This page is a collection of strategies, configurations, and best practices to help you optimize your AI testing agents for maximum speed, accuracy, and cost efficiency with Stably AI.

​
Improve Test Speed

The key to faster test execution is reducing unnecessary runtime AI calls while maximizing parallel processing. Focus on using cached Playwright actions for deterministic steps and running multiple tests simultaneously to achieve optimal speed and cost efficiency.
Use fewer AI Actions for deterministic steps. Cached Playwright steps run faster and cost less than AI actions.For relatively deterministic tests, use auto-heal step whenever possible. These tests execute cached Playwright actions by default, making them faster and cheaper.
When minor UI changes occur, auto-heal steps will fallback to AI Actions for repair. After these tests are healed, you can accept these changes to make future test runs even faster and cheaper.
Use auto-heal mode for most test steps. These execute cached Playwright actions by default, making them faster and cheaper.
When minor UI changes occur, auto-heal steps fallback to AI Actions for repair. Review and approve changes to maintain speed.

Learn More

See how auto-heal works and configure healing sensitivity.
Run multiple tests simultaneously to reduce overall execution time and improve testing efficiency.
1

Design for Parallelism

Ensure tests are idempotent and isolated from each other to enable safe parallel execution.Idempotent tests have their own setup/teardown and can run repeatedly without affecting other tests. Each test should:
  • Verify initial state and clean up any existing data at the start
  • Create its own test data during execution
  • Clean up after itself when finished
For example, a user registration test should first check if the test user already exists and delete it, then create a fresh user account, and finally clean up the created account.AI Action Prompt: "Check if user 'test@example.com' exists in the system and delete it if found, then verify the user list is clean before proceeding."
2

Configure Workers

Use auto-parallelism or manually set worker count in parallelism settings.
3

Review and Approve Changes

Regularly review auto-heal changes and approve them to reduce AI Actions across all tests.
Tests must be designed to run independently. Shared state or dependencies can cause race conditions in parallel execution.

​
Improve AI Accuracy

AI accuracy directly impacts test reliability and reduces maintenance overhead. The right model selection and comprehensive knowledge base are critical for generating tests that work consistently across different scenarios and application changes.
Stably offers multiple AI models optimized for different scenarios:
  • Flash Model: Fast, cost-effective default for >95% of routine tasks
  • DeepThink Model: Higher accuracy for complex scenarios, but slower and more expensive
  • Backup Model: Alternative for rare edge cases where other models struggle
Switch models in each action such as AI Action and AI Assertions. Flash model outperforms competitors in most benchmarks.
Many test failures are caused by the AI not having enough project-specific knowledge. Make sure to provide detailed documentation, unique workflows, and business rules so the AI can generate accurate and reliable tests for your application.
1

Import Documentation

Add product docs, API references, and help content to provide context-specific knowledge. Learn more
2

Add Manual Knowledge

Document business logic, edge cases, and integration details not found elsewhere. Learn more
3

Maintain Knowledge

Regularly update knowledge base as your application evolves. Learn more

Learn More

Discover all knowledge sources and best practices for building effective app knowledge.