Deep

Google Adds Managed Agents to Gemini API

Noah

21 May 2026 — 1 min read

Google has introduced Managed Agents to its Gemini API at I/O 2026, offering developers a complete, turnkey execution environment for AI agents that can be launched with a single API call. This move abstracts away complex infrastructure management, allowing teams to focus on agent behavior rather than backend systems.

Antigravity Agent Powered by Gemini 3.5 Flash Model

The core of the new offering is the Antigravity Agent, built on the newly announced Gemini 3.5 Flash model. It operates within a secure Linux sandbox hosted by Google, with default capabilities to execute code, perform Google searches, and read URLs. Developers can customize instructions, tools, and data, then save the configuration as a reusable managed agent for future deployments.

Free Infrastructure Preview with Token-Based Pricing

During the preview period, environmental computing resources (CPU, memory, and sandbox) are provided at no cost, with charges applied only for token consumption and tool usage. However, limitations remain: the preview does not support structured output, restricts input to text and images, and excludes certain tools such as file_search, computer_use, and google_maps. This abstracted approach eliminates the need for developers to manage sandboxes, file systems, or multi-process handling, enabling a sharper focus on product design and agent functionality.

NVIDIA and SK hynix Partner on AI Factory Memory

NVIDIA and SK hynix have entered a multiyear technology partnership to co-develop next-generation memory specifically designed for AI factories, aligning tightly with NVIDIA’s infrastructure roadmap. The collaboration spans memory for Vera Rubin supercomputers, Vera CPUs, RTX Spark PCs, and Jetson Thor robots, marking a deep integration between chip design

Goose: AI agent runs code, not just suggests

The Linux Foundation has officially unveiled Goose, an open-source AI agent that marks a paradigm shift from advisory tools to autonomous code execution, capable of installing, editing, testing, and running code without human intervention. Built in Rust and boasting over 70 extensions via the Model Context Protocol, Goose connects to

Accenture: Enterprise AI Reaches Scale in 2026

Enterprise AI is approaching a critical inflection point in 2026, as Accenture predicts a transition from experimental pilots to large-scale deployment, with customer-facing AI solutions and agentic systems moving into production. However, three major barriers—data infrastructure and cloud migration, enterprise knowledge bases, and governance plus workforce transformation—must be

Kimi Work Beta: Smarter Office Agent for Chinese Workplace

Chinese AI startup Kimi has launched Kimi Work, a local universal agent specifically designed for knowledge workers, offering a pragmatic alternative to broad-spectrum AI assistants. Early beta test results reveal strong performance in routine office tasks like report generation and data collection, though advanced capabilities remain behind a paywall. Kimi