The Test Pyramid & Quadrants — What to Automate First

The idea, in one line

Catch each bug at the cheapest place that can catch it. That's the whole rule. The picture that explains it has a proper name interviewers use — the test pyramid — so it's worth knowing.

Why that shape?

The shape is really about cost. Here's what each layer costs you:

• Unit tests run in milliseconds, point straight at the broken function, and rarely give false alarms.
• End-to-end tests are slow, break easily, and when they fail they only tell you something broke somewhere.
• So you write lots of the cheap ones and very few of the expensive ones.

See it work

A healthy test suite mirrors the pyramid: many unit, some service, few end-to-end. Here's a quick way to check the balance.

# Count of tests at each layer
suite = {"unit": 420, "service": 60, "e2e": 12}

total = sum(suite.values())
unit_share = suite["unit"] / total
print(f"unit share: {unit_share:.0%}")   # unit share: 85%

# Warning sign: if e2e outnumbers unit, the pyramid is upside down
assert suite["unit"] > suite["e2e"], "top-heavy: push tests down"

# Same bug, cheapest layer that catches it wins:
catches = {"discount math": "unit", "api contract": "service", "checkout flow": "e2e"}
for bug, layer in catches.items():
    print(f"{bug:14} -> test at: {layer}")

Read it top to bottom: you counted your tests, checked the shape wasn't upside down, then matched each kind of bug to the cheapest layer that can catch it.

Advanced — the trap and the wider map

The classic mistake is the ice-cream cone: a fat layer of screen tests balancing on almost no unit tests. It feels thorough, but it's slow and flaky — and it's exactly the shape a manual tester builds by default. If your top layer is bigger than your bottom, that's the warning.

There's also a companion map called the testing quadrants. It sorts checks by two questions: does this check help the team build or critique the finished product, and is it technical or business-facing? Its main lesson: not everything belongs in automation — some checks are still done by a human.

Grounded in Martin Fowler's 'The Practical Test Pyramid' (martinfowler.com)