Writing Failing Tests First, Then Asking AI to Implement

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-ai-coding-test-first-r12a1-creators

1. What advantage does a failing test give you when prompting AI to implement code?

   1. It looks more professional in the chat
   2. The success criterion is unambiguous and runnable
   3. It makes the model respond faster
   4. It removes the need for a code review

2. Which instruction belongs in a test-as-spec prompt?

   1. 'Modify the test until it passes.'
   2. 'Implement the minimum code to make this test pass; do not modify the test.'
   3. 'Add ten new features.'
   4. 'Delete the test if it bothers you.'

3. What does 'test-gaming' look like with AI-generated code?

   1. AI hardcodes return values to satisfy your one test
   2. AI writes documentation
   3. AI styles the code
   4. AI commits to git

4. How can you defend against test-gaming?

   1. Add a second, harder test before trusting the implementation
   2. Run only the first test forever
   3. Hide the test from the AI
   4. Disable assertions

5. Which task is AI well suited to once it has a precise failing test?

   1. Implementing the minimum code to make the test pass
   2. Choosing your team's roadmap
   3. Renegotiating contracts
   4. Recruiting interns

6. After a base test passes, how can AI extend the test suite usefully?

   1. Suggest additional edge-case tests once the base test exists
   2. Delete the original test
   3. Rename all variables to single letters
   4. Output only emojis

7. What is AI unable to decide for you, even with a great test?

   1. Whether the behavior described is correct for your domain
   2. How to syntax-check Python
   3. How to print a string
   4. How to declare a variable

8. Why is a test-as-spec less ambiguous than a prose request?

   1. Tests are written in a polite tone
   2. A test names exact inputs and expected outputs the code must satisfy
   3. Tests come with marketing copy
   4. Tests are always shorter than prose

9. A test passes after AI's implementation, but the function returns the same value for every input. What likely happened?

   1. AI hardcoded a value matching that single test case
   2. The compiler optimized away the function
   3. Tests are inherently broken
   4. Hardware fault

10. Why is 'do not add features the test does not require' a useful instruction?

    1. It keeps scope tight and avoids unintended changes
    2. It bans the use of variables
    3. It forces a specific framework
    4. It causes compile errors

11. Which workflow embodies test-first AI use?

    1. Write code, then ask AI for tests after the fact
    2. Write a failing test, hand it to AI, run the suite, add a harder test
    3. Skip tests; ask AI for production code only
    4. Have AI write tests that pass without any code

12. What's the trust value of a test the engineer wrote themselves?

    1. It encodes the engineer's intent, which the AI must meet
    2. It is automatically slower
    3. It must be deleted before commit
    4. It always uses random data

13. Why is 'test passes for the wrong reason' a real risk?

    1. A passing assertion can hide flawed logic that just happens to satisfy the check
    2. All passing tests are by definition correct
    3. Pass means deploy
    4. Pass and correctness are identical

14. Which is a healthy follow-up after AI's implementation passes the first test?

    1. Add adversarial inputs and edge-case tests
    2. Stop testing forever
    3. Remove the original test
    4. Push to production immediately

15. Why does test-first pair especially well with AI?

    1. It gives AI an executable definition of done it cannot misinterpret
    2. AI refuses to work without tests
    3. Tests change the model temperature
    4. Tests reduce token cost to zero