Anatomy of an AI Agent for Cisco ACI: Knowledge Assessment for Network Engineers
1. Introduction
Artificial Intelligence is becoming increasingly relevant to network operations. For Cisco ACI engineers, concepts such as AI agents, RAG, tool calling, ReAct, memory, planning, and Human-in-the-Loop can be applied to practical troubleshooting and automation scenarios.
This assessment converts the core Anatomy of an Agent concepts into Cisco ACI-focused scenarios.
The objective is not simply to test whether you know AI terminology. Instead, each question connects an AI-agent concept with something a Cisco ACI engineer may encounter during day-to-day operations.
Try answering each question before looking at the answer.
2. Question 1 – Plan-and-Execute in Cisco ACI
An AI agent is asked to investigate why an application EPG cannot communicate with a database EPG.
The agent first creates a troubleshooting roadmap covering endpoint learning, EPG configuration, contracts, Bridge Domain status, faults, and interfaces before executing the investigation.
What is the primary benefit of this approach?
A. It eliminates the need for APIC
B. It allows the agent to create an explicit roadmap before acting, making a complex troubleshooting task more manageable
C. It guarantees that the configuration is correct
D. It removes the need for network engineers
Answer: B – It allows the agent to create an explicit roadmap before acting, making a complex troubleshooting task more manageable.
This is the basic idea behind Plan-and-Execute: create a structured roadmap before performing the actual work.
3. Question 2 – System Prompt for an ACI Agent
You are building an AI agent for Cisco ACI troubleshooting.
The agent should be allowed to read APIC operational information but should not make production configuration changes without approval.
Which component should define these behavioral rules?
A. Context Window
B. RAG database
C. System Prompt
D. Memory
Answer: C – System Prompt
The System Prompt defines the agent's identity, scope, responsibilities, and behavioral guidelines.
For example, an ACI agent could be instructed to:
Investigate ACI issues using approved read-only tools and request human approval before production changes.
4. Question 3 – RAG for Cisco ACI
A network organization wants its AI agent to use the company's current ACI design standards, troubleshooting guides, internal runbooks, and approved change procedures.
Which capability would be most appropriate?
A. RAG
B. Context Window
C. Iteration Limit
D. Tool Description
Answer: A – RAG
RAG, or Retrieval-Augmented Generation, allows an AI system to retrieve relevant information from external knowledge sources before generating an answer.
For an ACI environment, the knowledge base could contain:
- ACI design documents
- Internal troubleshooting guides
- Change procedures
- Network diagrams
- Operational runbooks
- Approved configuration standards
If you are new to Generative AI, you can first read my beginner-friendly guide:
Related Article: Generative AI Fundamentals Explained for Beginners
5. Question 4 – Context Window
An AI agent is troubleshooting an ACI issue and needs to consider previous tool results, APIC faults, endpoint information, configuration data, and the current conversation.
What does the context window represent?
A. The time for which APIC retains an endpoint
B. The physical memory installed in APIC
C. The span of tokens the model can consider at one time when making decisions
D. The maximum number of ACI leaf switches
Answer: C – The span of tokens the model can consider at one time when making decisions.
For network engineers, you can think of context as the information available to the AI while it is working on the current problem.
6. Question 5 – Agent Termination
An AI agent is troubleshooting an ACI connectivity problem.
Which of the following is NOT a valid termination condition for the agent loop?
A. The requested troubleshooting task has been completed
B. The maximum iteration limit has been reached
C. The agent reaches a configured confidence threshold
D. The agent receives approval to continue
Answer: D – The agent receives approval to continue.
Approval to continue does not represent a termination condition.
7. Question 6 – Detecting an Agent Loop
An AI agent is troubleshooting an ACI interface.
Its activity looks like this:
Iteration 3 Action: Check interface Ethernet1/10 Result: No useful information Iteration 4 Action: Check interface Ethernet1/10 Result: No useful information Iteration 5 Action: Check interface Ethernet1/10 Result: No useful information
What problem is the agent demonstrating?
