Tech

Google unveils Gemini Spark, an always-on personal agent, and opens its agent harness to developers at I/O 2026

At I/O on 19 May, Google introduced Gemini Spark, a cloud-based agent that acts on users' behalf in the Gemini app, alongside Managed Agents in the Gemini API and an expanded Antigravity platform.

Abstract illustration of a glowing network of connected nodes, in orange and red.
Abstract illustration of a glowing network of connected nodes, in orange and red.

Google used the opening keynote of its I/O developer conference on 19 May to put agents at the centre of its product line-up. The main consumer launch is Gemini Spark, an agent inside the Gemini app that runs continuously in Google’s cloud and carries out tasks on a user’s behalf. For developers, Google introduced Managed Agents in the Gemini API, which start an agent in a sandboxed Linux environment with a single API call, and an expanded Antigravity platform that now includes a standalone desktop app, a command-line tool and an SDK.

What Gemini Spark does

Google describes Spark as “a 24/7 personal AI agent designed to proactively manage tasks”, working “all under your direction”. The difference from the existing Gemini assistant is that Spark does not wait for a question. It can run recurring jobs, respond to triggers and keep working after the user closes their laptop or locks their phone, because it runs on dedicated virtual machines on Google Cloud rather than on the device.

According to Google, Spark is powered by Gemini 3.5 and the Google Antigravity agent harness, and it connects to Workspace tools including Gmail, Docs and Slides. The examples Google gave are deliberately everyday ones:

  • checking monthly credit card statements and flagging new or hidden subscription fees;
  • scanning an inbox for school updates, pulling out deadlines and sending a combined daily digest to a user and their partner;
  • turning meeting notes scattered across email and chat into a Google Doc and drafting the companion email.

Controls and rollout

Spark is opt-in. Users decide whether to switch it on and which apps it can reach, and Google says the agent is “designed to ask you first before performing high-stakes actions like spending money or sending emails”.

The rollout is gradual. Trusted testers get access first, and Google said a beta would reach Google AI Ultra subscribers in the US the following week. Google’s subscription announcement lists Spark under both Ultra tiers and as US-only at launch. Those tiers changed on the same day: Google introduced a new $100-a-month AI Ultra plan with five times the usage limits of AI Pro, and cut the price of the top Ultra plan from $250 to $200 a month.

Google also set out what comes next. In the coming weeks, Spark is due to start using third-party tools through MCP, beginning with new connections to Canva, OpenTable and Instacart. Over the summer, Google plans to add:

  • the ability to text or email Spark directly;
  • custom sub-agents;
  • authorising payments within a set budget and with approved merchants;
  • operating the user’s browser, with Spark due to work directly inside Chrome later in the summer.

A Gemini app for macOS is available now, with Spark due to arrive on the Mac later in the summer.

Alongside Spark, Google launched Daily Brief, a simpler, ready-made agent that works in the background across connected apps, pulling together urgent Gmail messages, Calendar events and follow-ups into a prioritised morning digest. It is rolling out to Google AI Plus, Pro and Ultra subscribers, starting in the US.

Managed Agents for developers

The developer counterpart to Spark is Managed Agents in the Gemini API. Google says that with one API call, developers can start an agent that “reasons, uses tools and executes code in an isolated, ephemeral Linux environment”. The agent can also browse the web, and follow-up calls can resume the same environment with its files and state intact.

Managed Agents are powered by a new Antigravity agent built on Gemini 3.5 Flash, and are reached through Google’s Interactions API and Google AI Studio. Rather than writing orchestration code, developers can define an agent’s instructions and skills in Markdown files such as AGENTS.md and SKILL.md and register it as a managed agent. Google noted that its Deep Research agent, launched in December, was its first managed agent; this release opens the same harness and infrastructure to outside developers.

The feature was rolling out in preview at launch, with enterprise access in preview through Google’s Gemini Enterprise Agent Platform. Google’s launch post did not set out separate pricing.

Antigravity grows beyond coding

Antigravity, which began as an agent-first coding environment, is being broadened. The new version includes a standalone desktop application for running several agents in parallel, an Antigravity CLI for terminal users, and an Antigravity SDK that gives programmatic access to the same harness, including the option to host agents on a developer’s own infrastructure. Google is encouraging users of its existing Gemini CLI to migrate to the Antigravity CLI.

The scale behind it

Google framed the announcements against the growth of the Gemini app, which it says now has more than 900 million monthly users, up from 400 million at I/O a year earlier.

What it means

Analysis. The common thread across these launches is that Google wants one agent harness to sit behind its consumer app, its developer API and its coding tools. That gives Google a single system to improve, and means changes to the harness should reach all three audiences.

The bigger change is on the consumer side. An agent that works in the background with access to email and documents, and later to payments, asks users for a different kind of trust than a chatbot that only answers questions. Google’s approach at launch is opt-in access, per-app permissions, confirmation before high-stakes actions, and a staged rollout that starts with paying US subscribers. How those safeguards hold up in everyday use, and how quickly Spark reaches a wider audience, are the things to watch over the summer.

Sources

See All

More on Google

See All

Type to search. ↑ ↓ to navigate, ↵ to open.