Skip to main content
  1. Public practice questions
  2. AB-620 Practice Exams | AI Agent Builder Associate
  3. AB-620 Practice Exam 2

Question 17 — AB-620 Practice Exam 2

A Copilot Studio agent contains 33 available actions. Several tools have similar descriptions, and evaluation results show that customer requests are frequently being routed to the wrong action. The team initially planned to wait until the agent reached 40 actions before reconsidering the design. What should the team do? (Select one answer)

  1. Keep the architecture unchanged because routing degradation begins only after 40 actions

  2. Remove connected-agent support because fewer than 40 actions cannot justify specialization

  3. Add more actions until evaluation results establish a larger routing sample

  4. Consider separating responsibilities now because observed routing errors are more important than a fixed action count

Correct answers

  1. Consider separating responsibilities now because observed routing errors are more important than a fixed action count

Explanations

Correct:

Consider separating responsibilities now because observed routing errors are more important than a fixed action count

Routing degradation is a practical signal for reconsidering the architecture. The commonly used range of approximately 30–40 available actions is a guideline rather than a mandatory boundary. Similar tool descriptions can cause routing problems even before that range is exceeded, so evaluation results should guide the decision.

Wrong:

Keep the architecture unchanged because routing degradation begins only after 40 actions

There is no fixed requirement that routing problems begin only after the fortieth action. Action similarity and actual evaluation performance can expose orchestration problems earlier.

Remove connected-agent support because fewer than 40 actions cannot justify specialization

Agent separation can be appropriate below the approximate action-count range when routing quality or another structural boundary already supports the change.

Add more actions until evaluation results establish a larger routing sample

Adding actions can increase orchestration complexity. Existing evaluation evidence showing incorrect routing is already a relevant architectural signal.