Gemini managed agents add hooks and token budgets

Google’s Gemini API managed agents now support 3 model choices, pre- and post-tool hooks, scheduled triggers, and total-token caps, while stopped sandboxes persist for 7 days. Google’s product announcement says a hook can deny a tool call before execution, and a capped run pauses with its environment intact rather than burning past its budget. For teams weighing the economics and lock-in of hosted agent runtimes, these controls turn Google’s service into a more credible production option—but they also make policy code, budget ceilings, and resumability the baseline against which any in-house orchestration layer should be judged.