Glossary

Acceptance criteria

Acceptance criteria are the conditions a piece of work must meet to be accepted as done, written before the work starts. They turn "make it good" into checks anyone can apply, so the person doing the task, the one accepting it and an AI checking it all mean the same thing by done.

Acceptance criteria examples

For a task "onboarding guide for a new manager": every step has an owner; the guide is in the knowledge base and the task links to it; a new hire got through it without help. Each line is a yes or a no, which is what makes it checkable.

Acceptance criteria format

Two formats are common. A checklist of plain statements, each true or false. Or Given, When, Then from software teams: given a starting situation, when something happens, then this result follows. Whatever the format, a good criterion is specific, testable and agreed before the work begins.

Why AI work needs acceptance criteria

AI can produce a confident result that misses the point. Criteria written in advance let a person or an LLM as a judge check it the same way every time, and give a clear reason when work goes back. They are the base of harness engineering, the third stage of AI maturity.