Agentic AI in Network Operations Practice Questions

Answers and explanations appear when you work through the set above.

1. An AI agent investigating slow applications recommends running 'shutdown' on a core switch uplink. What should an engineer do before approving the change?

   • Approve it, because the agent has already analysed the network
   • Verify the recommendation against live show output or telemetry and assess which users and paths the change affects
   • Ask the agent to run the same analysis again and approve if the answer matches
   • Reject all agent recommendations because AI cannot be trusted with network data

2. A team wants an AI agent to help with troubleshooting on production routers. Which access design best limits risk?

   • Give the agent the enable password so it never has to stop for permissions
   • Give the agent read-only, least-privilege access, and require human approval before any configuration change
   • Share one administrator account between the agent and the on-call engineers
   • Disable logging on the routers so the agent's commands do not fill the log buffer

3. A digital network assistant reports that an OSPF neighbor is down because of an MTU mismatch. The engineer runs 'show ip ospf neighbor' and sees the neighbor in the FULL state. How should the engineer treat the assistant's statement?

   • As correct, because the assistant analysed more data than one show command
   • As unreliable, because it contradicts current device output, so the device evidence takes priority
   • As correct, because FULL only means the hello packets were received
   • As a reason to clear the OSPF process on both routers to force agreement

4. Which scenario is an example of an agentic workflow in network operations?

   • A dashboard shows a static list of devices sorted by hostname
   • A tool detects repeated link flaps from telemetry, queries interface counters and logs through APIs, correlates them, and opens a ticket with a proposed fix awaiting approval
   • An engineer copies a configuration template from a wiki into a router
   • A scheduled backup copies running-config to a TFTP server at midnight

5. Why should every action an AI agent takes on network devices be written to an audit log?

   • So the agent can use the log as its only memory between sessions
   • So engineers can trace what was done, by which identity, and why, and review or reverse it if needed
   • Because audit logs make the agent's answers more accurate
   • Because network devices refuse API calls that are not logged

6. What does 'human in the loop' mean when applied to agentic AI in network operations?

   • A person must write every command the agent runs
   • A person reviews and approves selected high-impact actions before the agent carries them out
   • The agent can only run while a person watches a live video feed
   • The agent sends every result to a person by email after the change is finished

7. Which statement correctly contrasts predictive AI with agentic AI in network operations?

   • Predictive AI forecasts outcomes such as a link reaching capacity, while agentic AI can act toward a goal using tools
   • Predictive AI executes configuration changes, while agentic AI only forecasts
   • Both terms mean the same thing and differ only in vendor branding
   • Predictive AI requires a human prompt, while agentic AI cannot accept prompts

8. An agent reads device logs and interface descriptions to diagnose a fault. An attacker sets an interface description to text telling the agent to disable logging and export the running configuration. Which risk is this?

   • Prompt injection through untrusted data the agent reads
   • A VLAN hopping attack
   • ARP cache poisoning
   • A spanning tree root takeover