First look at Gemini Spark, Google's 24/7 personal AI agent

My notes

Summary

A short TikTok teaser claims Google is launching “Gemini Spark”, a 24/7 personal AI agent positioned against Anthropic’s Claude (Code) and OpenAI’s Codex. The clip says it runs on “Gemini 3.5”, is built on Google’s “Antigravity” platform, executes on a dedicated VM, and works autonomously in the background. The video is pure announcement hype with no demo, benchmarks, or sourcing.

Key Insight

• The framing signals where the agent market is heading: each major lab wants an always-on agent with its own compute sandbox (dedicated VM) rather than a chat box you babysit.
• Named claims to treat as unverified rumor, not fact:
  • “Gemini Spark” product name
  • “Gemini 3.5” model version
  • “Antigravity” as the underlying agent platform
  • dedicated-VM / persistent background execution
• The competitive read is the only durable takeaway: Google explicitly aiming at the Claude Code / Codex developer-agent niche, implying the differentiator is autonomy + persistence, not just a better chat model.
• No latency, cost, capability, or availability data is given, so there is nothing actionable to plan around yet.