01What it is
AI test case generation reads a source that describes behaviour — a spec, a PRD, acceptance criteria, or an issue in your tracker — and drafts test cases from it: a title, steps, and an expected result for each. Instead of authoring every case from a blank page, you start from a reviewable draft and refine.
02How it works in practice
- Pick a source. Paste text directly, or point the generator at an item in a connected tool — a Linear issue, a Notion page, a Jira ticket, or an Azure DevOps work item.
- Generate. The model reads the source and drafts candidate cases mapped to your project's template — so the fields match what your team already uses, not a flattened generic structure.
- Review and save. You see the drafts, edit or discard as needed, and save the ones worth keeping. They become normal test cases.
- Keep traceability. Cases generated from a tracked item link back to the source, so coverage traces to the requirement that justified it.
03What good output looks like
- Template-aware — titles, priority, steps, and expected result populate the fields your project defines.
- Reviewable, not auto-applied — drafts wait for a human to approve.
- Linked to the source — each case points back to the spec or issue it came from.
04Keeping humans in the loop
Generation is a drafting aid, not a decision-maker. The team still decides what matters, what to keep, and whether an expected result is correct. The value is speed and completeness on the first pass — not removing review. See AI in software testing for where that line sits.
05Related
06Frequently asked questions
What does AI test case generation actually produce?
A draft set of test cases — each with a title, steps, and expected result — derived from a source you point it at (pasted text, or an item in a connected tracker). In a good tool the drafts map to your project template so they match the fields you already use, rather than a generic shape.
Do the generated cases get created automatically?
They should not. In TestOrchestrator, generation returns reviewable drafts — you select which ones to save, and they land as normal test cases you can edit. Nothing is created without a human deciding.
What makes a good source for generation?
Anything that describes desired behaviour: a feature spec, a PRD, acceptance criteria, or a well-written issue. The clearer the requirement, the better the drafted cases.
Is it accurate?
Treat output as a strong first draft. It removes the blank page and catches cases people forget, but you should always confirm steps and expected results against the real requirement before relying on them.