Always-On AI Arrives: Google's Gemini Spark Runs Your Tasks in the Cloud While You Sleep

At Google I/O 2026, Google unveiled Gemini Spark — a 24/7 personal AI agent that does something most consumer AI tools cannot: it keeps running when you walk away. Spark went live for trusted testers on May 19, the morning of the keynote, with an AI Ultra beta opening the week of May 25. The architectural detail that makes it a category move is that Spark tasks execute on dedicated virtual machines on Google Cloud, not on your phone or laptop.

Why 'Runs in the Cloud' Changes Everything

A normal Gemini or ChatGPT session ends the moment you close the tab. A Spark task is a persistent process with its own state, a trigger system, and a notification pathway back to you. Close your laptop and the Spark VM keeps working; lock your phone and it keeps working. The underlying model is Gemini 3.5 Flash, Google's mid-size multimodal model with a 1.05M-token context window — chosen because its speed-to-cost profile makes it viable to run ambient background work many times a day across millions of subscribers. Spark integrates natively with Gmail, Calendar, Docs, and Drive, plus three MCP-connected third-party apps at launch.

Tasks and Skills

Spark introduces two building blocks. Tasks are multi-step assignments you hand off — for example, 'find and track interior design internships in New Orleans for this summer,' which Spark monitors and updates without further prompting. Skills are reusable instruction sets you build over time: you might have Spark read your last 50 sent emails, derive a personal writing-style guide, and then apply that guide automatically every time it drafts a message. Together they turn the assistant from something you query into something you delegate to. May 2026 was the month several labs shipped exactly this kind of background, unsupervised agent — Spark is the most consumer-visible example.

Career and Business Implications

The skill that suddenly matters is delegation, not just prompting. Getting value from an always-on agent means decomposing your recurring work into well-scoped, verifiable tasks and writing reusable instructions an agent can follow unattended — closer to managing a junior employee than typing a clever prompt. For knowledge workers, the professionals who pull ahead will be the ones who audit their week for repeatable, rules-based work and hand it off, freeing attention for judgment-heavy work a background agent cannot do. For small businesses, ambient agents that monitor inboxes, track opportunities, and draft routine correspondence put a sliver of operations-assistant capability within reach of a solo founder.

One caution: persistent cloud agents with access to your email, calendar, and files raise the stakes on permissions and oversight. Before delegating anything sensitive, understand exactly what an agent can touch and how it reports back — autonomy is only an asset when you can verify what it did.