Gemini Spark

by Google

Google’s always-on personal AI agent — works in the background on your behalf, continuing tasks even when your phone is locked or your laptop is off, rather than the request/response chatbot model. Announced at Google I/O 2026 (2026-05-19).

See https://gemini.google/overview/agent/spark/

How it works

Three core building blocks:

  • Tasks — connect Spark to your Google Workspace and give it natural-language instructions, no coding required
  • Skills — customized actions for repetitive workflows, built up from your own usage patterns over time
  • Schedules — time-based or conditional triggers that kick off tasks autonomously

Architecture

Built on the Google Antigravity agentic harness (originally positioned as a coding platform for autonomous “junior engineer” agents), driven by the Gemini 3.5 Flash model. Spark is the productivity/personal-assistant application of that same underlying agent runtime, not a new harness of its own.

Integrations

Gmail (has its own dedicated email address you can send tasks to directly), Calendar, Drive,
Docs, Sheets, Slides, YouTube, Google Maps, and Chrome (can act on the live web). All
integrations are off by default and must be explicitly enabled in settings.

Guardrails

Doesn’t read email indiscriminately — acts only under explicit user direction, and requires
approval for major actions. Google frames it as requiring active supervision: users are
expected to check responses and interrupt when needed, not treat it as fully unattended.

Use cases

Finding and tracking opportunities, synthesizing inbox summaries, coordinating group
logistics, automating invoice management, building custom “skills” from repeated user
patterns.

Availability

Google AI Pro and Ultra subscribers (18+, select countries), plus select business users, as
of launch; wider rollout planned.

Sources