A. Oscillation
B. Loop
C. Successful termination
D. Human-in-the-Loop
Answer: B – Loop
The agent is repeatedly performing the same action without making progress.
This is an important concept for network automation because an automation system that continuously repeats the same diagnostic operation can waste resources without improving the outcome.
8. Question 7 – Tool Description
An AI agent has access to these ACI tools:
get_aci_faults get_aci_endpoints get_aci_contracts
Why is the description field in the tool schema important?
A. It determines the physical speed of the interface
B. It helps the LLM determine when the tool should be used
C. It determines APIC CPU utilization
D. It automatically fixes configuration errors
Answer: B – It helps the LLM determine when the tool should be used.
For example:
get_aci_faults— Retrieves current ACI faults for a specified node, tenant, EPG, or object.
Clear descriptions help the agent select the appropriate tool.
9. Question 8 – Selecting the Correct ACI Tool
An engineer asks:
"Show me all active critical faults on Leaf-101."
Which tool would be most appropriate?
A. get_aci_endpoints
B. get_aci_faults
C. change_aci_contract
D. send_email
Answer: B – get_aci_faults
The agent should select the tool whose purpose matches the requested operation.
10. Question 9 – Tool Categories
Match the following ACI agent activities with the most appropriate tool category.
| ACI Activity | Tool Category |
|---|---|
| Retrieve APIC configuration | Retrieval |
| Modify an approved ACI policy | Execution |
| Request engineer approval | Communication |
| Extract structured information from raw syslog | Perception |
Answer:
- Retrieve APIC configuration → Retrieval
- Modify an approved ACI policy → Execution
- Request engineer approval → Communication
- Extract structured information from raw syslog → Perception
11. Question 10 – Perception Tool
An AI agent receives thousands of unstructured ACI syslog messages.
It needs to extract:
- Leaf ID
- Interface
- Timestamp
- Fault type
- Severity
Which tool category would be most appropriate?
A. Retrieval
B. Execution
C. Communication
D. Perception
Answer: D – Perception
Perception is appropriate when the agent needs to interpret or extract structured information from unstructured information.
12. Question 11 – ReAct Observation
An ACI AI agent follows this sequence:
Thought ↓ Action ↓ Observation
The agent calls an APIC API to retrieve endpoint information.
What is the purpose of the Observation step?
A. Generate the final blog article
B. Receive and process the result of the tool call
C. Restart APIC
D. Delete the previous action
Answer: B – Receive and process the result of the tool call.
For example:
Thought: I should check whether the endpoint is learned. Action: get_aci_endpoint() Observation: Endpoint not found.
The observation provides information that can influence the agent's next decision.
13. Question 12 – ACI Agent Memory
An AI agent has already checked:
Endpoint learning EPG membership Bridge Domain Contract Interface status
It should not repeat the same investigation unnecessarily.
Which capability helps the agent retain information about previous actions and results?
A. Memory
B. Interface policy
C. VXLAN
D. BGP
Answer: A – Memory
Memory can help an agent retain relevant information from previous actions and interactions.
For a network engineer, think of this as the agent maintaining a useful history of its troubleshooting journey.
14. Question 13 – Short-Term Investigation Context
During an ACI troubleshooting session, the agent needs to remember:
"Leaf-101 was checked and is healthy."
It also needs to remember:
"EPG-App has endpoint learning problems."
Which capability is most directly related to maintaining information required during the current investigation?
A. Working context / short-term memory
B. L3Out
C. Contract
D. Interface policy group
Answer: A – Working context / short-term memory
This information is required to make the next troubleshooting decision without repeating previous steps.
15. Question 14 – Supervisor Architecture
An enterprise wants to build an AI-based network operations platform with specialized agents:
ACI Troubleshooting Agent Security Agent Monitoring Agent Ticketing Agent
The organization wants one central agent to coordinate these specialized agents.
Which orchestration pattern is most appropriate?
A. Peer-to-peer
B. Single agent
C. Supervisor
D. Random selection
Answer: C – Supervisor
A Supervisor architecture allows a central agent to coordinate specialized worker agents.
