Test cases for tool-using agents in Agentic AI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the main purpose of tool-using agents in AI?

Tool-using agents in AI are designed to:

A Learn from data without any external help

B Replace human decision-making completely without any tools

C Use external tools or APIs to enhance their problem-solving abilities

D Operate only within a fixed environment without interaction

💡 Hint

Think about how agents can improve their capabilities beyond their own programming.

Tool-using agents leverage external tools or APIs to perform tasks they cannot do alone, enhancing their problem-solving skills.

❓ Predict Output

intermediate

What output does this tool-using agent produce?

Consider this simplified Python code where an agent uses a calculator tool to add numbers:

Agentic AI

class CalculatorTool:
    def add(self, x, y):
        return x + y

class Agent:
    def __init__(self, tool):
        self.tool = tool
    def compute_sum(self, a, b):
        return self.tool.add(a, b)

calc = CalculatorTool()
agent = Agent(calc)
result = agent.compute_sum(5, 7)
print(result)

A None

B 57

C TypeError

D 12

💡 Hint

Check what the add method returns and how compute_sum uses it.

The add method returns the sum of 5 and 7, which is 12, so the agent prints 12.

❓ Model Choice

advanced

Which model architecture best suits a tool-using agent that must handle text and images?

You want to build a tool-using agent that can understand text commands and analyze images to decide which tool to use. Which model architecture is best?

A A multimodal transformer model that processes both text and images

B A simple feedforward neural network with only text input

C A convolutional neural network (CNN) for images only

D A recurrent neural network (RNN) for sequential text data only

💡 Hint

Think about models that can handle multiple types of data at once.

Multimodal transformers can process both text and images, making them ideal for agents that use multiple input types to decide tool usage.

❓ Metrics

advanced

Which metric best evaluates a tool-using agent's success in completing tasks?

You have an agent that uses external tools to complete tasks. Which metric best measures how well it completes these tasks?

A Task completion rate (percentage of tasks successfully finished)

B Loss value during training

C Accuracy of the agent's internal neural network weights

D Number of tools available to the agent

💡 Hint

Focus on the agent's real-world performance, not internal training stats.

Task completion rate directly measures how many tasks the agent finishes successfully, reflecting its practical effectiveness.

🔧 Debug

expert

Why does this tool-using agent code raise an AttributeError?

Examine the code below where an agent tries to use a tool but fails:

Agentic AI

class Tool:
    def execute(self, command):
        return f"Executed {command}"

class Agent:
    def __init__(self):
        self.tool = None
    def use_tool(self, cmd):
        return self.tool.execute(cmd)

agent = Agent()
output = agent.use_tool('run')
print(output)

A Because <code>Agent</code> class is missing an <code>execute</code> method

B Because <code>self.tool</code> is None and has no <code>execute</code> method

C Because <code>use_tool</code> method is missing a return statement

D Because <code>execute</code> method is misspelled

💡 Hint

Check what self.tool is when use_tool is called.

The agent's tool attribute is None, so calling execute on None raises an AttributeError.