: the more the system knows, the more helpful it can be, but the harder it is to ignore how much of a person is being mined to make that help possible.
Key points
- Spark is Google’s always-on AI agent and is being rolled out to the $99-per-month AI Ultra plan.
- It handled simple inbox and Docs tasks well before moving on to trip planning.
- For a Hershey weekend, it produced a highly personalized itinerary using details from Google data the author had not explicitly provided.
- It could organize and send information, but Airbnb blocked it from completing a booking.
- The story frames Spark as both a major AI step forward and a reminder of the privacy cost of that convenience.
If Spark works well, it could save people a lot of time by turning scattered emails and documents into useful plans and actions. For users already inside Google’s apps, it could feel like a real assistant instead of a chat window.
The same data access that makes Spark useful also exposes very personal details the user may not expect it to surface. If services block its actions or users feel watched, the promise of an agent that acts on your behalf can turn into discomfort and distrust.