This is particularly interesting for large network environments where different systems may require different specialized capabilities.
16. Question 15 – Human-in-the-Loop for ACI
An AI agent has determined that a production ACI contract needs to be modified.
The change could affect communication between multiple application environments.
What should the agent ideally do before making the production change?
A. Make the change immediately
B. Delete the existing contract
C. Request appropriate human approval
D. Restart the leaf switches
Answer: C – Request appropriate human approval
A safer production workflow is:
Investigate ↓ Analyze ↓ Recommend ↓ Human Approval ↓ Execute ↓ Verify
This is particularly important when AI agents are given access to production network infrastructure.
17. Question 16 – ACI Agent Safety
An AI agent identifies a possible problem with an ACI L3Out and proposes changing the configuration.
Which approach is safest for a production environment?
A. Allow unrestricted autonomous changes
B. Give the agent administrator access to everything
C. Use controlled permissions and require approval for high-impact changes
D. Disable logging to improve performance
Answer: C – Use controlled permissions and require approval for high-impact changes.
The objective should be to provide the agent with enough access to perform its task while limiting unnecessary privileges.
18. Question 17 – Agent Fails to Access APIC
An AI agent is troubleshooting Cisco ACI but its APIC API connection repeatedly fails.
What should the agent ideally do?
A. Continue calling the API indefinitely
B. Pretend that the API returned the expected result
C. Apply an approved fallback or escalate to a human
D. Change APIC configuration automatically
Answer: C – Apply an approved fallback or escalate to a human
A well-designed agent needs defined failure-handling behavior.
19. Question 18 – ACI Endpoint Investigation
An engineer asks:
"Why can't Server-A communicate with Server-B?"
The AI agent discovers:
Server-A → Endpoint learned Server-B → Endpoint learned EPG membership → Correct Bridge Domain → Operational Contract → Missing
What is the most logical next conclusion?
A. The leaf switch must be powered off
B. The missing contract should be investigated as a likely policy issue
C. The APIC cluster must be rebuilt
D. The spine switches must be replaced
Answer: B – The missing contract should be investigated as a likely policy issue.
This demonstrates how an agent can use observations from previous steps to determine what to investigate next.
20. Question 19 – RAG vs Live Operational Data
An ACI engineer asks:
"What does our company-approved procedure say about removing a leaf from production?"
Which information source would be most appropriate?
A. Generic model knowledge only
B. Organization-specific knowledge retrieved through RAG
C. Random internet content
D. Endpoint table only
Answer: B – Organization-specific knowledge retrieved through RAG
This distinction is important:
RAG → Knowledge
APIC API → Current operational state
A powerful ACI agent could potentially use both.
21. Question 20 – Complete AI Agent Workflow
Which sequence best represents a well-designed AI agent investigating a Cisco ACI incident?
A.
Change → Guess → Stop
B.
Goal → Plan → Tool → Observe → Analyze → Next Action → Verify → Stop
C.
Login → Change → Logout
D.
Prompt → Delete Fabric → Restart
Answer: B – Goal → Plan → Tool → Observe → Analyze → Next Action → Verify → Stop
This brings together the major concepts covered throughout the assessment.
22. Bonus Question – Think Like an ACI Engineer
An AI agent tells you:
"I am 98% confident that the contract is the cause of the connectivity problem."
Should the network engineer automatically implement the proposed configuration change?
A. Yes, because the confidence is 98%
B. Yes, because AI cannot make mistakes
C. No. Validate the evidence and follow the organization's change-control process
D. No, because AI agents can never troubleshoot networks
Answer: C – Validate the evidence and follow the organization's change-control process.
AI confidence should never replace engineering validation.
