code coverage

advancedwidespread

Code coverage shows in percentage how much of a program's code was actually executed by automated tests.

In plain terms

Imagine visiting an apartment before buying it. If you only enter seven out of ten rooms, your visit's 'coverage' is 70 percent: you know nothing about the remaining three rooms. Code coverage measures the same for software: what percentage of the program the tests have 'entered'.

Developers write automated tests to verify whether a program works correctly. Code coverage shows which parts were not touched by any test. These are blind spots: errors can remain there unnoticed.

An important distinction is that the number only indicates what was run, not whether it was thoroughly checked. Someone who entered the room may not have noticed the mold behind the cabinet.

In the AI era, the metric has taken on a new role: when a coding agent writes code and tests, coverage is one quick feedback on how well it tested its own work.

How it works

A measuring tool marks every line and branch of the program so it can later track which one ran.
Automated tests are launched, while the measuring tool records which lines and decision branches the program traversed.
At the end, it calculates the ratio: for example, if 820 out of 1000 executable lines ran, the line coverage is 82 percent. Empty lines and comments are not included.
The report uses colors to show what was missed: typically green for tested parts and red for never-run sections.
Many teams set a threshold, for example 80 percent: if a modification drops the metric below this, the system will not allow merging.

An everyday example

Your company orders a new customer management system, and the supplier reassures you that "our test coverage is 90 percent." This means that the tests ran through nine-tenths of the code; it does not mean that the program is 90 percent error-free. It is worth asking what is in the missing ten percent.

Who is affected

developerPart of your daily toolkit: it shows where tests are missing and also lets you check the code written by a coding agent. → Code

entrepreneur, executiveWhen ordering or receiving software, quality is often verified with this number, so you need to know what it means and what it does not.

industrial worker, engineerIn safety-critical areas such as aviation and the automotive industry, standards mandate what level of coverage must be verified. → Industry

A common misconception

Myth100 percent code coverage means the program is error-free.

In realityThe metric only indicates that the code ran during tests, not that the results were properly verified. Even with full coverage, a serious bug may remain, for instance if no one considered a critical scenario and no code was written for it.

Limits and risks

The figure is easy to game: one can write a test that runs the code but checks nothing about it.
It does not show missing features: what was never written cannot be covered.
If a percentage becomes a target in itself, the team starts chasing the metric instead of real quality, and achieving the last few percentage points can be disproportionately expensive.
Coding agents quickly generate tests to increase coverage, but humans should still review their value, as high numbers can mask weak tests.

Similar, but not the same

test coverage – In everyday usage, it is often a synonym for code coverage. In a stricter sense, it is a broader concept: it examines what proportion of requirements and use cases have tests, not just whether code lines have been executed.
mutation testing – It intentionally inserts errors into the code and measures whether the tests detect them, thus examining the power of the tests, not just the affected lines.
code quality – Code quality encompasses readability, maintainability, and security, while coverage is only one metric of this.

Related concepts

Broader

software testing, quality assurance, szoftvermetrika

Narrower

line coverage, branch coverage, function coverage, condition coverage

On the edge

unit test, automated testing, continuous integration, mutation testing, coding agent

A brief history

Code coverage is one of the oldest methods of systematic software testing: its first published description is linked to a 1963 article by Miller and Maloney, which appeared in the journal Communications of the ACM.

Where things stand

At the end of 2026, code coverage measurement is a standard part of software development, with most developer tools and automated checking pipelines offering it built in. As coding agents spread, machines increasingly write the tests too, which makes high coverage cheaper but the bare number less reliable.

Status: 2026 November

In the news

Appears in 25 Agent Age stories; the curve shows weekly occurrence in the news magazine.

AI liability in insurance: AI law firm Crosby buys malpractice cover as risk migrates to cyber insuranceNov 4, 2026 · Money
Aon and SAGE to use AI to price crop insurance more preciselyOct 29, 2026 · Money
Anthropic flags insurance gaps and copyright risk after $1.5bn settlementOct 27, 2026 · Jog
AI liability becomes more pressing as insurers and companies rethink coverageOct 26, 2026 · Money
Rockwell Automation launches FactoryTalk Conductor to orchestrate AI across industrial automationOct 26, 2026 · Ipar
Patients and experts oppose AI-based insurer customer service and prior authorizationOct 20, 2026 · Health
Meta Advertising Week: which AI ad tools are available now and which are in testingOct 19, 2026 · Marketing
Anthropic warns AI is months from self-replicating on smart devicesOct 13, 2026 · Agent Age

Also known as

code coverage · code coverage · test coverage · coverage metric · code coverage · test coverage · coverage · line coverage

Sources

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