integration test

advancedwidespread

A software test that checks whether individual, already functioning parts of a program work together flawlessly when connected.

In plain terms

Imagine you are installing a kitchen. The tap is good, the dishwasher is good, the drain pipe is good – each was tested separately at the factory. Yet the water still flows onto the floor because the pipe diameter does not match the connector. The integration test examines exactly this moment: what happens when you connect the pieces.

Software also consists of components: one part logs the user in, another talks to the database, and a third handles payments. A significant share of errors arise not inside the individual parts but at their boundaries: one sends the amount in forints while the other expects it in fillér, or one passes the date in a different format from the one the other understands.

This test therefore does not ask 'does this small code snippet calculate correctly' but 'do the parts understand each other.' In the AI era, this has become especially important: a coding agent can write a new feature in minutes, but whether it fits the existing system is shown primarily by integration tests.

How it works

The developer selects two or more related components, such as an order handler and a database, and connects them in a real (or near-real) environment.
It writes a scenario: what data goes in on one side and what must appear on the other—for example, "if I place an order, it must appear in the database and the stock must decrease."
The test automatically runs the script and compares the actual result with the expected one.
External services that would be expensive or risky to call in production (bank, SMS sender) are often replaced with a mock version.
Tests typically run automatically again when code changes (slower ones perhaps only daily or before release), so it quickly becomes apparent if a change broke something elsewhere.

An everyday example

After an update, you open the banking app: login works, the balance is visible, but you get an error message when clicking the transfer button. A common cause of such errors is that the parts were good individually, but their connection was not – a good integration test should have caught this before release.

Who is affected

developerThis person is responsible for ensuring these tests are completed and kept up to date, and uses them to verify code written by the coding agent. → Code

entrepreneur, executiveWhen ordering software, the existence of integration tests is a good sign that the finished system will not fall apart at the first update.

everyonePost-update crashes and strange errors often stem from insufficient testing of how components interact.

A common misconception

MythIf every component test is green, the entire program works.

In realityEven individually flawless parts can fail when connected, because they may interpret the same data differently or expect steps in a different order. Only joint testing reveals this.

Limits and risks

It is slower and more expensive than unit testing because it requires launching a real database, network, or multiple services.
If it fails, it is harder to identify which part is faulty, since multiple elements work simultaneously.
It only checks what a script was written for: a green test does not prove there are no bugs, only that the tested cases work.
A service replaced by a test version may behave differently than the real one, so the test can give a false sense of security.

Similar, but not the same

unit test – Unit testing examines a single small code snippet in isolation, while integration testing checks how multiple parts work together.
end-to-end test – End-to-end testing runs through the entire system from the user's perspective, while integration testing covers only a few connected parts.
regression test – Regression testing monitors whether a modification has broken previously working functionality, and this can happen at any level, not just at the interfaces between parts.

Related concepts

Broader

software testing, quality assurance, automated testing

Narrower

API-teszt, contract test

On the edge

unit test, end-to-end test, regression test, continuous integration, tesztpiramis

A brief history

Integration testing, alongside modular programming, became standard practice in software development in the 1970s. Glenford Myers' 1979 book, The Art of Software Testing, already detailed how to assemble and test modules: step-by-step or all at once, top-down or bottom-up.

Where things stand

By the end of 2026, this is standard practice in professional software development; tests typically run automatically when code is modified. Its importance will grow further because it is one of the main tools for verifying code written by coding agents.

Status: 2026 November

Also known as

integration test · integration testing · integration test · integration testing · integration trial

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.