Anatomy of an AI Agent : Knowledge Assessment

 

Question 1

In the plan-and-execute architecture, what is the key benefit of separating planning from execution?

• It reduces API costs
• It allows the agent to create an explicit roadmap before acting, making complex tasks more manageable
• It eliminates the need for tools
• It makes the agent respond faster

Ans - It allows the agent to create an explicit roadmap before acting, making complex tasks more manageable

Question 2

What is the primary role of the system prompt in an agent's configuration?

• To reduce API costs by limiting token usage
• To define the agent's identity, scope, and behavioral guidelines
• To specify which LLM model to use
• To store conversation history

Ans - To define the agent's identity, scope, and behavioral guidelines

 

Question 3

What is the primary benefit of RAG (Retrieval-Augmented Generation) for an agent?

• It makes the agent respond faster
• It allows the agent to access current, organization-specific knowledge
• It reduces the cost of API calls
• It eliminates the need for working memory

Ans - It allows the agent to access current, organization-specific knowledge

 

Question 4

What does the "context window" represent in agent architecture?

• A graphical interface that displays the agent's current status and activity
• The period of time during which the agent remains active and responsive
• The span of tokens the model can consider at once when making decisions
• A rule-based workflow that defines the steps an agent follows sequentially

Ans - The span of tokens the model can consider at once when making decisions

 

Question 5

Which of the following is NOT a valid termination condition for an agent loop?

• The agent has completed the requested task
• The maximum iteration limit has been reached
• The agent's confidence level drops below a threshold
• The agent receives approval to continue

Ans - The agent receives approval to continue

 

Question 6

An agent is troubleshooting a network issue. Review this trace excerpt:

{
  "iteration": 3,
  "thought": "I should check the router logs",
  "action": "get_router_logs",
  "observation": "Connection timed out"
}
{
  "iteration": 4,
  "thought": "I should check the router logs",
  "action": "get_router_logs",
  "observation": "Connection timed out"
}
{
  "iteration": 5,
  "thought": "I should check the router logs",
  "action": "get_router_logs",
  "observation": "Connection timed out"
}

What stuck state pattern is this agent exhibiting, and what mechanism would help prevent it?

• Oscillation - implement fallback strategies
• Loop - implement iteration limits and action history tracking
• Dead end - add more tools to the agent
• Hallucination - improve tool descriptions

Ans - Dead end - add more tools to the agent

 

Question 7

Why is the "description" field in a tool schema critically important?

• It determines how fast the tool executes
• It tells the LLM when to use this tool versus other tools
• Custom C++ implementation for performance
• It sets the default input values

Answer – It tells the LLM when to use this tool versus other tools

 

Question 8

Why is the "description" field in a tool schema critically important?

• It determines how fast the tool executes
• It tells the LLM when to use this tool versus other tools
• Custom C++ implementation for performance
• It sets the default input values

Ans - It tells the LLM when to use this tool versus other tools

 

Question 9

Match each tool to its correct category.

 

Ans

 

Question 10

An agent needs to extract structured data from unstructured syslog entries. Which tool category would be most appropriate?

• Retrieval
• Execution
• Communication
• Perception

Ans - Perception

 

Question 11

In the ReAct pattern, what is the purpose of the "Observation" step?

• To generate a response for the user
• To receive and process the result of a tool call
• To decide which tool to use next
• To summarize the conversation history

Ans - To receive and process the result of a tool call

 

Match each memory type to its appropriate use case.

Ans

Question 12

A financial services company needs to build an AI system to handle customer requests.

• Research customer account information
• Analyze transaction patterns for fraud detection
• Execute approved actions on accounts
• Generate audit reports

Each function requires specialized tools and domain expertise. The compliance team requires a clear audit trail of all decisions and a single point of accountability.

Which orchestration pattern would be most appropriate?

• Peer-to-peer - agents can collaborate like a team of specialists
• Hierarchical - because there are multiple levels of management
• Supervisor - add more workers to the central agent
• Single agent - because the tasks are simple enough for one agent

 Answer -Supervisor - add more workers to the central agent