Self-improving agents in Agentic AI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is a key characteristic of a self-improving agent?

Which of the following best describes a self-improving agent in AI?

A An agent that randomly changes its behavior without evaluation.

B An agent that requires manual updates from developers to improve.

C An agent that can modify its own code or parameters to improve performance over time.

D An agent that only follows fixed rules without any learning capability.

💡 Hint

Think about what makes an agent improve itself without external help.

Self-improving agents can autonomously adjust their own code or parameters to enhance their performance, unlike fixed or manually updated agents.

❓ Model Choice

intermediate

Choosing a model architecture for a self-improving agent

You want to build a self-improving agent that learns from its past decisions and adapts its strategy. Which model architecture is most suitable?

A A reinforcement learning model with policy gradient methods.

B A simple linear regression model without feedback.

C A clustering algorithm that groups data points.

D A static decision tree with fixed rules.

💡 Hint

Consider models that learn from interaction and improve policies over time.

Reinforcement learning models with policy gradients allow agents to learn from feedback and improve their decision-making policies, which is essential for self-improvement.

❓ Metrics

advanced

Evaluating improvement in a self-improving agent

Which metric is most appropriate to measure the improvement of a self-improving agent over multiple training episodes?

A Average cumulative reward per episode.

B Mean squared error on a fixed test set.

C Number of parameters in the model.

D Training time per epoch.

💡 Hint

Think about a metric that reflects how well the agent performs its task over time.

Average cumulative reward per episode directly measures how well the agent is performing and improving in its environment over time.

🔧 Debug

advanced

Debugging a self-improving agent stuck in local optimum

An agent using reinforcement learning is stuck with poor performance and does not improve despite training. What is the most likely cause?

A The agent's model architecture is too complex.

B The agent's learning rate is too high, causing unstable updates.

C The reward function is too sparse, providing no feedback.

D The agent's exploration rate is too low, causing it to exploit suboptimal actions.

💡 Hint

Consider why the agent might not try new actions to find better solutions.

If the exploration rate is too low, the agent mostly exploits known actions and fails to discover better strategies, causing it to get stuck in a local optimum.

❓ Predict Output

expert

Output of a self-modifying agent code snippet

Consider this Python code simulating a simple self-improving agent that updates its parameter to maximize reward. What is the output after running it?

Agentic AI

class SelfImprovingAgent:
    def __init__(self):
        self.param = 0
    def reward(self):
        return -(self.param - 5) ** 2 + 10
    def improve(self):
        best_param = self.param
        best_reward = self.reward()
        for delta in [-1, 1]:
            candidate = self.param + delta
            candidate_reward = -(candidate - 5) ** 2 + 10
            if candidate_reward > best_reward:
                best_param = candidate
                best_reward = candidate_reward
        self.param = best_param
agent = SelfImprovingAgent()
for _ in range(3):
    agent.improve()
print(agent.param)

A 4

B 3

C 5

D 2

💡 Hint

Trace the parameter updates step by step to see where it converges after 3 improvements.

The agent starts at param=0 (reward=-15). Each improve step moves to the neighbor with higher reward: 0→1 (-6 > -15), 1→2 (1 > -6), 2→3 (6 > 1). After 3 steps, param=3.