AI testing agents are AI software that write and run tests for an app by themselves, starting from a goal given in plain language. They work inside a developer's code editor or on a test grid, and they test APIs, websites and mobile apps. An agent is software that uses a large language model, the kind of AI behind chatbots, to plan steps and take actions on its own. This page covers one narrow use of AI agents.
How does an agent test? A tester gives it a goal, such as "Test the leave application". The agent opens the app, reads each screen, and works out which actions are possible. It decides which cases matter: the normal case, empty fields, wrong dates, and limits such as the most leave an employee may take. It writes each case as a test, runs it and checks the result. When a test fails, it records the steps, a screenshot and the error. The tests it writes are saved, so that they can run again on every later change.
Imagine a software start-up in Coimbatore that sells HR software to companies in India and abroad. It adds a new screen where employees apply for leave. A tester gives the agent the goal and a test login. In 20 minutes, the agent writes 18 test cases and runs them on a cloud test grid, on Chrome, Safari and two Android phones. Sixteen pass. Two fail. The app accepts a leave request that ends before it starts, and on one Android phone the date picker does not open. The tester reads both reports, confirms that they are real bugs, and files them. She also deletes one of the 16 passing tests, because it checked the wrong thing. It passed whenever the screen showed any message at all, even an error message.
That last step shows why a person still reviews the agent's work. An agent can write a test that passes for the wrong reason. It can also miss a case that a person who knows the business would test, such as a leave rule that applies only in one state. And each run costs money, because the agent calls the language model many times.
AI testing agents are the newest stage of test automation tools, after scripts, plain English and model-based tests. They are a special case of AI coding agents, which write and run code of any kind. How do they differ from a general AI coding agent? A coding agent may write tests for its own change, from inside the code. A testing agent tests the finished app from outside, the way a user would.