An agentic workflow is a delegation decision as well as a technical design. The useful question is not simply whether a system can take an action. It is whether that action fits the authority and evidence available to it.
Begin with the task
Describe a concrete outcome. Identify the systems involved, the information required and the people affected. A precise task makes it easier to decide which actions should be automatic and which deserve review.
Make authority explicit
Reading a document, preparing a draft and sending a message are different kinds of action. Define permissions at that level. Decide who can grant access, how it can be withdrawn and where human approval is required.
Design for interruption
A useful workflow needs a way to stop, inspect progress and recover from mistakes. Keep records of important actions and make failure visible. An agent should not need to pretend it has succeeded when a tool or source is unavailable.
Evaluate the whole workflow
Review outcomes, exceptions and the work left for people. A fluent answer is only one part of a useful system. The larger test is whether the workflow improves the task without hiding new burdens.