QuestionQ70

VerifiedSecuring and governing agentic workflows

You are designing an autonomous payment processing multiagent system on Agent Runtime in Google Cloud. The system has an orchestrator agent that delegates corporate transaction instructions to specialized banking execution tools. Safety governance policies require transactions above a defined threshold to be intercepted and approved by an authorized compliance officer before tool invocation. You need to configure a safety control mechanism that meets this requirement with minimal operational complexity.

What should you do?

Explanation

Approval breakpoints (human-in-the-loop tool confirmation) on the banking tool definitions pause execution before the tool is invoked when the threshold condition is met. Managed sessions persist the execution state until the compliance officer's response resumes the run via callback. This is the platform-native mechanism with minimal operational complexity. Cloud Tasks with polling, a custom Cloud Run wrapper with an approval database, or Pub/Sub with approval-token functions all require building and operating custom infrastructure.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!