Functional tests check whether an AI system performs the task it is supposed to perform.
A functional test checks whether an AI system produces the required result for a defined task.
It tests behavior that a user or another system can observe. The test can check the answer, format, tool action, refusal, or handling of missing information.
Think of it like a driving test.
The test gives the driver a known situation and checks the observable result. A functional test checks what the AI system does for a known input.
Input: “How do I reset my password?”
System: returns the approved reset steps.
Output: a helpful answer with no invented policy.
Input: a request to create a calendar event.
System: calls the calendar tool with the required fields.
Output: a valid event or a clear error.
Functional test: checks whether a task works.
Quality score: measures how good the result is along a chosen dimension.
Functional test: checks observable system behavior.
Unit test: checks a smaller code component in isolation.
Functional test: runs against a planned case.
Production monitoring: watches live behavior after release.
Explore next
A functional test checks whether an AI system does the required job for a defined case.