Google Gemini Spark rolls out as 24/7 AI agent

With Gemini Spark now rolling out to paying users, Google is moving beyond demos and blogs into real, persistent AI agents that live in people’s everyday workflows. Spark runs continuously on dedicated Google Cloud VMs, watches your email, calendar and files, and can act autonomously on longer‑running tasks instead of waiting for prompts. Combined with agentic search (“information agents”) and Gemini 3.5 Flash powering AI Mode, this is Google’s clearest execution step toward its ambition of turning Gemini into a full decision‑and‑action layer over its ecosystem.([9to5google.com](https://9to5google.com/2026/05/29/gemini-spark-ultra-us/?extended-comments=1&utm_source=openai))

For the race to AGI, this matters less for raw capability than for distribution. Billions of users already live in Gmail, Docs, YouTube and Android; if those environments quietly gain competent, semi‑autonomous agents, Google can amass unparalleled behavioral data on how humans want agents to behave, where they fail, and what guardrails work. That feedback loop could accelerate both product polish and the training of more general agentic systems. It also raises the stakes for rivals: OpenAI has to turn ChatGPT into a true “do‑things” layer just as fast, and Anthropic must find comparable surfaces through partners like AWS, Microsoft and enterprise SaaS vendors.([datacamp.com](https://www.datacamp.com/blog/google-i-o-2026?utm_source=openai))

At the same time, an always‑on agent embedded this deeply in personal data raises new questions about consent, oversight and competition policy that regulators haven’t yet fully grappled with.