Two different questions
Use “agentic” when you talk about how a test came to be: an AI agent scans the page, drafts the steps and executes them in real browsers while approval gates the run. Use “autonomous” when you talk about a kept test running unattended.
| Autonomous | Agentic | |
|---|---|---|
| Describes | How a kept test runs | How the test is produced |
| Who is present | Nobody: it fires on a push, deploy, schedule, CI call or coding-agent request | You: the agent drafts and runs it, you approve the plan |
| What it may change | Nothing: it replays the approved steps | It proposes steps and repairs as new versions you approve |
What autonomous testing means here
A test you approved and kept reruns without you present: on every GitHub push, every deploy, a daily or hourly schedule (by plan), from CI, or when your coding agent asks. The first run is always supervised, and every approval and repair lands in an append-only audit log.
Three explicit levels of autonomy
| Level | Applies to | What happens |
|---|---|---|
| Supervised | The first run, or any materially changed plan | Scan and draft are automatic; execution needs your approval |
| Approved automation | A saved, unchanged, approved test | Manual reruns stay pinned to the approved version |
| Runs itself | Kept tests in a suite | Rerun on every push, deploy or schedule, report by email and can fail your pipeline, still bounded by the approved steps |