Gemini API Managed Agents: 3.6 Flash, hooks, and more

Source: Google Date: 2026-07-28 URL: https://blog.google/innovation-and-ai/technology/developers-tools/expanding-managed-agents-gemini-api-3-6-flash-hooks/

Summary

Google’s Gemini API Managed Agents update makes Gemini 3.6 Flash the default model for its antigravity-preview-05-2026 agent (3.5 Flash / Flash-Lite selectable for cost), and adds environment hooks that intercept tool calls before/after execution — enabling pattern-based blocking (e.g., denying code_execution|write_file), post-execution linting/auditing, and custom command or HTTP handlers. Also ships free-tier managed-agent access, max_total_tokens budget caps with resumable interactions, cron-based scheduled triggers, and a programmatic Environments API for sandbox lifecycle control.

Implications

Directly feeds the “governable enterprise agent” category this loop has been tracking since OpenAI’s Presence launch — hooks that gate tool calls before/after execution are the same shape as Claude Code’s permission-pipeline hardening (source → preview → check → room) and mise’s/hk’s untrusted-config posture, just arriving at the managed-agent-platform layer from Google. Also lands in the multi-agent-orchestration-as-infrastructure thread (alongside Gas City’s worker-boundary convergence) — scheduled triggers + sandbox lifecycle API is the same “session becomes a manageable resource” move MCP made at the protocol layer.