23. Quick Answer Key
| Question | Answer |
|---|---|
| Q1 | B |
| Q2 | C |
| Q3 | A |
| Q4 | C |
| Q5 | D |
| Q6 | B |
| Q7 | B |
| Q8 | B |
| Q9 | Matching |
| Q10 | D |
| Q11 | B |
| Q12 | A |
| Q13 | A |
| Q14 | C |
| Q15 | C |
| Q16 | C |
| Q17 | C |
| Q18 | B |
| Q19 | B |
| Q20 | B |
| Bonus | C |
24. Related Cisco ACI & AI Articles from Netterrene
If you are learning AI as a network engineer, I recommend following these articles in sequence.
AI Fundamentals
Generative AI Fundamentals Explained for Beginners
A beginner-friendly introduction to Generative AI, LLMs, context, prompting, Human-in-the-Loop, AI agents, and networking examples.
Read: Generative AI Fundamentals for Beginners
Agentic AI for Network Engineers
Agentic AI for Network Engineers: What It Actually Means for BGP, ACI, and Your NOC
This is a natural follow-up to this assessment because it explains Agentic AI using real network operations scenarios and compares agents with traditional automation.
Read: Agentic AI for Network Engineers
Reactive Automation vs Agentic AI
Reactive Automation vs Generative AI vs Agentic AI: A Decision Framework for Network Engineers
Useful for understanding where traditional automation, Generative AI, and Agentic AI fit into network operations.
Read: Reactive Automation vs Generative AI vs Agentic AI
Cisco ACI Fundamentals
Cisco ACI Explained: Concepts, Learning Prerequisites, Benefits, and Limitations
If you are new to ACI, start here before moving into AI-powered ACI automation.
Cisco ACI MoQuery
Cisco ACI MoQuery – Advanced Commands for Day-to-Day Operations
MoQuery is particularly relevant when thinking about how an AI agent could retrieve operational information from the ACI Management Information Tree.
Cisco ACI vPC
Cisco ACI vPC Explained: Architecture, Working, Configuration, Traffic Flow & Interview Questions
Useful for engineers who want to understand ACI vPC before applying AI-assisted troubleshooting concepts to it.
Cisco ACI Service Graph
Why Service Graphs Matter in Cisco ACI
A useful advanced ACI topic because service insertion introduces additional relationships that an AI troubleshooting agent may need to understand.
Read: Cisco ACI Service Graph Guide
ACI Switch Lifecycle
Cisco ACI Decommission Only vs Remove vs Secure Remove
This is another good example of where an AI agent could potentially help an engineer investigate dependencies before a fabric change.
Read: Cisco ACI Decommission vs Remove vs Secure Remove
25. Recommended Learning Path
If you are using this article for self-study, follow this sequence:
Step 1 → Learn Cisco ACI fundamentals
Step 2 → Learn Generative AI fundamentals
Step 3 → Understand AI agents
Step 4 → Understand Reactive Automation vs Generative AI vs Agentic AI
Step 5 → Learn RAG and tool calling
Step 6 → Learn ReAct and agent memory
Step 7 → Apply these concepts to ACI troubleshooting
Step 8 → Start with read-only automation
Step 9 → Introduce Human-in-the-Loop approval
Step 10 → Gradually explore controlled autonomous operations
This gives you a strong bridge between your existing Cisco ACI expertise and your new AI/Agentic AI learning journey.
26. Final Takeaway
The anatomy of an AI agent becomes much easier to understand when it is mapped to a Cisco ACI troubleshooting problem.
Plan-and-Execute → ACI troubleshooting plan
RAG → ACI documentation and internal runbooks
Tool Calling → APIC APIs and operational tools
ReAct → Think → Act → Observe
Memory → Incident investigation history
Perception → Extracting information from logs
Supervisor → Coordinating specialized network agents
Human-in-the-Loop → Production change approval
Termination → Knowing when the troubleshooting task should stop
The important lesson for network engineers is that Agentic AI doesn't require abandoning your networking fundamentals.
Instead, your existing knowledge of ACI, APIC, contracts, EPGs, Bridge Domains, L3Outs, endpoints, faults, and troubleshooting methodology becomes the foundation on which you can build AI-assisted network operations.
The future network engineer won't just understand the network. They will increasingly understand how to teach intelligent systems to work with the network safely.
No comments:
Post a Comment