Tests as Prompts — an Unexpected Superpower

A test says: when I give the function this input, I expect exactly this output. That is the most precise way to describe behavior that exists. If you paste a failing test and ask the AI to make it pass, you get amazingly focused results.

The red-green-refactor loop, with AI

1. 1Write a test that fails (red)
2. 2Ask the AI to write the function to make it pass (green)
3. 3Check the code works and is readable
4. 4Ask the AI to clean it up without breaking tests (refactor)

A concrete example in Python

# Step 1: write the tests first
def test_add_positive():
    assert add(2, 3) == 5

def test_add_negative():
    assert add(-1, -1) == -2

def test_add_zero():
    assert add(0, 42) == 42

# Step 2: paste the tests into your AI chat and say:
#   "Write the `add` function so all three tests pass."

# Step 3: the AI almost always produces:
def add(a, b):
    return a + b

# Step 4: run pytest. If green, ship it.

Ask the AI to help write tests too

You do not need to write every test by hand. Give the AI the function signature, ask for test cases including edge cases, then review the list. You will often find a case you had not considered — that is the point.

Fast frameworks to know

“Tests are the only prompt that runs.”

The big idea: a test is a prompt the machine can verify. Writing tests first turns every AI coding session into a tight loop: target, generate, check, done.