Buried in the I/O 2026 dev announcements: Managed Agents in the Gemini API. A single API call gives you an agent that reasons, calls tools, and executes code inside an isolated Linux environment — no infra to provision. Environments are persistent and resumable across multi-turn sessions, and you define agent behavior with plain markdown files.

This collapses a whole category of backend work — sandboxing, session state, tool routing — into a managed primitive. Combined with OpenAI's agent-tooling push, the "run untrusted agent code safely" problem is quietly becoming a checkbox.

Source: blog.google