An insurer already knows the claim severity. Posting the reserve is then a fixed sequence of connector calls, with no further judgment. Which design fits that posting step?
Select an answer to reveal the explanation.
Short Explanation
Once the severity is known, posting the reserve is a checklist of connector calls, not another opinion. You design an agent flow that runs those steps in the fixed order. A prompt that asks the model to judge the reserve again reopens a closed decision.
Full Explanation
An agent flow is the design for a fixed path of connector steps. Severity is already known, and posting the reserve does not ask the model to judge anything, so the flow runs the connectors in that order and posts the reserve. Another prompt that asks the model to judge the reserve again turns a settled input into a new opinion and can post a different amount than the severity requires. A topic that only explains what a reserve is answers a question, and it does not call the connectors that post the money. An autonomous agent that reopens severity scoring on a schedule redoes the decision the process already finished and then posts whatever it prefers, which is a second judgment rather than a fixed post. The flow can still be started by the agent once severity is known. The posting steps themselves stay deterministic. Exam caveat: a prompt action belongs on a step that needs a model judgment. This step was defined as having none. Operational check: run a test claim with a known severity and confirm the reserve posted matches that severity, with no prompt step changing the amount.