Agent architecture (observe, think, act) in Prompt Engineering / GenAI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding the 'Observe' step in agent architecture

In the agent architecture model, what is the primary role of the 'Observe' step?

A To decide the best action based on past experiences

B To collect data from the environment through sensors or inputs

C To execute actions that change the environment

D To store the results of previous actions for future use

💡 Hint

Think about how an agent first learns about its surroundings.

The 'Observe' step is when the agent gathers information from its environment using sensors or inputs. This data is essential before the agent can think or act.

❓ Predict Output

intermediate

Output of agent decision-making code snippet

What will be the output of the following code simulating an agent's 'Think' step?

Prompt Engineering / GenAI

environment_data = {'temperature': 30, 'light': 'bright'}

# Agent thinks based on temperature
if environment_data['temperature'] > 25:
    action = 'turn_on_cooler'
else:
    action = 'do_nothing'

print(action)

A turn_on_cooler

B do_nothing

C KeyError

D SyntaxError

💡 Hint

Check the temperature value and the condition in the if statement.

The temperature is 30, which is greater than 25, so the agent decides to 'turn_on_cooler'.

❓ Model Choice

advanced

Choosing the best model for the 'Act' step in an agent

Which type of model is most suitable for the 'Act' step where an agent must perform actions in a continuous environment?

A Linear regression model

B Unsupervised clustering model

C Autoencoder neural network

D Reinforcement learning policy network

💡 Hint

Consider models that learn to make decisions based on rewards.

Reinforcement learning policy networks are designed to select actions in environments to maximize rewards, making them ideal for the 'Act' step.

❓ Metrics

advanced

Evaluating agent performance metrics

Which metric best measures how well an agent balances exploring new actions and exploiting known good actions during learning?

A Exploration rate (epsilon) in epsilon-greedy policy

B Precision of classification

C Mean squared error of predictions

D Cumulative reward over time

💡 Hint

Think about the parameter controlling randomness in action choice.

The exploration rate (epsilon) controls how often the agent tries new actions versus exploiting known ones, directly measuring this balance.

🔧 Debug

expert

Debugging an agent's 'Think' step causing wrong action

Given the code below, why does the agent always choose 'do_nothing' even when the temperature is 30?

Prompt Engineering / GenAI

environment_data = {'temperature': 30}

# Agent thinks and decides action
if environment_data['temperature'] < 25:
    action = 'turn_on_cooler'
else:
    action = 'do_nothing'

print(action)

A The dictionary key 'temperature' is misspelled causing a KeyError

B The agent lacks an 'Observe' step to get temperature data

C The condition is reversed; it should be '>' instead of '<' to turn on the cooler when hot

D The print statement is outside the if-else block causing no output

💡 Hint

Check the logic condition comparing temperature to 25.

The condition uses '< 25' which means the cooler turns on only if temperature is less than 25. Since temperature is 30, it chooses 'do_nothing'. Reversing the condition fixes this.