Three ways to create a test case — write the steps, describe the flow, or record it once.

A test case is the unit Autonomy defines and executes: one named journey with an ordered list of steps. You can author it three ways, and every one of them ends in the same place — the step editor, where you review the steps before saving.

Start a new test case from Test Cases → New Test Case. Step 1 collects the name, platforms, and environment. Step 2 asks how you want to author it.

Choosing a method

Method	Badge	Best for
Write the steps	Manual	Short, precise flows where you already know the exact sequence.
Describe the flow	AI Draft	Exploratory journeys. You describe intent in plain language and Autonomy drafts intent-level steps.
Record the flow	Recording	Long flows. You perform the journey once in a live browser session and Autonomy drafts the steps from what it observed.

Writing and describing are four-step wizards (Details → Method → Steps → Review). Recording adds one step (Details → Method → Record → Steps → Review), because the recording session sits between choosing the method and reviewing the drafted steps.

Steps are intent, not a script

Whichever method you use, the saved steps describe intent. Autonomy's agent resolves them against the live product at run time — it does not replay coordinates or a fixed selector chain. A recording is authoring input, not an execution artifact.

That distinction matters in practice:

A step recorded as a click on "Continue" still passes when the button moves or the markup changes, as long as a control with that intent exists.
Recording the journey once does not lock the test to that session's data. Credential-like fields become agent-chosen values.
Editing a step after recording is normal and expected. The recording is a draft.

Step actions

Drafted and hand-written steps use the same vocabulary:

Navigate — go to a URL.
Type — fill a text field.
Tap/Click — click or tap an element.
Scroll — scroll the page or a container.
Wait — pause for a condition.
Assert — verify text, an element, or the URL.
Visual Compare — compare the screen against a reference. See Visual Checks.
Validate Network — assert on the requests the UI caused.

Values, credentials, and environment variables

Do not paste real credentials into steps.

For email, password, OTP, and verification-code fields, leave the value as an agent-chosen instruction such as Agent-chosen at run time: strong password. Autonomy generates a high-entropy value at run time and reuses it for the rest of the run. Common example passwords are frequently rejected by real auth providers.
To reference environment configuration, use the {{env.VAR_NAME}} form. The value resolves from the selected environment when the run starts.
Values that contain a number or a currency symbol make poor targets. Prefer the checkout button over Pay $50.00, because the amount changes and the label does not identify the control.

Attachments and AI drafting

The Describe the flow method accepts attachments alongside the prompt — screenshots, PDFs, and an OpenAPI or Swagger document (.yaml, .json). An attached API spec lets Autonomy draft network assertions next to the UI steps, and connects the case to API Assurance.

When you re-prompt an existing case from the edit dialog, Autonomy asks whether the new draft should replace the current steps or be added after them. Replace is the right choice when the flow changed; add is the right choice when you are extending a journey.

Where to go next

Read Record the Flow for the recording session in detail. Read Visual Checks to attach design references. Read Test Plans to group cases and run them together.

Overview