Code generation agent design in Agentic AI - Model Metrics & Evaluation

Metrics & Evaluation - Code generation agent design

Which metric matters for Code Generation Agent Design and WHY

For code generation agents, the key metrics are accuracy of generated code, BLEU score or similar text similarity metrics, and functional correctness. Accuracy here means how often the generated code matches the expected solution or passes tests. BLEU score measures how close the generated code is to reference code in wording and structure. Functional correctness is the most important because code must run correctly, not just look similar.

Confusion Matrix or Equivalent Visualization

Unlike classification, code generation does not use a confusion matrix. Instead, we can think in terms of pass/fail on test cases:

Test Cases: 100
Passed: 85
Failed: 15

Pass Rate = 85/100 = 85%
Fail Rate = 15/100 = 15%
    

This pass/fail count acts like a simple confusion matrix for code correctness.

Tradeoff: Precision vs Recall (or Equivalent)

In code generation, the tradeoff is between generating code that compiles and runs (precision) and covering all requested features or requirements (recall).

If the agent generates very safe but minimal code, it has high precision (few errors) but low recall (misses features). If it tries to generate complex code covering all features, it may have higher recall but lower precision due to bugs.

Example: A code agent that always generates a simple "Hello World" program has perfect precision but poor recall for complex tasks. One that tries to generate full apps may fail often, lowering precision.

What "Good" vs "Bad" Metric Values Look Like

Good: Pass rate above 90%, BLEU score above 0.7, and generated code passes all functional tests.

Bad: Pass rate below 50%, BLEU score below 0.3, and generated code frequently fails to compile or run.

Good code generation means the agent reliably produces working code that meets requirements. Bad means frequent errors or incomplete solutions.

Common Pitfalls in Metrics

• Overfitting: Agent memorizes training code but fails on new tasks.
• Data Leakage: Test code appears in training data, inflating pass rates.
• Accuracy Paradox: High BLEU score but generated code does not run correctly.
• Ignoring Functional Tests: Relying only on text similarity without running code.

Self Check

Your code generation agent has 98% accuracy on training data but only 12% pass rate on new test cases. Is it good for production? Why or why not?

Answer: No, it is not good. The high training accuracy suggests overfitting, meaning the agent memorized training code but cannot generate correct new code. The low pass rate on test cases shows poor generalization, which is critical for production use.

Key Result

Functional correctness (pass rate) is the key metric to evaluate code generation agents, ensuring generated code runs correctly beyond text similarity.

Practice

1.

What is the main purpose of a code generation agent in AI?

easy

A. To execute code faster than a computer

B. To manually debug code written by humans

C. To automatically write code from given instructions

D. To replace all human programmers completely

Solution

1. Step 1: Understand the role of a code generation agent

   A code generation agent is designed to write code automatically based on instructions it receives.

2. Step 2: Compare options with this role

   Only To automatically write code from given instructions matches this purpose. Other options describe unrelated tasks.

3. Final Answer:

   To automatically write code from given instructions -> Option C

4. Quick Check:

   Code generation agent purpose = automatic code writing [OK]

Hint: Focus on automatic code writing, not manual or execution tasks [OK]

Common Mistakes:

• Confusing code generation with debugging
• Thinking it executes code faster
• Assuming it replaces all programmers

2.

Which of the following is the correct way to instruct a code generation agent to create a Python function named add that returns the sum of two numbers?

easy

A. Define add to subtract two numbers

B. Function add returns x minus y

C. Create add function that multiplies x and y

D. Write a function add(x, y) that returns x + y

Solution

1. Step 1: Identify the correct instruction for addition

   The instruction must specify a function named add that returns the sum (x + y).

2. Step 2: Check each option

   Write a function add(x, y) that returns x + y correctly instructs to write a function add(x, y) returning x + y. Others describe subtraction or multiplication.

3. Final Answer:

   Write a function add(x, y) that returns x + y -> Option D

4. Quick Check:

   Correct function instruction = Write a function add(x, y) that returns x + y [OK]

Hint: Look for 'returns x + y' to identify addition function [OK]

Common Mistakes:

• Choosing instructions for subtraction or multiplication
• Ignoring function name or return statement
• Confusing wording of instructions

3.

Given this instruction to a code generation agent: Write a Python function multiply that returns the product of two numbers. Which of the following code outputs is correct when calling multiply(3, 4)?

medium

A. 7

B. 12

C. 34

D. Error

Solution

1. Step 1: Understand the function's purpose

   The function multiply should return the product of two numbers, so multiply(3, 4) should return 3 * 4 = 12.

2. Step 2: Evaluate each output option

   12 is 12, which matches the expected product. Others are incorrect or errors.

3. Final Answer:

   12 -> Option B

4. Quick Check:

   3 * 4 = 12 [OK]

Hint: Multiply inputs to find correct output [OK]

Common Mistakes:

• Adding instead of multiplying
• Concatenating numbers as strings
• Assuming function causes error

4.

Consider this code generated by an agent:

def divide(x, y):
    return x / y

result = divide(10, 0)

What is the main issue with this code?

medium

A. Runtime error due to division by zero

B. Logical error returning wrong result

C. Syntax error due to missing colon

D. No issue, code runs correctly

Solution

1. Step 1: Analyze the function call

   The function divide is called with y=0, which causes division by zero.

2. Step 2: Identify the error type

   Division by zero causes a runtime error (ZeroDivisionError) in Python.

3. Final Answer:

   Runtime error due to division by zero -> Option A

4. Quick Check:

   Divide by zero causes runtime error [OK]

Hint: Check for zero in denominator to spot division errors [OK]

Common Mistakes:

• Thinking it's a syntax error
• Assuming code runs without error
• Confusing logical error with runtime error

5.

You want a code generation agent to create a Python function that filters out all negative numbers from a list and returns the positive numbers only. Which instruction will most likely produce the correct function?

hard

A. Write a function that returns only positive numbers from the list

B. Write a function that returns all numbers less than zero from the list

C. Write a function that returns the sum of all numbers in the list

D. Write a function that returns the list sorted in descending order

Solution

1. Step 1: Understand the filtering goal

   The goal is to keep only positive numbers, so the instruction must specify returning positive numbers.

2. Step 2: Evaluate each instruction

   Write a function that returns only positive numbers from the list correctly asks for a function returning only positive numbers. Others do different tasks.

3. Final Answer:

   Write a function that returns only positive numbers from the list -> Option A

4. Quick Check:

   Filter positive numbers = Write a function that returns only positive numbers from the list [OK]

Hint: Look for 'returns only positive numbers' in instruction [OK]

Common Mistakes:

• Choosing instructions that filter negatives instead
• Confusing filtering with summing or sorting
• Ignoring the word 'positive' in instruction