Write the Test First and Tell Claude to Make It Pass — TDD with AI

Tests are the most precise way to tell Claude what you want. Instead of describing a feature in English and hoping Claude interprets it correctly, write a test that defines the exact behavior — then let Claude figure out the implementation.

> I just wrote a failing test in tests/Feature/DiscountTest.php.
> Read it, then write the code to make it pass. Run the test to verify.

Claude reads your test, understands the expected inputs, outputs, edge cases, and error conditions, then writes exactly the code needed to satisfy those assertions — nothing more, nothing less.

This works because tests are unambiguous specifications. Instead of "add a discount feature," your test says:

it('applies a percentage discount to the order total', function () {
    $order = Order::factory()->create(['subtotal' => 10000]);
    $discount = new PercentageDiscount(15);

    $result = $discount->apply($order);

    expect($result->total)->toBe(8500);
    expect($result->discount_amount)->toBe(1500);
});

Claude now knows exactly what class to create, what method signature to use, what the return type looks like, and what the math should be. No ambiguity.

The TDD loop with Claude looks like this:

1. You write a failing test that describes the next piece of behavior
2. Claude reads the test and writes the minimum code to make it pass
3. Claude runs the test to verify it passes
4. You write the next test — repeat

This approach gives you several advantages:

• Precision — the test is the spec, no room for misinterpretation
• Built-in verification — Claude proves its implementation works before moving on
• No over-engineering — Claude writes only what the test requires
• Instant test coverage — every feature ships with tests already written

Write the test, hand Claude the red bar, and let it turn it green.

via Claude Code