Google has rolled out a major update to Managed Agents in the Gemini API, adding new controls for model selection, security, spending, and automated execution. The update builds on the earlier release that introduced background tasks, custom agent configurations, and remote MCP (Model Context Protocol) server support.
Gemini 3.6 Flash Is Now the Default
The antigravity-preview-05-2026 agent now runs on Gemini 3.6 Flash by default, requiring no code changes from existing users. Developers can still pick a specific model explicitly when creating an interaction or managed agent, using agent_config.model.
Supported models include:
- Gemini 3.6 Flash (default) balanced model for reasoning, coding, and tool use
- Gemini 3.5 Flash previous generation model for general agentic workflows
- Gemini 3.5 Flash-Lite lowest latency and cost on the Gemini 3.5 family
Environment Hooks: Block, Lint, and Audit Tool Calls
A new feature called Environment Hooks lets developers run custom scripts before or after tool calls made inside the sandbox. Users can add a .agents/hooks.json file into their environment to execute handlers on pre_tool_execution or post_tool_execution events.
The matcher field in hooks supports regular expressions, letting developers target multiple tools at once or catch all tool calls with a wildcard.
These hooks are already being used by teams to build production-grade validation pipelines. For example, AI-native investment bank Offdeal uses post-tool-execution hooks to automate image verification inside a remote sandbox as part of its pitch deck generation workflow, ensuring every visual meets brand and quality standards before it’s approved.
Cost Control and Automation Features
Alongside model and hook updates, Google introduced several operational controls:
- Free tier availability Managed agents are now available on free tier projects, letting developers test agentic workflows using an API key without activating billing.
- Budget controls Developers can pass max_total_tokens inside agent_config to cap total token consumption (input, output, and thinking combined) and prevent runaway multi-turn tasks from overspending.
- Scheduled execution with triggers Developers can automate recurring agent tasks using scheduled triggers, allowing a persistent agent environment to prompt itself and run cleanup or maintenance jobs without manual intervention.
New Environments API
Google also launched an Environments API that lets developers list, inspect, and delete sandbox sessions. This gives developers a way to recover environment IDs after a disconnect or clean up unneeded sessions instead of waiting for the standard seven-day expiration period.
Getting Started
The updates are aimed at moving managed agents from experimental use into cost-controlled, scheduled real-world deployment letting teams run autonomous workers without exceeding budgets or needing constant external orchestration. Developers can check the Gemini Interactions API overview and the managed agents quickstart guide to explore custom agent definitions, environment rules, network access settings, and advanced deployment patterns.
