Interview Guides
You are building an LLM agent that can take actions for users in a business workflow. Some actions are safe to run automatically, while others could cause financial, legal, or customer-facing harm if the model is wrong.
How do you decide when an agent action needs human approval before execution?
