The AI Constraint Generator accepts a plain language description of the rules in your system under test and suggests constraints automatically. Rather than manually clicking through parameter values to define each constraint, you can type a plain language description and let the AI service suggest constraints for you to review and accept.
Describing your constraints
The AI Constraint Generator appears as a banner at the top of the Constraints screen, in both the regular constraints view and the bulk view. Clicking anywhere on the banner header expands it to reveal a text area accepting a plain language description of the rules in your system under test.
You don't need to use any special syntax to describe the rules of your system under test. Write naturally, as you would explain the rule to a colleague. Once you've finished writing your description and click Generate Constraints, the AI service will attempt to map your description onto the parameters in your model.
Example descriptions
"Read-only users should not be able to edit or delete records. Admin settings are only available for users with a manager role or above."
"Mobile browsers don't support wide screen views. IE is not compatible with macOS."
At least 5 words are required before the Generate Constraints button becomes active. If your description is too short, a prompt will appear beneath the text area.
Design Guidance
If Design Guidance has been set up for your model, the AI service will use your testing goal and any uploaded context documents to improve its suggestions. If Design Guidance is not yet configured, clicking Generate Constraints will show a dialog offering two options:
Proceed without Guidance generates constraints using only your description and the existing model. This can still produce useful suggestions, especially with a clearly-worded and detailed description.
Provide Guidance opens the Design Guidance setup so you can add a testing goal and context before generating.
Reviewing suggested constraints
Once the AI service has processed your description, the results appear in the banner. Each suggested constraint shows:
The constraint type as a color-coded pill: Invalid, Bound, or Advanced
The parameter values involved, in the same format used by the constraints table
Any explanation the AI service included with the suggestion
Accepting and rejecting suggestions
Standard constraints view
Suggested constraints can be accepted or rejected individually, or all at once.
Click the ✓ icon on a row to accept that suggestion. The constraint is saved immediately and appears in the constraints table below.
Click the ✕ icon on a row to reject that suggestion. Rejected constraints are discarded.
Use Accept All to accept every suggestion at once
You can also use Reject All to discard all of them.
AI-suggested constraints include a comment describing the suggestion, so you can identify which constraints were generated by the AI service after the fact. These comments are prepended with an icon (✨), to denote them as AI-suggested.
Advanced constraints view
The AI Constraint Generator is also available in the advanced constraints view. Just as with the standard constraints view, suggested constraints can be accepted or rejected individually, or all at once.
In this view, AI-suggested constraints are appended to the bulk editor directly, pre-filled with the correct syntax.
Tips for writing effective constraint descriptions
Be specific. The more detail you provide in your description, the more accurate the suggestions will be. Targeted and specific phrasing, such as "IE is not compatible with macOS or Linux" is likely to produce a useful suggestion.
Use Design Guidance. If your model has a testing goal or context documents configured, the AI service can use that background to produce suggestions more relevant to your testing objectives.
Generate in rounds. If the first set of suggestions doesn't cover everything you had in mind, you can generate again with a different or more specific description.
If no suggestions are returned, try rephrasing your description or adding more detail. The AI service may not always be able to map a description onto the specific parameters and values in your model.