Cuestionario vinculado a una guía · Medio Nivel

Prueba de automatización del flujo de trabajo de IA

Prueba los límites del flujo de trabajo, la validación semántica, la autorización, los reintentos, la finalización parcial, las aprobaciones y los resultados de un extremo a otro.

Rutas de guía relacionadasAutomatización del flujo de trabajo de IA
Pregunta 1 de 8

When mapping an automated workflow, which parts should be marked separately?

Vista previa de cada pregunta8 preguntas

Las respuestas y explicaciones aparecen a medida que realiza el cuestionario.

  1. When mapping an automated workflow, which parts should be marked separately?
    • Only the model prompt and vendor
    • Trigger, inputs, decision points, actions, and completion condition
    • Only the first and last screen
    • The number of steps that use AI
  2. A model classifies an invoice, and code sends payment. What validation is needed before payment?
    • Check only that the response is valid JSON
    • Ask the same model whether it is safe
    • Check the schema and meaning, including account, amount, destination, and authorization
    • Trust the classification label as permission
  3. A request times out after an external action may have succeeded. What should a retry do?
    • Repeat immediately because there was no response
    • Assume the operation definitely failed
    • Start a duplicate workflow under a new identifier
    • Reconcile durable state and operation identifiers before repeating side effects
  4. What does a valid structured model output establish?
    • That it meets a format contract, not that its meaning or scope is correct
    • That the action is authorized
    • That all values match source records
    • That downstream execution is safe
  5. Why narrow the model-dependent step in an automation?
    • It removes the need for retry handling
    • A smaller uncertain part is easier to test and contain
    • It guarantees that the model will not fail
    • It makes permissions unnecessary
  6. A workflow partially completes before a timeout. Which design helps avoid duplicate effects?
    • Randomly changing the identifier on every retry
    • Discarding all completion records
    • Durable state and stable operation identifiers
    • Using a longer model prompt
  7. What makes human approval useful for a proposed automated action?
    • A single approve button with no details
    • A notification after the action has run
    • A model-generated claim that the action is low risk
    • Enough context for the reviewer to assess the action and its scope
  8. Which outcome better measures automation quality than step count?
    • Correct completed workflows and manual recovery burden
    • Number of model calls
    • Number of automated steps executed
    • Average prompt length