Google Gemini Spark Planned the Birthday Party, But Missed the Most Important Thing

Google introduced Gemini Spark—a new AI agent that can access your personal information. A Wired journalist tested the technology by planning a birthday party and discovered it already works quite well. But she also uncovered strange blind spots in its perception.

How Spark Works

Google Gemini Spark is a new generation AI agent that can interact not just with text, but with a user's personal content. It's not simply another chatbot with a trendy interface. Spark can access the most private data and use it to help with tasks. Technically, the agent can:

• Read and analyze emails from Gmail
• View documents in Google Docs and Google Drive
• Check events in Google Calendar
• Synthesize information from different sources
• Suggest actions based on identified patterns

The logic is sound: if you give AI full access to your life, it should help much better than a regular chatbot that knows nothing about you.

The Birthday Test

A Wired journalist decided to test Spark in practice. The task was simple: plan the perfect birthday party. To do this, she gave the agent full access to her emails, documents, and calendar—everything Google knows about her life. The result was quite good. Spark successfully extracted information about her food preferences from emails, found potential guests, and checked their schedules in shared calendars. The agent even suggested several reasonable options for a party, taking into account the season and venue availability. Technically, it was impressive.

But There's One Strange Problem

And this is where things get interesting. Despite having full access to correspondence and documents, Spark missed the most important person for the journalist—her boyfriend. Her boyfriend appeared in emails, was mentioned in documents, and was listed in shared calendar events. But when the agent analyzed all the information, it somehow overlooked this obvious detail. This isn't just an interface error or some bug. It points to a fundamental problem: even with access to your personal life, AI can still fail to understand what matters and what doesn't. The agent can process data, but its logic remains superficial.

What This Means

Spark shows that AI agents have already become advanced enough for serious tasks involving personal data. But the birthday story serves as a reminder: full access to information doesn't guarantee proper understanding of context. Before giving AI full access to your emails and calendar, it's worth understanding its limitations.