Gemini Spark, Your 24/7 personal AI agent for productivity

My notes

Summary

Google’s Gemini Spark is an autonomous personal AI agent that runs background, multi-step tasks across the Google ecosystem (Gmail, Calendar, Drive, Docs, Sheets, Maps). It works under user direction, checks in before major actions, and is built around three primitives: Tasks, Skills, and Schedules. Initially limited to Google AI Ultra subscribers (18+, US) and select business users.

Key Insight

• Runs on Gemini 3.5 Flash + Antigravity, a cheaper/faster model tier, signalling Google is optimising agent loops for cost, not flagship reasoning.
• Three-primitive architecture worth noting for anyone building agent products:
  • Tasks, connect agent to Workspace apps to take action
  • Skills, reusable, user-defined action patterns (“how I want it done”) to avoid repetitive prompting
  • Schedules, time-based or conditional triggers (cron + event-driven)
• App connections are off by default; user opts in per-app in settings. Privacy framing is explicit: does not read email “indiscriminately”.
• Positioned against “standard chatbots”, proactive multi-step execution vs reactive Q&A is the dividing line Google draws between “agent” and “tool”.
• Concrete agent use cases shown: invoice/receipt hunting in inbox, turning email chains into trip plans (logging receipts to Sheets + emailing the group), home-maintenance forecasting from past invoices (recurring Calendar reminders + Keep supply list), newsletter/subscription synthesis, live web research-to-booking.
• This is Google’s consumer-grade answer to the agent-orchestration problem that teams currently solve with custom tooling; the Skills/Schedules pattern mirrors slash-command plus cron workflows.