Building custom tools in Agentic AI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding Tool Integration in Agentic AI

Which of the following best describes the primary purpose of building custom tools in an agentic AI system?

A To replace the AI's core reasoning engine with simpler rules

B To increase the AI's training dataset size automatically

C To reduce the AI model's size for faster inference

D To enable the AI to perform specific tasks by interacting with external systems or APIs

💡 Hint

Think about why an AI would need to connect to outside services or perform actions beyond just generating text.

Custom tools allow agentic AI to extend its capabilities by interacting with external APIs or systems, enabling it to perform tasks like fetching data, controlling devices, or executing commands.

❓ Predict Output

intermediate

Output of a Custom Tool Invocation

Given the following Python code snippet for a custom tool in an agentic AI framework, what will be the output when the tool is called with input '5'?

Agentic AI

def multiply_by_two(input_value):
    return f"Result: {int(input_value) * 2}"

output = multiply_by_two('5')
print(output)

A Result: 10

B Result: 25

C Result: 5

D TypeError

💡 Hint

Check how the input string is converted and multiplied.

The input '5' is converted to integer 5, then multiplied by 2, resulting in 10. The function returns the string 'Result: 10'.

❓ Hyperparameter

advanced

Choosing Timeout for External Tool Calls

When building a custom tool that calls an external API within an agentic AI system, which timeout setting is most appropriate to balance responsiveness and reliability?

A Set timeout to 0.1 seconds to ensure the AI never waits long

B Set timeout to 5 seconds to allow reasonable wait but avoid long delays

C Set timeout to 30 seconds to allow slow responses without blocking

D Do not set a timeout to wait indefinitely for the API response

💡 Hint

Consider user experience and network variability.

A 5-second timeout balances waiting long enough for typical API responses while preventing the AI from hanging too long, maintaining responsiveness.

❓ Metrics

advanced

Evaluating Custom Tool Effectiveness

Which metric is most suitable to evaluate the success of a custom tool integrated into an agentic AI that performs data retrieval tasks?

A Accuracy of the AI's language generation

B Latency of the tool's response time

C Precision and recall of the retrieved data matching user queries

D Number of parameters in the AI model

💡 Hint

Think about how well the tool returns relevant information.

Precision and recall measure how accurately and completely the tool retrieves relevant data, which is key for data retrieval tasks.

🔧 Debug

expert

Debugging a Custom Tool Integration Failure

Consider this Python code snippet for a custom tool in an agentic AI system that queries a weather API. The tool always returns 'None' instead of the expected temperature string. What is the most likely cause?

import requests

def get_temperature(city):
    response = requests.get(f"https://api.weather.com/temp?city={city}")
    if response.status_code == 200:
        data = response.json()
        return data['temperature']

print(get_temperature('London'))

A The API URL is incorrect or missing required parameters, causing no valid data returned

B The function does not convert the temperature to string before returning

C The requests library is not imported correctly

D The function should use POST instead of GET method

💡 Hint

Check if the API URL and parameters match the expected API specification.

If the API URL or parameters are wrong, the API may return a non-200 status or empty data, causing the function to return None.