Google has made Gemini 3.6 Flash the standard model for its Gemini API managed agents, streamlining the developer experience by removing the need for code changes during the upgrade. Along with this shift, the tech giant introduced new capabilities such as tool call hooks, token limits, scheduled task execution, and sandbox environment management.
Enhanced Developer Controls and Security
The Antigravity preview agent now automatically runs on Gemini 3.6 Flash, though developers can still switch to other models like Gemini 3.5 Flash or Gemini 3.5 Flash Lite, which offers lower costs and latency. The upgrade focuses on improving coding, reasoning, and multimodal analysis while reducing the number of reasoning steps and debugging cycles required for workflows compared to its predecessor.
The update also brings environment hooks that allow developers to run custom scripts before or after an agent uses tools. This feature lets them block risky actions such as code execution or file modifications before they occur, or validate outputs afterward by linting code or auditing files. When a hook denies an operation, the reason is fed back to the model so that the agent can adapt rather than proceed blindly. For example, AI investment banking platform OffDeal is already using these hooks to verify company logos before embedding them in presentations.
Another key addition is token budget controls. Developers can now set maximum token limits that cover the entire lifecycle of the task, including input, output, and reasoning steps. Once the token limit is reached, the agent pauses with an incomplete status instead of consuming more resources, preserving the sandbox state so the task can resume later with a fresh token budget.
This set of new features enhances security, spending control, and automation across Google's managed agents ecosystem, making them more solid for integrators and reducing friction when managing complex workflows.
This content is for informational purposes and does not constitute financial advice.
