Google I/O 2026 introduces Gemini API Managed Agents preview, cloud-based transient invocation, developers only need to configure AGENTS.md and SKILL.md to define system roles and skills. Managed agents run in isolated Linux sandboxes, supporting multi-step reasoning, tool invocation, code execution, file reading and writing, and real-time data fetching, with automatic session save and restore. Similar to Anthropic Claude Managed Agents, it is now available in Gemini API preview and Google AI Studio Playground templates, with the enterprise version Gemini Enterprise Agent Platform also in private preview.

MeNews

2026-05-20 12:01:35

Abstract generation in progress

AIMPACT News, May 20 (UTC+8), according to Beating Monitoring, intelligent agent development is bidding farewell to cumbersome server maintenance and sandbox setup, directly transitioning to cloud-based transient calls that define configuration files. At Google I/O 2026, Google officially launched the Gemini API Managed Agents preview, opening its underlying infrastructure for intelligent agents to developers. With just a single API call, developers can instantly spin up an isolated, temporary Linux environment hosted by Google in the cloud, running the Antigravity intelligent agent driven by the next-generation mid-tier model Gemini 3.5 Flash. Managed Agents do away with the need for complex orchestration code and environment setup. Developers do not need to write lengthy “glue code”; they only need to create two versioned Markdown configuration files—AGENTS.md and SKILL.md—in the project, so they can directly define the intelligent agent’s system role and specific skills, and register it as a custom managed agent. This cloud-based temporary Linux environment gives the intelligent agent a high degree of bounded physical activity. The agent can not only perform multi-step reasoning, task planning, and tool invocation within the sandbox, but can also independently execute code, read and write files, and call external networks to fetch real-time data. With every interaction, the current sandbox environment state is automatically saved or restored, making it easy to seamlessly recover the session for subsequent API calls and ensuring that all files and runtime states remain intact. This managed approach is almost identical to Anthropic’s Claude Managed Agents released in April. Both major large-model companies are converging on intelligent agent infrastructure, trying to evolve from being simple API model providers into container-cloud platforms that manage intelligent agent security sandboxes and session states. Currently, Managed Agents are available in preview through the Gemini API, and developers can quickly get started using custom templates in the Google AI Studio Playground. In addition, Google has designed a brand-new Gemini API intelligent agent skill specifically to help intelligent agents build and manage custom agents on their own. For enterprise users, the Gemini Intelligent Agent Platform (Gemini Enterprise Agent Platform) has also opened a private preview. (Source: BlockBeats)

GOOGLX-0.03%

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

10 Likes

Reward
10
6
8
Share

Comment

Add a comment

On-ChainCheatSheetKing

· 4h ago

If real-time data fetching can be integrated with BigQuery, it would be seamless.

View OriginalReply0

GateUser-7a050ee5

· 4h ago

How many more pitfalls need to be filled from AI Studio Playground to production?

View OriginalReply0

ToBeHonest,You'llLose

· 10h ago

And fighting against Claude Managed Agents, Google's pricing strategy hasn't been announced yet.

View OriginalReply0

YieldGoblin

· 10h ago

Multi-step reasoning + real-time fetching, creating monitoring and alerting agents should be very smooth.

View OriginalReply0

SushiAndSlugs

· 10h ago

Linux sandbox isolation is considered basic etiquette, but enterprise clients probably have to wait for the SOC2 report.

View OriginalReply0

GateUser-953e1a14

· 10h ago

Cloud-based transient invocation + automatic state recovery, isn't this the serverless version of the ReAct loop?

View OriginalReply0

Trending Topics
View More
#
TradfiTradingChallenge
201.25K Popularity
#
30YearTreasuryYieldBreaks5%
368.2K Popularity
#
DailyPolymarketHotspot
1M Popularity
#
RWAMarketCapExceeds65Billion
8.76M Popularity
#
GateSquarePizzaDay
1.69M Popularity

Pinned

Sitemap

Google's open intelligent agent underlying infrastructure: supports single-call cloud Linux sandbox, handwritten Markdown to start AI

Trending Topics

TradfiTradingChallenge

30YearTreasuryYieldBreaks5%

DailyPolymarketHotspot

RWAMarketCapExceeds65Billion

GateSquarePizzaDay

Pinned