Google turns Gemini Managed Agents into scheduled, sandboxed workers

Google announced a set of upgrades to Managed Agents in the Gemini API on July 28, moving the product closer to a hosted platform for autonomous AI work rather than a simple model endpoint.

Managed Agents can coordinate reasoning, code execution, package installation, file management and web retrieval inside an isolated cloud sandbox. The new release makes Gemini 3.6 Flash the default model for the current agent preview and adds model selection for teams that want to optimize for cost or latency.

The most important addition is environment hooks: custom scripts that run before or after tool calls inside the sandbox. That lets teams block risky actions, lint outputs or audit what an agent is doing without building a separate orchestration layer. Google also added token-budget caps, scheduled triggers for recurring jobs and free-tier access for projects without active billing.

For a general AI reader, the takeaway is that agent products are becoming more like managed cloud workers: persistent, scheduled, budgeted and surrounded by guardrails. The question is less whether a model can answer a prompt, and more whether it can run useful work repeatedly without surprising its operator.

Sources:

• https://blog.google/innovation-and-ai/technology/developers-tools/expanding-managed-agents-gemini-api-3-6-flash-hooks/