Intermediate result handling in Agentic AI - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

What is the output of this agentic AI code snippet?

Consider an agentic AI system that processes tasks in steps and stores intermediate results in a dictionary. What will be the final content of the results dictionary after running this code?

Agentic AI

results = {}
for step in range(3):
    results[f'step_{step}'] = step * 2
results['final'] = sum(results.values())
print(results)

A {'step_0': 0, 'step_1': 2, 'step_2': 4, 'final': 6}

B {'step_0': 0, 'step_1': 1, 'step_2': 2, 'final': 3}

C {'step_0': 0, 'step_1': 2, 'step_2': 4, 'final': 0}

D {'step_0': 1, 'step_1': 2, 'step_2': 3, 'final': 6}

💡 Hint

Remember that each step multiplies the step number by 2 before storing.

Each step key stores step * 2: step_0=0, step_1=2, step_2=4. The final key sums these values: 0+2+4=6.

❓ Model Choice

intermediate

Which model type best handles intermediate result caching in agentic AI?

You want an AI model that can store and reuse intermediate results during multi-step reasoning to improve efficiency. Which model architecture is best suited for this?

A Recurrent neural network without state preservation

B Simple feedforward neural network without memory

C Convolutional neural network for image processing

D Transformer with attention layers that cache key-value pairs

💡 Hint

Think about models that remember past information efficiently.

Transformers use attention with cached key-value pairs to reuse intermediate computations, ideal for multi-step tasks.

❓ Hyperparameter

advanced

Which hyperparameter adjustment improves intermediate result stability in iterative agentic AI?

In an iterative agentic AI process, which hyperparameter change helps reduce fluctuations in intermediate outputs across iterations?

A Use gradient clipping to limit update size

B Add dropout with high rate

C Increase learning rate significantly

D Remove batch normalization layers

💡 Hint

Think about controlling how much the model changes per step.

Gradient clipping limits update size, stabilizing intermediate results during iterative updates.

❓ Metrics

advanced

Which metric best evaluates the quality of intermediate results in agentic AI?

You want to measure how accurate intermediate results are during a multi-step agentic AI task. Which metric is most appropriate?

A Number of parameters in the model

B Model training loss on final output only

C Mean Squared Error (MSE) between predicted and true intermediate values

D Inference time per step

💡 Hint

Focus on measuring difference between predicted and actual intermediate values.

MSE directly measures the average squared difference between predicted and true intermediate results.

🔧 Debug

expert

Why does this agentic AI code fail to update intermediate results correctly?

Given this code snippet, why does the intermediate result dictionary not update as expected after each iteration? results = {} for i in range(3): temp = results temp['step'] = i print(results)

A The variable 'temp' is a reference to 'results', so updates affect 'results' correctly.

B The dictionary key 'step' is overwritten each iteration, so only the last value remains.

C The loop does not run because range(3) is empty.

D 'temp' is a copy of 'results', so changes to 'temp' do not affect 'results'.

💡 Hint

Check how dictionary keys are updated inside the loop.

Each iteration overwrites the same key 'step', so only the last value (2) remains in results.