AI for Testing and QA

TL;DR

• AI can generate test cases, test code, and exploratory scenarios.
• It optimizes for coverage, not risk. You decide what matters.
• Use AI to expand test ideas. You own the strategy and the "would a user hit this?" judgment.

Testing is about finding problems before users do. AI can produce a lot of tests. The question is: do they find the right problems?

Test Case Generation

Good use cases:

• "Generate test cases for this user flow: [steps]. Include happy path and 3 edge cases"
• "Suggest negative test cases for this API endpoint"
• "What scenarios should we test for this checkout flow?"

What AI misses:

• Domain-specific edge cases ("what if they're a returning customer with a expired promo?")
• Business rules you haven't stated
• Prioritization — AI gives you 50 cases; you need to pick the 10 that matter

Use AI to brainstorm. You filter and prioritize.

Test Code Generation

Good use cases:

• "Generate pytest/ Jest / etc. tests for this function"
• "Add integration tests for this API. Use our existing test pattern"
• "Convert these manual test steps to Playwright/Cypress"

Cautions:

• Tests must actually run. AI can generate syntactically correct, logically wrong tests.
• Assertions — AI may assert the wrong thing. "Test passes" ≠ "we tested the right thing."
• Flakiness — AI doesn't know your timing issues, race conditions, or flaky selectors.

Workflow: Generate → run → fix. Expect to fix.

Exploratory Testing Support

Good use cases:

• "Suggest exploratory scenarios for this feature"
• "What could a malicious user try here?"
• "List risk areas for this release"

AI can expand your mental map. You still do the exploring.

Visual Regression and Accessibility

Good use cases:

• "Suggest selectors for this component for visual regression"
• "Generate accessibility test cases for this form"
• "What WCAG criteria apply to this UI?"

AI knows the patterns. You verify they match your implementation.

When Not to Use AI

• Critical path testing — Don't let AI own the tests for payment, auth, or data integrity without deep review.
• Test strategy — What to test, when, and how much — that's you.
• Interpretation — AI can't tell you why a test failed or whether it's a real bug or a bad test.

Do This Next

1. Generate test cases for one feature with AI. Run through them. How many were useful? How many were noise?
2. Use AI to generate test code for one function. Run it. Fix what breaks. Note the pattern of errors.